From 601480122970ebc3a9b5ce8059f6b20dd8390e78 Mon Sep 17 00:00:00 2001
From: Mario Zechner <badlogicgames@gmail.com>
Date: Tue, 2 Jun 2026 15:09:36 +0200
Subject: [PATCH 01/10] Add NVIDIA NIM provider

---
 packages/ai/CHANGELOG.md                      |   4 +
 packages/ai/README.md                         |   6 +-
 packages/ai/scripts/generate-models.ts        |  97 +++++
 packages/ai/src/env-api-keys.ts               |   1 +
 packages/ai/src/models.generated.ts           | 363 ++++++++++++++++++
 .../ai/src/providers/openai-completions.ts    |  10 +-
 packages/ai/src/types.ts                      |   1 +
 packages/ai/test/stream.test.ts               |  24 ++
 packages/coding-agent/CHANGELOG.md            |   1 +
 packages/coding-agent/README.md               |   5 +-
 packages/coding-agent/docs/providers.md       |   2 +
 packages/coding-agent/docs/usage.md           |   2 +-
 packages/coding-agent/src/cli/args.ts         |   1 +
 .../coding-agent/src/core/agent-session.ts    |   1 +
 .../core/compaction/branch-summarization.ts   |  31 +-
 .../coding-agent/src/core/model-resolver.ts   |   1 +
 .../src/core/provider-attribution.ts          |  97 +++++
 .../src/core/provider-display-names.ts        |   1 +
 packages/coding-agent/src/core/sdk.ts         |  52 +--
 .../test/sdk-openrouter-attribution.test.ts   |  66 +++-
 20 files changed, 700 insertions(+), 66 deletions(-)
 create mode 100644 packages/coding-agent/src/core/provider-attribution.ts

diff --git a/packages/ai/CHANGELOG.md b/packages/ai/CHANGELOG.md
index af63cdc4a..4a616f893 100644
--- a/packages/ai/CHANGELOG.md
+++ b/packages/ai/CHANGELOG.md
@@ -2,6 +2,10 @@
 
 ## [Unreleased]
 
+### Added
+
+- Added NVIDIA NIM as a built-in OpenAI-compatible provider, exposing public NIM models that support tool use.
+
 ### Fixed
 
 - Fixed Amazon Bedrock requests to replace blank required user/tool-result text with a placeholder and skip blank replay text blocks ([#4975](https://github.com/earendil-works/pi/issues/4975)).
diff --git a/packages/ai/README.md b/packages/ai/README.md
index e6624d3b8..76a987692 100644
--- a/packages/ai/README.md
+++ b/packages/ai/README.md
@@ -54,6 +54,7 @@ Unified LLM API with automatic model discovery, provider configuration, token an
 - **Azure OpenAI (Responses)**
 - **OpenAI Codex** (ChatGPT Plus/Pro subscription, requires OAuth, see below)
 - **DeepSeek**
+- **NVIDIA NIM**
 - **Anthropic**
 - **Google**
 - **Vertex AI** (Gemini via Vertex AI)
@@ -801,7 +802,7 @@ A **provider** offers models through a specific API. For example:
 - **Google** models use the `google-generative-ai` API
 - **OpenAI** models use the `openai-responses` API
 - **Mistral** models use the `mistral-conversations` API
-- **xAI, Cerebras, Groq, Together AI, etc.** models use the `openai-completions` API (OpenAI-compatible)
+- **xAI, Cerebras, Groq, NVIDIA NIM, Together AI, etc.** models use the `openai-completions` API (OpenAI-compatible)
 
 ### Querying Providers and Models
 
@@ -923,7 +924,7 @@ const ollamaReasoningModel: Model<'openai-completions'> = {
 
 ### OpenAI Compatibility Settings
 
-The `openai-completions` API is implemented by many providers with minor differences. By default, the library auto-detects compatibility settings based on `baseUrl` for a small set of known OpenAI-compatible providers (Cerebras, xAI, Chutes, DeepSeek, Together AI, zAi, OpenCode, Cloudflare Workers AI, etc.). For custom proxies or unknown endpoints, you can override these settings via the `compat` field. For `openai-responses` models, the compat field only supports Responses-specific flags.
+The `openai-completions` API is implemented by many providers with minor differences. By default, the library auto-detects compatibility settings based on `baseUrl` for a small set of known OpenAI-compatible providers (Cerebras, xAI, Chutes, DeepSeek, NVIDIA NIM, Together AI, zAi, OpenCode, Cloudflare Workers AI, etc.). For custom proxies or unknown endpoints, you can override these settings via the `compat` field. For `openai-responses` models, the compat field only supports Responses-specific flags.
 
 ```typescript
 interface OpenAICompletionsCompat {
@@ -1102,6 +1103,7 @@ In Node.js environments, you can set environment variables to avoid passing API
 | Azure OpenAI | `AZURE_OPENAI_API_KEY` + `AZURE_OPENAI_BASE_URL` (e.g. `https://{resource}.openai.azure.com`) or `AZURE_OPENAI_RESOURCE_NAME`. Supports `*.openai.azure.com` and `*.cognitiveservices.azure.com`; root endpoints auto-normalize to `/openai/v1`. Optional: `AZURE_OPENAI_API_VERSION` (default `v1`), `AZURE_OPENAI_DEPLOYMENT_NAME_MAP`. |
 | Anthropic | `ANTHROPIC_API_KEY` or `ANTHROPIC_OAUTH_TOKEN` |
 | DeepSeek | `DEEPSEEK_API_KEY` |
+| NVIDIA NIM | `NVIDIA_API_KEY` |
 | Google | `GEMINI_API_KEY` |
 | Vertex AI | `GOOGLE_CLOUD_API_KEY` or `GOOGLE_CLOUD_PROJECT` (or `GCLOUD_PROJECT`) + `GOOGLE_CLOUD_LOCATION` + ADC |
 | Mistral | `MISTRAL_API_KEY` |
diff --git a/packages/ai/scripts/generate-models.ts b/packages/ai/scripts/generate-models.ts
index 50df88f2b..8fb62d145 100644
--- a/packages/ai/scripts/generate-models.ts
+++ b/packages/ai/scripts/generate-models.ts
@@ -32,12 +32,17 @@ interface ModelsDevModel {
 	};
 	modalities?: {
 		input?: string[];
+		output?: string[];
 	};
 	provider?: {
 		npm?: string;
 	};
 }
 
+interface NvidiaNimModelListItem {
+	id: string;
+}
+
 interface AiGatewayModel {
 	id: string;
 	name?: string;
@@ -117,6 +122,39 @@ const TOGETHER_TOGGLE_REASONING_LEVEL_MAP = {
 
 const AI_GATEWAY_MODELS_URL = "https://ai-gateway.vercel.sh/v1";
 const AI_GATEWAY_BASE_URL = "https://ai-gateway.vercel.sh";
+const NVIDIA_BASE_URL = "https://integrate.api.nvidia.com/v1";
+const NVIDIA_HEADERS = {
+	"NVCF-POLL-SECONDS": "3600",
+} as const;
+const NVIDIA_OPENAI_COMPAT: OpenAICompletionsCompat = {
+	supportsStore: false,
+	supportsDeveloperRole: false,
+	supportsReasoningEffort: false,
+	maxTokensField: "max_tokens",
+	supportsStrictMode: false,
+	supportsLongCacheRetention: false,
+};
+const NVIDIA_NIM_UNSUPPORTED_MODELS = new Set([
+	"abacusai/dracarys-llama-3.1-70b-instruct",
+	"bytedance/seed-oss-36b-instruct",
+	"deepseek-ai/deepseek-v4-flash",
+	"deepseek-ai/deepseek-v4-pro",
+	"google/gemma-2-2b-it",
+	"google/gemma-3n-e2b-it",
+	"google/gemma-3n-e4b-it",
+	"google/gemma-4-31b-it",
+	"meta/llama-3.2-1b-instruct",
+	"meta/llama-4-maverick-17b-128e-instruct",
+	"microsoft/phi-4-mini-instruct",
+	"minimaxai/minimax-m2.7",
+	"mistralai/mistral-nemotron",
+	"nvidia/nemotron-mini-4b-instruct",
+	"qwen/qwen3-next-80b-a3b-instruct",
+	"qwen/qwen3.5-122b-a10b",
+	"qwen/qwen3.5-397b-a17b",
+	"sarvamai/sarvam-m",
+	"upstage/solar-10.7b-instruct",
+]);
 const ZAI_TOOL_STREAM_UNSUPPORTED_MODELS = new Set(["glm-4.5", "glm-4.5-air", "glm-4.5-flash", "glm-4.5v"]);
 const EAGER_TOOL_INPUT_STREAMING_UNSUPPORTED_ANTHROPIC_MODELS = new Set([
 	"github-copilot:claude-haiku-4.5",
@@ -312,6 +350,30 @@ function getBedrockBaseUrl(modelId: string): string {
 		: "https://bedrock-runtime.us-east-1.amazonaws.com";
 }
 
+function normalizeNvidiaModelId(modelId: string): string {
+	return modelId.toLowerCase().replaceAll("_", ".");
+}
+
+async function fetchNvidiaNimModelIds(): Promise<Map<string, string>> {
+	try {
+		console.log("Fetching models from NVIDIA NIM API...");
+		const response = await fetch(`${NVIDIA_BASE_URL}/models`);
+		const data = (await response.json()) as { data?: NvidiaNimModelListItem[] };
+		const modelIds = new Map<string, string>();
+
+		for (const model of data.data ?? []) {
+			modelIds.set(model.id, model.id);
+			modelIds.set(normalizeNvidiaModelId(model.id), model.id);
+		}
+
+		console.log(`Fetched ${data.data?.length ?? 0} model IDs from NVIDIA NIM`);
+		return modelIds;
+	} catch (error) {
+		console.error("Failed to fetch NVIDIA NIM models:", error);
+		return new Map();
+	}
+}
+
 async function fetchOpenRouterModels(): Promise<Model<any>[]> {
 	try {
 		console.log("Fetching models from OpenRouter API...");
@@ -435,6 +497,7 @@ async function loadModelsDevData(): Promise<Model<any>[]> {
 		const data = await response.json();
 
 		const models: Model<any>[] = [];
+		const nvidiaNimModelIds = data.nvidia?.models ? await fetchNvidiaNimModelIds() : new Map<string, string>();
 
 		// Process Amazon Bedrock models
 		if (data["amazon-bedrock"]?.models) {
@@ -836,6 +899,40 @@ async function loadModelsDevData(): Promise<Model<any>[]> {
 			}
 		}
 
+		// Process NVIDIA NIM models
+		if (data.nvidia?.models) {
+			for (const [modelId, model] of Object.entries(data.nvidia.models)) {
+				const m = model as ModelsDevModel;
+				if (m.tool_call !== true) continue;
+				if (!m.modalities?.input?.includes("text")) continue;
+				if (!m.modalities?.output?.includes("text")) continue;
+
+				const liveModelId = nvidiaNimModelIds.get(modelId) ?? nvidiaNimModelIds.get(normalizeNvidiaModelId(modelId));
+				if (!liveModelId) continue;
+				if (NVIDIA_NIM_UNSUPPORTED_MODELS.has(liveModelId)) continue;
+
+				models.push({
+					id: liveModelId,
+					name: m.name || liveModelId,
+					api: "openai-completions",
+					provider: "nvidia",
+					baseUrl: NVIDIA_BASE_URL,
+					headers: { ...NVIDIA_HEADERS },
+					reasoning: m.reasoning === true,
+					input: m.modalities.input.includes("image") ? ["text", "image"] : ["text"],
+					cost: {
+						input: m.cost?.input || 0,
+						output: m.cost?.output || 0,
+						cacheRead: m.cost?.cache_read || 0,
+						cacheWrite: m.cost?.cache_write || 0,
+					},
+					compat: NVIDIA_OPENAI_COMPAT,
+					contextWindow: m.limit?.context || 4096,
+					maxTokens: m.limit?.output || 4096,
+				});
+			}
+		}
+
 		// Process Together AI models
 		const togetherProvider = data.together ?? data.togetherai ?? data["together-ai"];
 		if (togetherProvider?.models) {
diff --git a/packages/ai/src/env-api-keys.ts b/packages/ai/src/env-api-keys.ts
index a00e8470a..6321a03a1 100644
--- a/packages/ai/src/env-api-keys.ts
+++ b/packages/ai/src/env-api-keys.ts
@@ -101,6 +101,7 @@ function getApiKeyEnvVars(provider: string): readonly string[] | undefined {
 	const envMap: Record<string, string> = {
 		openai: "OPENAI_API_KEY",
 		"azure-openai-responses": "AZURE_OPENAI_API_KEY",
+		nvidia: "NVIDIA_API_KEY",
 		deepseek: "DEEPSEEK_API_KEY",
 		google: "GEMINI_API_KEY",
 		"google-vertex": "GOOGLE_CLOUD_API_KEY",
diff --git a/packages/ai/src/models.generated.ts b/packages/ai/src/models.generated.ts
index 8e83c5808..cd022b795 100644
--- a/packages/ai/src/models.generated.ts
+++ b/packages/ai/src/models.generated.ts
@@ -6335,6 +6335,369 @@ export const MODELS = {
 			maxTokens: 262144,
 		} satisfies Model<"openai-completions">,
 	},
+	"nvidia": {
+		"meta/llama-3.1-70b-instruct": {
+			id: "meta/llama-3.1-70b-instruct",
+			name: "Llama 3.1 70b Instruct",
+			api: "openai-completions",
+			provider: "nvidia",
+			baseUrl: "https://integrate.api.nvidia.com/v1",
+			headers: {"NVCF-POLL-SECONDS":"3600"},
+			compat: {"supportsStore":false,"supportsDeveloperRole":false,"supportsReasoningEffort":false,"maxTokensField":"max_tokens","supportsStrictMode":false,"supportsLongCacheRetention":false},
+			reasoning: false,
+			input: ["text"],
+			cost: {
+				input: 0,
+				output: 0,
+				cacheRead: 0,
+				cacheWrite: 0,
+			},
+			contextWindow: 128000,
+			maxTokens: 4096,
+		} satisfies Model<"openai-completions">,
+		"meta/llama-3.1-8b-instruct": {
+			id: "meta/llama-3.1-8b-instruct",
+			name: "Llama 3.1 8B Instruct",
+			api: "openai-completions",
+			provider: "nvidia",
+			baseUrl: "https://integrate.api.nvidia.com/v1",
+			headers: {"NVCF-POLL-SECONDS":"3600"},
+			compat: {"supportsStore":false,"supportsDeveloperRole":false,"supportsReasoningEffort":false,"maxTokensField":"max_tokens","supportsStrictMode":false,"supportsLongCacheRetention":false},
+			reasoning: false,
+			input: ["text"],
+			cost: {
+				input: 0,
+				output: 0,
+				cacheRead: 0,
+				cacheWrite: 0,
+			},
+			contextWindow: 16000,
+			maxTokens: 4096,
+		} satisfies Model<"openai-completions">,
+		"meta/llama-3.2-11b-vision-instruct": {
+			id: "meta/llama-3.2-11b-vision-instruct",
+			name: "Llama 3.2 11b Vision Instruct",
+			api: "openai-completions",
+			provider: "nvidia",
+			baseUrl: "https://integrate.api.nvidia.com/v1",
+			headers: {"NVCF-POLL-SECONDS":"3600"},
+			compat: {"supportsStore":false,"supportsDeveloperRole":false,"supportsReasoningEffort":false,"maxTokensField":"max_tokens","supportsStrictMode":false,"supportsLongCacheRetention":false},
+			reasoning: false,
+			input: ["text", "image"],
+			cost: {
+				input: 0,
+				output: 0,
+				cacheRead: 0,
+				cacheWrite: 0,
+			},
+			contextWindow: 128000,
+			maxTokens: 4096,
+		} satisfies Model<"openai-completions">,
+		"meta/llama-3.2-90b-vision-instruct": {
+			id: "meta/llama-3.2-90b-vision-instruct",
+			name: "Llama-3.2-90B-Vision-Instruct",
+			api: "openai-completions",
+			provider: "nvidia",
+			baseUrl: "https://integrate.api.nvidia.com/v1",
+			headers: {"NVCF-POLL-SECONDS":"3600"},
+			compat: {"supportsStore":false,"supportsDeveloperRole":false,"supportsReasoningEffort":false,"maxTokensField":"max_tokens","supportsStrictMode":false,"supportsLongCacheRetention":false},
+			reasoning: false,
+			input: ["text", "image"],
+			cost: {
+				input: 0,
+				output: 0,
+				cacheRead: 0,
+				cacheWrite: 0,
+			},
+			contextWindow: 128000,
+			maxTokens: 8192,
+		} satisfies Model<"openai-completions">,
+		"meta/llama-3.3-70b-instruct": {
+			id: "meta/llama-3.3-70b-instruct",
+			name: "Llama 3.3 70b Instruct",
+			api: "openai-completions",
+			provider: "nvidia",
+			baseUrl: "https://integrate.api.nvidia.com/v1",
+			headers: {"NVCF-POLL-SECONDS":"3600"},
+			compat: {"supportsStore":false,"supportsDeveloperRole":false,"supportsReasoningEffort":false,"maxTokensField":"max_tokens","supportsStrictMode":false,"supportsLongCacheRetention":false},
+			reasoning: false,
+			input: ["text"],
+			cost: {
+				input: 0,
+				output: 0,
+				cacheRead: 0,
+				cacheWrite: 0,
+			},
+			contextWindow: 128000,
+			maxTokens: 4096,
+		} satisfies Model<"openai-completions">,
+		"mistralai/mistral-large-3-675b-instruct-2512": {
+			id: "mistralai/mistral-large-3-675b-instruct-2512",
+			name: "Mistral Large 3 675B Instruct 2512",
+			api: "openai-completions",
+			provider: "nvidia",
+			baseUrl: "https://integrate.api.nvidia.com/v1",
+			headers: {"NVCF-POLL-SECONDS":"3600"},
+			compat: {"supportsStore":false,"supportsDeveloperRole":false,"supportsReasoningEffort":false,"maxTokensField":"max_tokens","supportsStrictMode":false,"supportsLongCacheRetention":false},
+			reasoning: false,
+			input: ["text", "image"],
+			cost: {
+				input: 0,
+				output: 0,
+				cacheRead: 0,
+				cacheWrite: 0,
+			},
+			contextWindow: 262144,
+			maxTokens: 262144,
+		} satisfies Model<"openai-completions">,
+		"mistralai/mistral-small-4-119b-2603": {
+			id: "mistralai/mistral-small-4-119b-2603",
+			name: "mistral-small-4-119b-2603",
+			api: "openai-completions",
+			provider: "nvidia",
+			baseUrl: "https://integrate.api.nvidia.com/v1",
+			headers: {"NVCF-POLL-SECONDS":"3600"},
+			compat: {"supportsStore":false,"supportsDeveloperRole":false,"supportsReasoningEffort":false,"maxTokensField":"max_tokens","supportsStrictMode":false,"supportsLongCacheRetention":false},
+			reasoning: false,
+			input: ["text"],
+			cost: {
+				input: 0,
+				output: 0,
+				cacheRead: 0,
+				cacheWrite: 0,
+			},
+			contextWindow: 128000,
+			maxTokens: 8192,
+		} satisfies Model<"openai-completions">,
+		"moonshotai/kimi-k2.6": {
+			id: "moonshotai/kimi-k2.6",
+			name: "Kimi K2.6",
+			api: "openai-completions",
+			provider: "nvidia",
+			baseUrl: "https://integrate.api.nvidia.com/v1",
+			headers: {"NVCF-POLL-SECONDS":"3600"},
+			compat: {"supportsStore":false,"supportsDeveloperRole":false,"supportsReasoningEffort":false,"maxTokensField":"max_tokens","supportsStrictMode":false,"supportsLongCacheRetention":false},
+			reasoning: true,
+			input: ["text", "image"],
+			cost: {
+				input: 0,
+				output: 0,
+				cacheRead: 0,
+				cacheWrite: 0,
+			},
+			contextWindow: 262144,
+			maxTokens: 262144,
+		} satisfies Model<"openai-completions">,
+		"nvidia/llama-3.3-nemotron-super-49b-v1": {
+			id: "nvidia/llama-3.3-nemotron-super-49b-v1",
+			name: "Llama 3.3 Nemotron Super 49B v1",
+			api: "openai-completions",
+			provider: "nvidia",
+			baseUrl: "https://integrate.api.nvidia.com/v1",
+			headers: {"NVCF-POLL-SECONDS":"3600"},
+			compat: {"supportsStore":false,"supportsDeveloperRole":false,"supportsReasoningEffort":false,"maxTokensField":"max_tokens","supportsStrictMode":false,"supportsLongCacheRetention":false},
+			reasoning: true,
+			input: ["text"],
+			cost: {
+				input: 0,
+				output: 0,
+				cacheRead: 0,
+				cacheWrite: 0,
+			},
+			contextWindow: 131072,
+			maxTokens: 131072,
+		} satisfies Model<"openai-completions">,
+		"nvidia/llama-3.3-nemotron-super-49b-v1.5": {
+			id: "nvidia/llama-3.3-nemotron-super-49b-v1.5",
+			name: "Llama 3.3 Nemotron Super 49B v1.5",
+			api: "openai-completions",
+			provider: "nvidia",
+			baseUrl: "https://integrate.api.nvidia.com/v1",
+			headers: {"NVCF-POLL-SECONDS":"3600"},
+			compat: {"supportsStore":false,"supportsDeveloperRole":false,"supportsReasoningEffort":false,"maxTokensField":"max_tokens","supportsStrictMode":false,"supportsLongCacheRetention":false},
+			reasoning: true,
+			input: ["text"],
+			cost: {
+				input: 0,
+				output: 0,
+				cacheRead: 0,
+				cacheWrite: 0,
+			},
+			contextWindow: 131072,
+			maxTokens: 131072,
+		} satisfies Model<"openai-completions">,
+		"nvidia/nemotron-3-nano-30b-a3b": {
+			id: "nvidia/nemotron-3-nano-30b-a3b",
+			name: "nemotron-3-nano-30b-a3b",
+			api: "openai-completions",
+			provider: "nvidia",
+			baseUrl: "https://integrate.api.nvidia.com/v1",
+			headers: {"NVCF-POLL-SECONDS":"3600"},
+			compat: {"supportsStore":false,"supportsDeveloperRole":false,"supportsReasoningEffort":false,"maxTokensField":"max_tokens","supportsStrictMode":false,"supportsLongCacheRetention":false},
+			reasoning: true,
+			input: ["text"],
+			cost: {
+				input: 0,
+				output: 0,
+				cacheRead: 0,
+				cacheWrite: 0,
+			},
+			contextWindow: 131072,
+			maxTokens: 131072,
+		} satisfies Model<"openai-completions">,
+		"nvidia/nemotron-3-nano-omni-30b-a3b-reasoning": {
+			id: "nvidia/nemotron-3-nano-omni-30b-a3b-reasoning",
+			name: "Nemotron 3 Nano Omni",
+			api: "openai-completions",
+			provider: "nvidia",
+			baseUrl: "https://integrate.api.nvidia.com/v1",
+			headers: {"NVCF-POLL-SECONDS":"3600"},
+			compat: {"supportsStore":false,"supportsDeveloperRole":false,"supportsReasoningEffort":false,"maxTokensField":"max_tokens","supportsStrictMode":false,"supportsLongCacheRetention":false},
+			reasoning: true,
+			input: ["text", "image"],
+			cost: {
+				input: 0,
+				output: 0,
+				cacheRead: 0,
+				cacheWrite: 0,
+			},
+			contextWindow: 256000,
+			maxTokens: 65536,
+		} satisfies Model<"openai-completions">,
+		"nvidia/nemotron-3-super-120b-a12b": {
+			id: "nvidia/nemotron-3-super-120b-a12b",
+			name: "Nemotron 3 Super",
+			api: "openai-completions",
+			provider: "nvidia",
+			baseUrl: "https://integrate.api.nvidia.com/v1",
+			headers: {"NVCF-POLL-SECONDS":"3600"},
+			compat: {"supportsStore":false,"supportsDeveloperRole":false,"supportsReasoningEffort":false,"maxTokensField":"max_tokens","supportsStrictMode":false,"supportsLongCacheRetention":false},
+			reasoning: true,
+			input: ["text"],
+			cost: {
+				input: 0.2,
+				output: 0.8,
+				cacheRead: 0,
+				cacheWrite: 0,
+			},
+			contextWindow: 262144,
+			maxTokens: 262144,
+		} satisfies Model<"openai-completions">,
+		"nvidia/nvidia-nemotron-nano-9b-v2": {
+			id: "nvidia/nvidia-nemotron-nano-9b-v2",
+			name: "nvidia-nemotron-nano-9b-v2",
+			api: "openai-completions",
+			provider: "nvidia",
+			baseUrl: "https://integrate.api.nvidia.com/v1",
+			headers: {"NVCF-POLL-SECONDS":"3600"},
+			compat: {"supportsStore":false,"supportsDeveloperRole":false,"supportsReasoningEffort":false,"maxTokensField":"max_tokens","supportsStrictMode":false,"supportsLongCacheRetention":false},
+			reasoning: true,
+			input: ["text"],
+			cost: {
+				input: 0,
+				output: 0,
+				cacheRead: 0,
+				cacheWrite: 0,
+			},
+			contextWindow: 131072,
+			maxTokens: 131072,
+		} satisfies Model<"openai-completions">,
+		"openai/gpt-oss-20b": {
+			id: "openai/gpt-oss-20b",
+			name: "GPT OSS 20B",
+			api: "openai-completions",
+			provider: "nvidia",
+			baseUrl: "https://integrate.api.nvidia.com/v1",
+			headers: {"NVCF-POLL-SECONDS":"3600"},
+			compat: {"supportsStore":false,"supportsDeveloperRole":false,"supportsReasoningEffort":false,"maxTokensField":"max_tokens","supportsStrictMode":false,"supportsLongCacheRetention":false},
+			reasoning: true,
+			input: ["text"],
+			cost: {
+				input: 0,
+				output: 0,
+				cacheRead: 0,
+				cacheWrite: 0,
+			},
+			contextWindow: 131072,
+			maxTokens: 32768,
+		} satisfies Model<"openai-completions">,
+		"qwen/qwen3-coder-480b-a35b-instruct": {
+			id: "qwen/qwen3-coder-480b-a35b-instruct",
+			name: "Qwen3 Coder 480B A35B Instruct",
+			api: "openai-completions",
+			provider: "nvidia",
+			baseUrl: "https://integrate.api.nvidia.com/v1",
+			headers: {"NVCF-POLL-SECONDS":"3600"},
+			compat: {"supportsStore":false,"supportsDeveloperRole":false,"supportsReasoningEffort":false,"maxTokensField":"max_tokens","supportsStrictMode":false,"supportsLongCacheRetention":false},
+			reasoning: false,
+			input: ["text"],
+			cost: {
+				input: 0,
+				output: 0,
+				cacheRead: 0,
+				cacheWrite: 0,
+			},
+			contextWindow: 262144,
+			maxTokens: 66536,
+		} satisfies Model<"openai-completions">,
+		"stepfun-ai/step-3.5-flash": {
+			id: "stepfun-ai/step-3.5-flash",
+			name: "Step 3.5 Flash",
+			api: "openai-completions",
+			provider: "nvidia",
+			baseUrl: "https://integrate.api.nvidia.com/v1",
+			headers: {"NVCF-POLL-SECONDS":"3600"},
+			compat: {"supportsStore":false,"supportsDeveloperRole":false,"supportsReasoningEffort":false,"maxTokensField":"max_tokens","supportsStrictMode":false,"supportsLongCacheRetention":false},
+			reasoning: true,
+			input: ["text"],
+			cost: {
+				input: 0,
+				output: 0,
+				cacheRead: 0,
+				cacheWrite: 0,
+			},
+			contextWindow: 256000,
+			maxTokens: 16384,
+		} satisfies Model<"openai-completions">,
+		"stepfun-ai/step-3.7-flash": {
+			id: "stepfun-ai/step-3.7-flash",
+			name: "Step 3.7 Flash",
+			api: "openai-completions",
+			provider: "nvidia",
+			baseUrl: "https://integrate.api.nvidia.com/v1",
+			headers: {"NVCF-POLL-SECONDS":"3600"},
+			compat: {"supportsStore":false,"supportsDeveloperRole":false,"supportsReasoningEffort":false,"maxTokensField":"max_tokens","supportsStrictMode":false,"supportsLongCacheRetention":false},
+			reasoning: true,
+			input: ["text", "image"],
+			cost: {
+				input: 0,
+				output: 0,
+				cacheRead: 0,
+				cacheWrite: 0,
+			},
+			contextWindow: 256000,
+			maxTokens: 16384,
+		} satisfies Model<"openai-completions">,
+		"z-ai/glm-5.1": {
+			id: "z-ai/glm-5.1",
+			name: "GLM-5.1",
+			api: "openai-completions",
+			provider: "nvidia",
+			baseUrl: "https://integrate.api.nvidia.com/v1",
+			headers: {"NVCF-POLL-SECONDS":"3600"},
+			compat: {"supportsStore":false,"supportsDeveloperRole":false,"supportsReasoningEffort":false,"maxTokensField":"max_tokens","supportsStrictMode":false,"supportsLongCacheRetention":false},
+			reasoning: true,
+			input: ["text"],
+			cost: {
+				input: 0,
+				output: 0,
+				cacheRead: 0,
+				cacheWrite: 0,
+			},
+			contextWindow: 131072,
+			maxTokens: 131072,
+		} satisfies Model<"openai-completions">,
+	},
 	"openai": {
 		"gpt-4": {
 			id: "gpt-4",
diff --git a/packages/ai/src/providers/openai-completions.ts b/packages/ai/src/providers/openai-completions.ts
index ae06326e1..c58cbf46c 100644
--- a/packages/ai/src/providers/openai-completions.ts
+++ b/packages/ai/src/providers/openai-completions.ts
@@ -1078,8 +1078,10 @@ function detectCompat(model: Model<"openai-completions">): ResolvedOpenAIComplet
 	const isOpenRouter = provider === "openrouter" || baseUrl.includes("openrouter.ai");
 	const isCloudflareWorkersAI = provider === "cloudflare-workers-ai" || baseUrl.includes("api.cloudflare.com");
 	const isCloudflareAiGateway = provider === "cloudflare-ai-gateway" || baseUrl.includes("gateway.ai.cloudflare.com");
+	const isNvidia = provider === "nvidia" || baseUrl.includes("integrate.api.nvidia.com");
 
 	const isNonStandard =
+		isNvidia ||
 		provider === "cerebras" ||
 		baseUrl.includes("cerebras.ai") ||
 		provider === "xai" ||
@@ -1094,7 +1096,7 @@ function detectCompat(model: Model<"openai-completions">): ResolvedOpenAIComplet
 		isCloudflareWorkersAI ||
 		isCloudflareAiGateway;
 
-	const useMaxTokens = baseUrl.includes("chutes.ai") || isMoonshot || isCloudflareAiGateway || isTogether;
+	const useMaxTokens = baseUrl.includes("chutes.ai") || isMoonshot || isCloudflareAiGateway || isTogether || isNvidia;
 
 	const isGrok = provider === "xai" || baseUrl.includes("api.x.ai");
 	const isDeepSeek = provider === "deepseek" || baseUrl.includes("deepseek.com");
@@ -1103,7 +1105,7 @@ function detectCompat(model: Model<"openai-completions">): ResolvedOpenAIComplet
 	return {
 		supportsStore: !isNonStandard,
 		supportsDeveloperRole: !isNonStandard && !isOpenRouter,
-		supportsReasoningEffort: !isGrok && !isZai && !isMoonshot && !isTogether && !isCloudflareAiGateway,
+		supportsReasoningEffort: !isGrok && !isZai && !isMoonshot && !isTogether && !isCloudflareAiGateway && !isNvidia,
 		supportsUsageInStreaming: true,
 		maxTokensField: useMaxTokens ? "max_tokens" : "max_completion_tokens",
 		requiresToolResultName: false,
@@ -1122,10 +1124,10 @@ function detectCompat(model: Model<"openai-completions">): ResolvedOpenAIComplet
 		openRouterRouting: {},
 		vercelGatewayRouting: {},
 		zaiToolStream: false,
-		supportsStrictMode: !isMoonshot && !isTogether && !isCloudflareAiGateway,
+		supportsStrictMode: !isMoonshot && !isTogether && !isCloudflareAiGateway && !isNvidia,
 		cacheControlFormat,
 		sendSessionAffinityHeaders: false,
-		supportsLongCacheRetention: !(isTogether || isCloudflareWorkersAI || isCloudflareAiGateway),
+		supportsLongCacheRetention: !(isTogether || isCloudflareWorkersAI || isCloudflareAiGateway || isNvidia),
 	};
 }
 
diff --git a/packages/ai/src/types.ts b/packages/ai/src/types.ts
index d0b053ea3..7a4487720 100644
--- a/packages/ai/src/types.ts
+++ b/packages/ai/src/types.ts
@@ -28,6 +28,7 @@ export type KnownProvider =
 	| "openai"
 	| "azure-openai-responses"
 	| "openai-codex"
+	| "nvidia"
 	| "deepseek"
 	| "github-copilot"
 	| "xai"
diff --git a/packages/ai/test/stream.test.ts b/packages/ai/test/stream.test.ts
index 298104dd4..1693f4066 100644
--- a/packages/ai/test/stream.test.ts
+++ b/packages/ai/test/stream.test.ts
@@ -787,6 +787,30 @@ describe("Generate E2E Tests", () => {
 		});
 	});
 
+	describe.skipIf(!process.env.NVIDIA_API_KEY)("NVIDIA NIM Provider (Nemotron 3 Super via OpenAI Completions)", () => {
+		const llm = getModel("nvidia", "nvidia/nemotron-3-super-120b-a12b");
+
+		it("should complete basic text generation", { retry: 3 }, async () => {
+			await basicTextGeneration(llm);
+		});
+
+		it("should handle tool calling", { retry: 3 }, async () => {
+			await handleToolCall(llm);
+		});
+
+		it("should handle streaming", { retry: 3 }, async () => {
+			await handleStreaming(llm);
+		});
+
+		it("should handle thinking mode", { retry: 3 }, async () => {
+			await handleThinking(llm, { reasoningEffort: "high" });
+		});
+
+		it("should handle multi-turn with thinking and tools", { retry: 3 }, async () => {
+			await multiTurn(llm, { reasoningEffort: "high" });
+		});
+	});
+
 	describe.skipIf(!process.env.OPENROUTER_API_KEY)("OpenRouter Provider (glm-4.5v via OpenAI Completions)", () => {
 		const llm = getModel("openrouter", "z-ai/glm-4.5v");
 
diff --git a/packages/coding-agent/CHANGELOG.md b/packages/coding-agent/CHANGELOG.md
index 85b26f8f7..28401d532 100644
--- a/packages/coding-agent/CHANGELOG.md
+++ b/packages/coding-agent/CHANGELOG.md
@@ -4,6 +4,7 @@
 
 ### Added
 
+- Added NVIDIA NIM provider selection, setup documentation, and direct NIM request attribution headers.
 - Added `ctx.mode` to extension contexts so extensions can distinguish TUI, RPC, JSON, and print mode.
 - Added `ctx.getSystemPromptOptions()` for extension commands to inspect the current base system prompt inputs.
 
diff --git a/packages/coding-agent/README.md b/packages/coding-agent/README.md
index 0cf352551..585a93a0c 100644
--- a/packages/coding-agent/README.md
+++ b/packages/coding-agent/README.md
@@ -113,6 +113,7 @@ For each built-in provider, pi maintains a list of tool-capable models, updated
 - OpenAI
 - Azure OpenAI
 - DeepSeek
+- NVIDIA NIM
 - Google Gemini
 - Google Vertex
 - Amazon Bedrock
@@ -290,7 +291,7 @@ See [docs/settings.md](docs/settings.md) for all options.
 Pi has two separate startup features:
 
 - **Update check:** fetches `https://pi.dev/api/latest-version` to check whether a newer Pi version exists. Disable it with `PI_SKIP_VERSION_CHECK=1`. Disabling update checks only turns off this check.
-- **Install/update telemetry:** after first install or a changelog-detected update, sends an anonymous version ping to `https://pi.dev/api/report-install`. Opt out by setting `enableInstallTelemetry` to `false` in `settings.json`, or by setting `PI_TELEMETRY=0`. This does not disable update checks; Pi may still contact `pi.dev` for the latest version unless update checks are disabled or offline mode is enabled.
+- **Install/update telemetry:** after first install or a changelog-detected update, sends an anonymous version ping to `https://pi.dev/api/report-install`. This setting also controls optional provider attribution headers for OpenRouter, Cloudflare, and direct NVIDIA NIM requests. Opt out by setting `enableInstallTelemetry` to `false` in `settings.json`, or by setting `PI_TELEMETRY=0`. This does not disable update checks; Pi may still contact `pi.dev` for the latest version unless update checks are disabled or offline mode is enabled.
 
 Use `--offline` or `PI_OFFLINE=1` to disable all startup network operations described here, including update checks, package update checks, and install/update telemetry.
 
@@ -641,7 +642,7 @@ pi --thinking high "Solve this complex problem"
 | `PI_PACKAGE_DIR` | Override package directory (useful for Nix/Guix where store paths tokenize poorly) |
 | `PI_OFFLINE` | Disable startup network operations, including update checks, package update checks, and install/update telemetry |
 | `PI_SKIP_VERSION_CHECK` | Skip the Pi version update check at startup. This prevents the `pi.dev` latest-version request |
-| `PI_TELEMETRY` | Override install/update telemetry. Use `1`/`true`/`yes` to enable or `0`/`false`/`no` to disable. This does not disable update checks |
+| `PI_TELEMETRY` | Override install/update telemetry and provider attribution headers. Use `1`/`true`/`yes` to enable or `0`/`false`/`no` to disable. This does not disable update checks |
 | `PI_CACHE_RETENTION` | Set to `long` for extended prompt cache (Anthropic: 1h, OpenAI: 24h) |
 | `VISUAL`, `EDITOR` | External editor for Ctrl+G |
 
diff --git a/packages/coding-agent/docs/providers.md b/packages/coding-agent/docs/providers.md
index 07b6ba085..e7a77c5aa 100644
--- a/packages/coding-agent/docs/providers.md
+++ b/packages/coding-agent/docs/providers.md
@@ -52,6 +52,7 @@ pi
 | Azure OpenAI Responses | `AZURE_OPENAI_API_KEY` | `azure-openai-responses` |
 | OpenAI | `OPENAI_API_KEY` | `openai` |
 | DeepSeek | `DEEPSEEK_API_KEY` | `deepseek` |
+| NVIDIA NIM | `NVIDIA_API_KEY` | `nvidia` |
 | Google Gemini | `GEMINI_API_KEY` | `google` |
 | Mistral | `MISTRAL_API_KEY` | `mistral` |
 | Groq | `GROQ_API_KEY` | `groq` |
@@ -86,6 +87,7 @@ Store credentials in `~/.pi/agent/auth.json`:
   "anthropic": { "type": "api_key", "key": "sk-ant-..." },
   "openai": { "type": "api_key", "key": "sk-..." },
   "deepseek": { "type": "api_key", "key": "sk-..." },
+  "nvidia": { "type": "api_key", "key": "nvapi-..." },
   "google": { "type": "api_key", "key": "..." },
   "opencode": { "type": "api_key", "key": "..." },
   "opencode-go": { "type": "api_key", "key": "..." },
diff --git a/packages/coding-agent/docs/usage.md b/packages/coding-agent/docs/usage.md
index 47d08769c..2f2a64493 100644
--- a/packages/coding-agent/docs/usage.md
+++ b/packages/coding-agent/docs/usage.md
@@ -275,7 +275,7 @@ pi --exclude-tools ask_question
 | `PI_PACKAGE_DIR` | Override package directory, useful for Nix/Guix store paths |
 | `PI_OFFLINE` | Disable startup network operations, including update checks, package update checks, and install/update telemetry |
 | `PI_SKIP_VERSION_CHECK` | Skip the Pi version update check at startup. This prevents the `pi.dev` latest-version request |
-| `PI_TELEMETRY` | Override install/update telemetry: `1`/`true`/`yes` or `0`/`false`/`no`. This does not disable update checks |
+| `PI_TELEMETRY` | Override install/update telemetry and provider attribution headers: `1`/`true`/`yes` or `0`/`false`/`no`. This does not disable update checks |
 | `PI_CACHE_RETENTION` | Set to `long` for extended prompt cache where supported |
 | `VISUAL`, `EDITOR` | External editor for Ctrl+G |
 
diff --git a/packages/coding-agent/src/cli/args.ts b/packages/coding-agent/src/cli/args.ts
index 2decae247..8156bd03f 100644
--- a/packages/coding-agent/src/cli/args.ts
+++ b/packages/coding-agent/src/cli/args.ts
@@ -335,6 +335,7 @@ ${chalk.bold("Environment Variables:")}
   AZURE_OPENAI_API_VERSION         - Azure OpenAI API version (default: v1)
   AZURE_OPENAI_DEPLOYMENT_NAME_MAP - Azure OpenAI model=deployment map (comma-separated)
   DEEPSEEK_API_KEY                 - DeepSeek API key
+  NVIDIA_API_KEY                   - NVIDIA NIM API key
   GEMINI_API_KEY                   - Google Gemini API key
   GROQ_API_KEY                     - Groq API key
   CEREBRAS_API_KEY                 - Cerebras API key
diff --git a/packages/coding-agent/src/core/agent-session.ts b/packages/coding-agent/src/core/agent-session.ts
index 556ca20e2..de76a6198 100644
--- a/packages/coding-agent/src/core/agent-session.ts
+++ b/packages/coding-agent/src/core/agent-session.ts
@@ -2787,6 +2787,7 @@ export class AgentSession {
 					customInstructions,
 					replaceInstructions,
 					reserveTokens: branchSummarySettings.reserveTokens,
+					streamFn: this.agent.streamFn,
 				});
 				if (result.aborted) {
 					return { cancelled: true, aborted: true };
diff --git a/packages/coding-agent/src/core/compaction/branch-summarization.ts b/packages/coding-agent/src/core/compaction/branch-summarization.ts
index 0039bc79d..f6ce3d677 100644
--- a/packages/coding-agent/src/core/compaction/branch-summarization.ts
+++ b/packages/coding-agent/src/core/compaction/branch-summarization.ts
@@ -5,8 +5,8 @@
  * a summary of the branch being left so context isn't lost.
  */
 
-import type { AgentMessage } from "@earendil-works/pi-agent-core";
-import type { Model } from "@earendil-works/pi-ai";
+import type { AgentMessage, StreamFn } from "@earendil-works/pi-agent-core";
+import type { Model, SimpleStreamOptions } from "@earendil-works/pi-ai";
 import { completeSimple } from "@earendil-works/pi-ai";
 import {
 	convertToLlm,
@@ -77,6 +77,8 @@ export interface GenerateBranchSummaryOptions {
 	replaceInstructions?: boolean;
 	/** Tokens reserved for prompt + LLM response (default 16384) */
 	reserveTokens?: number;
+	/** Optional session stream function. Used to preserve SDK request behavior without mutating agent state. */
+	streamFn?: StreamFn;
 }
 
 // ============================================================================
@@ -284,7 +286,16 @@ export async function generateBranchSummary(
 	entries: SessionEntry[],
 	options: GenerateBranchSummaryOptions,
 ): Promise<BranchSummaryResult> {
-	const { model, apiKey, headers, signal, customInstructions, replaceInstructions, reserveTokens = 16384 } = options;
+	const {
+		model,
+		apiKey,
+		headers,
+		signal,
+		customInstructions,
+		replaceInstructions,
+		reserveTokens = 16384,
+		streamFn,
+	} = options;
 
 	// Token budget = context window minus reserved space for prompt + response
 	const contextWindow = model.contextWindow || 128000;
@@ -320,12 +331,14 @@ export async function generateBranchSummary(
 		},
 	];
 
-	// Call LLM for summarization
-	const response = await completeSimple(
-		model,
-		{ systemPrompt: SUMMARIZATION_SYSTEM_PROMPT, messages: summarizationMessages },
-		{ apiKey, headers, signal, maxTokens: 2048 },
-	);
+	// Call LLM for summarization. Prefer the session stream function so SDK
+	// request behavior (timeouts, retries, attribution headers) stays consistent
+	// without running through agent state/events.
+	const context = { systemPrompt: SUMMARIZATION_SYSTEM_PROMPT, messages: summarizationMessages };
+	const requestOptions: SimpleStreamOptions = { apiKey, headers, signal, maxTokens: 2048 };
+	const response = streamFn
+		? await (await streamFn(model, context, requestOptions)).result()
+		: await completeSimple(model, context, requestOptions);
 
 	// Check if aborted or errored
 	if (response.stopReason === "aborted") {
diff --git a/packages/coding-agent/src/core/model-resolver.ts b/packages/coding-agent/src/core/model-resolver.ts
index 87f7354e7..e3af076da 100644
--- a/packages/coding-agent/src/core/model-resolver.ts
+++ b/packages/coding-agent/src/core/model-resolver.ts
@@ -17,6 +17,7 @@ export const defaultModelPerProvider: Record<KnownProvider, string> = {
 	openai: "gpt-5.4",
 	"azure-openai-responses": "gpt-5.4",
 	"openai-codex": "gpt-5.5",
+	nvidia: "nvidia/nemotron-3-super-120b-a12b",
 	deepseek: "deepseek-v4-pro",
 	google: "gemini-3.1-pro-preview",
 	"google-vertex": "gemini-3.1-pro-preview",
diff --git a/packages/coding-agent/src/core/provider-attribution.ts b/packages/coding-agent/src/core/provider-attribution.ts
new file mode 100644
index 000000000..91265c92d
--- /dev/null
+++ b/packages/coding-agent/src/core/provider-attribution.ts
@@ -0,0 +1,97 @@
+import type { Api, Model } from "@earendil-works/pi-ai";
+import type { SettingsManager } from "./settings-manager.ts";
+import { isInstallTelemetryEnabled } from "./telemetry.ts";
+
+const OPENROUTER_HOST = "openrouter.ai";
+const NVIDIA_NIM_HOST = "integrate.api.nvidia.com";
+const CLOUDFLARE_API_HOST = "api.cloudflare.com";
+const CLOUDFLARE_AI_GATEWAY_HOST = "gateway.ai.cloudflare.com";
+const OPENCODE_HOST = "opencode.ai";
+
+function matchesHost(baseUrl: string, expectedHost: string): boolean {
+	try {
+		return new URL(baseUrl).hostname === expectedHost;
+	} catch {
+		return false;
+	}
+}
+
+function isOpenRouterModel(model: Model<Api>): boolean {
+	return model.provider === "openrouter" || model.baseUrl.includes(OPENROUTER_HOST);
+}
+
+function isNvidiaNimModel(model: Model<Api>): boolean {
+	return model.provider === "nvidia" || matchesHost(model.baseUrl, NVIDIA_NIM_HOST);
+}
+
+function isCloudflareModel(model: Model<Api>): boolean {
+	return (
+		model.provider === "cloudflare-workers-ai" ||
+		model.provider === "cloudflare-ai-gateway" ||
+		matchesHost(model.baseUrl, CLOUDFLARE_API_HOST) ||
+		matchesHost(model.baseUrl, CLOUDFLARE_AI_GATEWAY_HOST)
+	);
+}
+
+function getDefaultAttributionHeaders(
+	model: Model<Api>,
+	settingsManager: SettingsManager,
+): Record<string, string> | undefined {
+	if (!isInstallTelemetryEnabled(settingsManager)) {
+		return undefined;
+	}
+
+	if (isOpenRouterModel(model)) {
+		return {
+			"HTTP-Referer": "https://pi.dev",
+			"X-OpenRouter-Title": "pi",
+			"X-OpenRouter-Categories": "cli-agent",
+		};
+	}
+
+	if (isNvidiaNimModel(model)) {
+		return {
+			"X-BILLING-INVOKE-ORIGIN": "Pi",
+		};
+	}
+
+	if (isCloudflareModel(model)) {
+		return {
+			"User-Agent": "pi-coding-agent",
+		};
+	}
+
+	return undefined;
+}
+
+function getSessionHeaders(model: Model<Api>, sessionId: string | undefined): Record<string, string> | undefined {
+	if (!sessionId) return undefined;
+	if (
+		model.provider !== "opencode" &&
+		model.provider !== "opencode-go" &&
+		!matchesHost(model.baseUrl, OPENCODE_HOST)
+	) {
+		return undefined;
+	}
+	return { "x-opencode-session": sessionId, "x-opencode-client": "pi" };
+}
+
+export function mergeProviderAttributionHeaders(
+	model: Model<Api>,
+	settingsManager: SettingsManager,
+	sessionId: string | undefined,
+	...headerSources: Array<Record<string, string> | undefined>
+): Record<string, string> | undefined {
+	const merged = {
+		...getSessionHeaders(model, sessionId),
+		...getDefaultAttributionHeaders(model, settingsManager),
+	};
+
+	for (const headers of headerSources) {
+		if (headers) {
+			Object.assign(merged, headers);
+		}
+	}
+
+	return Object.keys(merged).length > 0 ? merged : undefined;
+}
diff --git a/packages/coding-agent/src/core/provider-display-names.ts b/packages/coding-agent/src/core/provider-display-names.ts
index 24f7b3ced..789f31530 100644
--- a/packages/coding-agent/src/core/provider-display-names.ts
+++ b/packages/coding-agent/src/core/provider-display-names.ts
@@ -17,6 +17,7 @@ export const BUILT_IN_PROVIDER_DISPLAY_NAMES: Record<string, string> = {
 	"minimax-cn": "MiniMax (China)",
 	moonshotai: "Moonshot AI",
 	"moonshotai-cn": "Moonshot AI (China)",
+	nvidia: "NVIDIA NIM",
 	opencode: "OpenCode Zen",
 	"opencode-go": "OpenCode Go",
 	openai: "OpenAI",
diff --git a/packages/coding-agent/src/core/sdk.ts b/packages/coding-agent/src/core/sdk.ts
index 7a1c922b5..e6db747b7 100644
--- a/packages/coding-agent/src/core/sdk.ts
+++ b/packages/coding-agent/src/core/sdk.ts
@@ -11,11 +11,11 @@ import type { ExtensionRunner, LoadExtensionsResult, SessionStartEvent, ToolDefi
 import { convertToLlm } from "./messages.ts";
 import { ModelRegistry } from "./model-registry.ts";
 import { findInitialModel } from "./model-resolver.ts";
+import { mergeProviderAttributionHeaders } from "./provider-attribution.ts";
 import type { ResourceLoader } from "./resource-loader.ts";
 import { DefaultResourceLoader } from "./resource-loader.ts";
 import { getDefaultSessionDir, SessionManager } from "./session-manager.ts";
 import { SettingsManager } from "./settings-manager.ts";
-import { isInstallTelemetryEnabled } from "./telemetry.ts";
 import { time } from "./timings.ts";
 import {
 	createBashTool,
@@ -128,44 +128,6 @@ function getDefaultAgentDir(): string {
 	return getAgentDir();
 }
 
-function getAttributionHeaders(
-	model: Model<any>,
-	settingsManager: SettingsManager,
-	sessionId?: string,
-): Record<string, string> | undefined {
-	if (
-		sessionId &&
-		(model.provider === "opencode" || model.provider === "opencode-go" || model.baseUrl.includes("opencode.ai"))
-	) {
-		return { "x-opencode-session": sessionId, "x-opencode-client": "pi" };
-	}
-
-	if (!isInstallTelemetryEnabled(settingsManager)) {
-		return undefined;
-	}
-
-	if (model.provider === "openrouter" || model.baseUrl.includes("openrouter.ai")) {
-		return {
-			"HTTP-Referer": "https://pi.dev",
-			"X-OpenRouter-Title": "pi",
-			"X-OpenRouter-Categories": "cli-agent",
-		};
-	}
-
-	if (
-		model.provider === "cloudflare-workers-ai" ||
-		model.provider === "cloudflare-ai-gateway" ||
-		model.baseUrl.includes("api.cloudflare.com") ||
-		model.baseUrl.includes("gateway.ai.cloudflare.com")
-	) {
-		return {
-			"User-Agent": "pi-coding-agent",
-		};
-	}
-
-	return undefined;
-}
-
 /**
  * Create an AgentSession with the specified options.
  *
@@ -349,7 +311,6 @@ export async function createAgentSession(options: CreateAgentSessionOptions = {}
 			const timeoutMs = options?.timeoutMs ?? providerRetrySettings.timeoutMs ?? effectiveTimeoutMs;
 			const websocketConnectTimeoutMs =
 				options?.websocketConnectTimeoutMs ?? settingsManager.getWebSocketConnectTimeoutMs();
-			const attributionHeaders = getAttributionHeaders(model, settingsManager, options?.sessionId);
 			return streamSimple(model, context, {
 				...options,
 				apiKey: auth.apiKey,
@@ -357,10 +318,13 @@ export async function createAgentSession(options: CreateAgentSessionOptions = {}
 				websocketConnectTimeoutMs,
 				maxRetries: options?.maxRetries ?? providerRetrySettings.maxRetries,
 				maxRetryDelayMs: options?.maxRetryDelayMs ?? providerRetrySettings.maxRetryDelayMs,
-				headers:
-					attributionHeaders || auth.headers || options?.headers
-						? { ...attributionHeaders, ...auth.headers, ...options?.headers }
-						: undefined,
+				headers: mergeProviderAttributionHeaders(
+					model,
+					settingsManager,
+					options?.sessionId,
+					auth.headers,
+					options?.headers,
+				),
 			});
 		},
 		onPayload: async (payload, _model) => {
diff --git a/packages/coding-agent/test/sdk-openrouter-attribution.test.ts b/packages/coding-agent/test/sdk-openrouter-attribution.test.ts
index c70321030..9f035189a 100644
--- a/packages/coding-agent/test/sdk-openrouter-attribution.test.ts
+++ b/packages/coding-agent/test/sdk-openrouter-attribution.test.ts
@@ -15,14 +15,14 @@ import { createAgentSession } from "../src/core/sdk.ts";
 import { SessionManager } from "../src/core/session-manager.ts";
 import { SettingsManager } from "../src/core/settings-manager.ts";
 
-describe("createAgentSession OpenRouter attribution headers", () => {
+describe("createAgentSession provider attribution headers", () => {
 	let tempDir: string;
 	let cwd: string;
 	let agentDir: string;
 	let originalTelemetryEnv: string | undefined;
 
 	beforeEach(() => {
-		tempDir = join(tmpdir(), `pi-sdk-openrouter-test-${Date.now()}-${Math.random().toString(36).slice(2)}`);
+		tempDir = join(tmpdir(), `pi-sdk-attribution-test-${Date.now()}-${Math.random().toString(36).slice(2)}`);
 		cwd = join(tempDir, "project");
 		agentDir = join(tempDir, "agent");
 		mkdirSync(cwd, { recursive: true });
@@ -42,9 +42,9 @@ describe("createAgentSession OpenRouter attribution headers", () => {
 		}
 	});
 
-	function createModel(provider: string, baseUrl: string): Model<Api> {
+	function createModel(provider: string, baseUrl: string, id = `${provider}-test-model`): Model<Api> {
 		return {
-			id: `${provider}-test-model`,
+			id,
 			name: `${provider} Test Model`,
 			api: "openai-completions",
 			provider,
@@ -172,6 +172,14 @@ describe("createAgentSession OpenRouter attribution headers", () => {
 		expect(headers?.["X-OpenRouter-Categories"]).toBe("cli-agent");
 	});
 
+	it("preserves legacy OpenRouter base URL substring attribution matching", async () => {
+		const headers = await captureHeaders(createModel("custom-openrouter", "not-a-url-openrouter.ai"));
+
+		expect(headers?.["HTTP-Referer"]).toBe("https://pi.dev");
+		expect(headers?.["X-OpenRouter-Title"]).toBe("pi");
+		expect(headers?.["X-OpenRouter-Categories"]).toBe("cli-agent");
+	});
+
 	it("lets provider and request headers override the defaults", async () => {
 		const headers = await captureHeaders(createModel("openrouter", "https://openrouter.ai/api/v1"), {
 			providerHeaders: {
@@ -188,6 +196,56 @@ describe("createAgentSession OpenRouter attribution headers", () => {
 		expect(headers?.["X-OpenRouter-Categories"]).toBe("provider-category");
 	});
 
+	it("adds default attribution headers for direct NVIDIA NIM endpoints", async () => {
+		const headers = await captureHeaders(createModel("custom-nim", "https://integrate.api.nvidia.com/v1"));
+
+		expect(headers?.["X-BILLING-INVOKE-ORIGIN"]).toBe("Pi");
+	});
+
+	it("adds default attribution headers for the NVIDIA provider", async () => {
+		const headers = await captureHeaders(createModel("nvidia", "https://example.test/v1"));
+
+		expect(headers?.["X-BILLING-INVOKE-ORIGIN"]).toBe("Pi");
+	});
+
+	it("does not add NVIDIA NIM attribution headers when telemetry is disabled", async () => {
+		const headers = await captureHeaders(createModel("nvidia", "https://integrate.api.nvidia.com/v1"), {
+			telemetryEnabled: false,
+		});
+
+		expect(headers?.["X-BILLING-INVOKE-ORIGIN"]).toBeUndefined();
+	});
+
+	it("lets provider and request headers override NVIDIA NIM defaults", async () => {
+		const headers = await captureHeaders(createModel("nvidia", "https://integrate.api.nvidia.com/v1"), {
+			providerHeaders: {
+				"X-BILLING-INVOKE-ORIGIN": "Provider",
+			},
+			requestHeaders: {
+				"X-BILLING-INVOKE-ORIGIN": "Request",
+			},
+		});
+
+		expect(headers?.["X-BILLING-INVOKE-ORIGIN"]).toBe("Request");
+	});
+
+	it("does not add NVIDIA NIM attribution headers for NVIDIA models routed through OpenRouter", async () => {
+		const headers = await captureHeaders(
+			createModel("openrouter", "https://openrouter.ai/api/v1", "nvidia/nemotron-3-super-120b-a12b"),
+		);
+
+		expect(headers?.["HTTP-Referer"]).toBe("https://pi.dev");
+		expect(headers?.["X-BILLING-INVOKE-ORIGIN"]).toBeUndefined();
+	});
+
+	it("does not add NVIDIA NIM attribution headers for NVIDIA models routed through Vercel AI Gateway", async () => {
+		const headers = await captureHeaders(
+			createModel("vercel-ai-gateway", "https://ai-gateway.vercel.sh/v1", "nvidia/nemotron-3-super-120b-a12b"),
+		);
+
+		expect(headers?.["X-BILLING-INVOKE-ORIGIN"]).toBeUndefined();
+	});
+
 	it("adds OpenCode session headers", async () => {
 		const headers = await captureHeaders(createModel("opencode", "https://opencode.ai/zen/v1"), {
 			sessionId: "opencode-session",

From 6e269edfd0497cc124482e66acde7e62de2509f1 Mon Sep 17 00:00:00 2001
From: Mario Zechner <badlogicgames@gmail.com>
Date: Tue, 2 Jun 2026 15:10:36 +0200
Subject: [PATCH 02/10] Restore NVIDIA Qwen 3.5 122B model

---
 packages/ai/scripts/generate-models.ts |  1 -
 packages/ai/src/models.generated.ts    | 19 +++++++++++++++++++
 2 files changed, 19 insertions(+), 1 deletion(-)

diff --git a/packages/ai/scripts/generate-models.ts b/packages/ai/scripts/generate-models.ts
index 8fb62d145..12c926d13 100644
--- a/packages/ai/scripts/generate-models.ts
+++ b/packages/ai/scripts/generate-models.ts
@@ -150,7 +150,6 @@ const NVIDIA_NIM_UNSUPPORTED_MODELS = new Set([
 	"mistralai/mistral-nemotron",
 	"nvidia/nemotron-mini-4b-instruct",
 	"qwen/qwen3-next-80b-a3b-instruct",
-	"qwen/qwen3.5-122b-a10b",
 	"qwen/qwen3.5-397b-a17b",
 	"sarvamai/sarvam-m",
 	"upstage/solar-10.7b-instruct",
diff --git a/packages/ai/src/models.generated.ts b/packages/ai/src/models.generated.ts
index cd022b795..1dacf48d5 100644
--- a/packages/ai/src/models.generated.ts
+++ b/packages/ai/src/models.generated.ts
@@ -6640,6 +6640,25 @@ export const MODELS = {
 			contextWindow: 262144,
 			maxTokens: 66536,
 		} satisfies Model<"openai-completions">,
+		"qwen/qwen3.5-122b-a10b": {
+			id: "qwen/qwen3.5-122b-a10b",
+			name: "Qwen3.5 122B-A10B",
+			api: "openai-completions",
+			provider: "nvidia",
+			baseUrl: "https://integrate.api.nvidia.com/v1",
+			headers: {"NVCF-POLL-SECONDS":"3600"},
+			compat: {"supportsStore":false,"supportsDeveloperRole":false,"supportsReasoningEffort":false,"maxTokensField":"max_tokens","supportsStrictMode":false,"supportsLongCacheRetention":false},
+			reasoning: true,
+			input: ["text", "image"],
+			cost: {
+				input: 0,
+				output: 0,
+				cacheRead: 0,
+				cacheWrite: 0,
+			},
+			contextWindow: 262144,
+			maxTokens: 65536,
+		} satisfies Model<"openai-completions">,
 		"stepfun-ai/step-3.5-flash": {
 			id: "stepfun-ai/step-3.5-flash",
 			name: "Step 3.5 Flash",

From 0d38e17b6894b689bb09e1cc3c944d82726ab396 Mon Sep 17 00:00:00 2001
From: Mario Zechner <badlogicgames@gmail.com>
Date: Tue, 2 Jun 2026 15:25:27 +0200
Subject: [PATCH 03/10] Fix empty self-rendered tool rows

closes #5299
---
 packages/coding-agent/CHANGELOG.md            |  1 +
 .../interactive/components/tool-execution.ts  | 25 +++++++++++++++
 .../test/tool-execution-component.test.ts     | 31 +++++++++++++++++++
 3 files changed, 57 insertions(+)

diff --git a/packages/coding-agent/CHANGELOG.md b/packages/coding-agent/CHANGELOG.md
index 28401d532..a9f3c9268 100644
--- a/packages/coding-agent/CHANGELOG.md
+++ b/packages/coding-agent/CHANGELOG.md
@@ -13,6 +13,7 @@
 - Fixed SDK embedding in bundled Node apps failing with `ENOENT` when `package.json` is not present next to the bundle entrypoint. The package metadata reader now gracefully handles missing `package.json` by using defaults, enabling `createAgentSession()` without requiring package-adjacent files at runtime ([#5226](https://github.com/earendil-works/pi/issues/5226)).
 - Fixed HTTP timeout setting not being respected for non-Codex providers (e.g., llama.cpp via OpenAI-compatible API). The `httpIdleTimeoutMs` setting (set via `/settings` HTTP timeout) now applies as the default SDK request timeout for all providers that support it, not just OpenAI Codex Responses. Disabling the timeout (HTTP timeout = false) now correctly disables SDK timeouts for all supported providers by sending a maximum int32 value (effectively infinite) instead of 0, since SDKs treat timeout=0 as an immediate timeout ([#5294](https://github.com/earendil-works/pi/issues/5294)).
 - Fixed opening and listing very large JSONL session files by reading session entries line-by-line instead of materializing the full file as one string ([#5231](https://github.com/earendil-works/pi/issues/5231)).
+- Fixed `renderShell: "self"` tool renderers that emit no component lines leaving a blank chat row ([#5299](https://github.com/earendil-works/pi/issues/5299)).
 
 ## [0.78.0] - 2026-05-29
 
diff --git a/packages/coding-agent/src/modes/interactive/components/tool-execution.ts b/packages/coding-agent/src/modes/interactive/components/tool-execution.ts
index b941dfc88..ad84f4413 100644
--- a/packages/coding-agent/src/modes/interactive/components/tool-execution.ts
+++ b/packages/coding-agent/src/modes/interactive/components/tool-execution.ts
@@ -222,6 +222,31 @@ export class ToolExecutionComponent extends Container {
 		if (this.hideComponent) {
 			return [];
 		}
+
+		if (this.hasRendererDefinition() && this.getRenderShell() === "self") {
+			const contentLines = this.selfRenderContainer.render(width);
+			if (contentLines.length === 0 && this.imageComponents.length === 0) {
+				return [];
+			}
+
+			const lines: string[] = [];
+			if (contentLines.length > 0) {
+				lines.push("");
+				lines.push(...contentLines);
+			}
+			for (let i = 0; i < this.imageComponents.length; i++) {
+				const spacer = this.imageSpacers[i];
+				if (spacer) {
+					lines.push(...spacer.render(width));
+				}
+				const imageComponent = this.imageComponents[i];
+				if (imageComponent) {
+					lines.push(...imageComponent.render(width));
+				}
+			}
+			return lines;
+		}
+
 		return super.render(width);
 	}
 
diff --git a/packages/coding-agent/test/tool-execution-component.test.ts b/packages/coding-agent/test/tool-execution-component.test.ts
index 7a4a813da..bf890959d 100644
--- a/packages/coding-agent/test/tool-execution-component.test.ts
+++ b/packages/coding-agent/test/tool-execution-component.test.ts
@@ -67,6 +67,37 @@ describe("ToolExecutionComponent parity", () => {
 		expect(rendered).toContain("custom result");
 	});
 
+	test("self-rendered empty tool rows take no layout space", () => {
+		const toolDefinition: ToolDefinition = {
+			...createBaseToolDefinition(),
+			renderShell: "self",
+			renderCall: () => new Text("", 0, 0),
+			renderResult: () => new Text("", 0, 0),
+		};
+
+		const component = new ToolExecutionComponent(
+			"custom_tool",
+			"tool-empty-self-render",
+			{},
+			{},
+			toolDefinition,
+			createFakeTui(),
+			process.cwd(),
+		);
+		expect(component.render(120)).toEqual([]);
+
+		component.updateResult(
+			{
+				content: [],
+				details: {},
+				isError: false,
+			},
+			false,
+		);
+
+		expect(component.render(120)).toEqual([]);
+	});
+
 	test("uses built-in rendering for built-in overrides without custom renderers", () => {
 		const overrideDefinition: ToolDefinition = {
 			...createBaseToolDefinition("edit"),

From 2125221bfce8e1a6fa02f8db5ff7820890f986a8 Mon Sep 17 00:00:00 2001
From: Mario Zechner <badlogicgames@gmail.com>
Date: Tue, 2 Jun 2026 15:36:52 +0200
Subject: [PATCH 04/10] Fix OpenRouter Kimi reasoning compat

closes #5309
---
 packages/ai/CHANGELOG.md                      |  1 +
 packages/ai/scripts/generate-models.ts        |  6 ++-
 packages/ai/src/models.generated.ts           |  4 +-
 .../ai/src/providers/openai-completions.ts    |  4 +-
 .../openai-completions-tool-choice.test.ts    | 37 ++++++++++++++++++-
 5 files changed, 47 insertions(+), 5 deletions(-)

diff --git a/packages/ai/CHANGELOG.md b/packages/ai/CHANGELOG.md
index 4a616f893..e927bdcaf 100644
--- a/packages/ai/CHANGELOG.md
+++ b/packages/ai/CHANGELOG.md
@@ -10,6 +10,7 @@
 
 - Fixed Amazon Bedrock requests to replace blank required user/tool-result text with a placeholder and skip blank replay text blocks ([#4975](https://github.com/earendil-works/pi/issues/4975)).
 - Fixed OpenAI GPT-5.5 generated metadata to omit unsupported minimal thinking ([#5243](https://github.com/earendil-works/pi/issues/5243)).
+- Fixed OpenRouter Kimi K2.6 thinking replay and preserved developer-role instructions for OpenRouter OpenAI and Anthropic models ([#5309](https://github.com/earendil-works/pi/issues/5309)).
 - Fixed GitHub Copilot and OpenRouter test model references that became stale after model regeneration.
 
 ## [0.78.0] - 2026-05-29
diff --git a/packages/ai/scripts/generate-models.ts b/packages/ai/scripts/generate-models.ts
index 12c926d13..b0f495711 100644
--- a/packages/ai/scripts/generate-models.ts
+++ b/packages/ai/scripts/generate-models.ts
@@ -1343,7 +1343,11 @@ async function generateModels() {
 			candidate.maxTokens = 4096;
 		}
 		if (candidate.provider === "openrouter" && candidate.id.startsWith("moonshotai/kimi-k2.6")) {
-			candidate.compat = { ...candidate.compat, supportsDeveloperRole: false };
+			candidate.compat = {
+				...candidate.compat,
+				supportsDeveloperRole: false,
+				requiresReasoningContentOnAssistantMessages: true,
+			};
 		}
 		if (candidate.provider === "openrouter" && candidate.id === "z-ai/glm-5") {
 			candidate.cost.input = 0.6;
diff --git a/packages/ai/src/models.generated.ts b/packages/ai/src/models.generated.ts
index 1dacf48d5..b3ff24a97 100644
--- a/packages/ai/src/models.generated.ts
+++ b/packages/ai/src/models.generated.ts
@@ -10281,7 +10281,7 @@ export const MODELS = {
 			api: "openai-completions",
 			provider: "openrouter",
 			baseUrl: "https://openrouter.ai/api/v1",
-			compat: {"supportsDeveloperRole":false},
+			compat: {"supportsDeveloperRole":false,"requiresReasoningContentOnAssistantMessages":true},
 			reasoning: true,
 			input: ["text", "image"],
 			cost: {
@@ -10299,7 +10299,7 @@ export const MODELS = {
 			api: "openai-completions",
 			provider: "openrouter",
 			baseUrl: "https://openrouter.ai/api/v1",
-			compat: {"supportsDeveloperRole":false},
+			compat: {"supportsDeveloperRole":false,"requiresReasoningContentOnAssistantMessages":true},
 			reasoning: true,
 			input: ["text", "image"],
 			cost: {
diff --git a/packages/ai/src/providers/openai-completions.ts b/packages/ai/src/providers/openai-completions.ts
index c58cbf46c..ae6d583e3 100644
--- a/packages/ai/src/providers/openai-completions.ts
+++ b/packages/ai/src/providers/openai-completions.ts
@@ -1100,11 +1100,13 @@ function detectCompat(model: Model<"openai-completions">): ResolvedOpenAIComplet
 
 	const isGrok = provider === "xai" || baseUrl.includes("api.x.ai");
 	const isDeepSeek = provider === "deepseek" || baseUrl.includes("deepseek.com");
+	const isOpenRouterDeveloperRoleModel =
+		isOpenRouter && (model.id.startsWith("anthropic/") || model.id.startsWith("openai/"));
 	const cacheControlFormat = provider === "openrouter" && model.id.startsWith("anthropic/") ? "anthropic" : undefined;
 
 	return {
 		supportsStore: !isNonStandard,
-		supportsDeveloperRole: !isNonStandard && !isOpenRouter,
+		supportsDeveloperRole: isOpenRouterDeveloperRoleModel || (!isNonStandard && !isOpenRouter),
 		supportsReasoningEffort: !isGrok && !isZai && !isMoonshot && !isTogether && !isCloudflareAiGateway && !isNvidia,
 		supportsUsageInStreaming: true,
 		maxTokensField: useMaxTokens ? "max_tokens" : "max_completion_tokens",
diff --git a/packages/ai/test/openai-completions-tool-choice.test.ts b/packages/ai/test/openai-completions-tool-choice.test.ts
index 5f40c8176..c06ac344e 100644
--- a/packages/ai/test/openai-completions-tool-choice.test.ts
+++ b/packages/ai/test/openai-completions-tool-choice.test.ts
@@ -817,7 +817,7 @@ describe("openai-completions tool_choice", () => {
 		expect(writeCall).not.toHaveProperty("partialArgs");
 	});
 
-	it("uses system messages for OpenRouter reasoning model instructions", async () => {
+	it("uses system messages for non-OpenAI/Anthropic OpenRouter reasoning model instructions", async () => {
 		const model = getModel("openrouter", "deepseek/deepseek-v4-pro")!;
 		let payload: unknown;
 
@@ -839,6 +839,33 @@ describe("openai-completions tool_choice", () => {
 		expect(params.messages?.[0]?.role).toBe("system");
 	});
 
+	it("keeps developer messages for OpenAI and Anthropic OpenRouter reasoning model instructions", async () => {
+		for (const model of [
+			getModel("openrouter", "openai/gpt-5.2-codex"),
+			getModel("openrouter", "anthropic/claude-sonnet-4.5"),
+		]) {
+			expect(model).toBeDefined();
+			let payload: unknown;
+
+			await streamSimple(
+				model!,
+				{
+					systemPrompt: "Follow instructions.",
+					messages: [{ role: "user", content: "Hi", timestamp: Date.now() }],
+				},
+				{
+					apiKey: "test",
+					onPayload: (params: unknown) => {
+						payload = params;
+					},
+				},
+			).result();
+
+			const params = payload as { messages?: Array<{ role?: string }> };
+			expect(params.messages?.[0]?.role).toBe("developer");
+		}
+	});
+
 	it("keeps developer messages for OpenAI reasoning model instructions", async () => {
 		const { compat: _compat, ...baseModel } = getModel("openai", "gpt-5.5")!;
 		const model = { ...baseModel, api: "openai-completions" } as const;
@@ -862,6 +889,14 @@ describe("openai-completions tool_choice", () => {
 		expect(params.messages?.[0]?.role).toBe("developer");
 	});
 
+	it("stores OpenRouter Kimi K2.6 reasoning replay compat in built-in metadata", () => {
+		for (const modelId of ["moonshotai/kimi-k2.6", "moonshotai/kimi-k2.6:free"] as const) {
+			const model = getModel("openrouter", modelId)!;
+			expect(model.compat?.supportsDeveloperRole).toBe(false);
+			expect(model.compat?.requiresReasoningContentOnAssistantMessages).toBe(true);
+		}
+	});
+
 	it("stores Xiaomi MiMo reasoning replay compat in built-in metadata", () => {
 		const providers = ["xiaomi", "xiaomi-token-plan-cn", "xiaomi-token-plan-ams", "xiaomi-token-plan-sgp"] as const;
 

From 276cb1bb89977d83651bae4ea9b00f634c1670ad Mon Sep 17 00:00:00 2001
From: Mario Zechner <badlogicgames@gmail.com>
Date: Tue, 2 Jun 2026 15:53:37 +0200
Subject: [PATCH 05/10] Fix stale SDK stream options test after
 httpIdleTimeoutMs universalization

---
 packages/coding-agent/test/sdk-stream-options.test.ts | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/packages/coding-agent/test/sdk-stream-options.test.ts b/packages/coding-agent/test/sdk-stream-options.test.ts
index e30eff7a8..1d565c7e3 100644
--- a/packages/coding-agent/test/sdk-stream-options.test.ts
+++ b/packages/coding-agent/test/sdk-stream-options.test.ts
@@ -119,10 +119,10 @@ describe("createAgentSession stream options", () => {
 		expect(options?.timeoutMs).toBe(1234);
 	});
 
-	it("does not default timeoutMs from httpIdleTimeoutMs for other providers", async () => {
+	it("defaults timeoutMs from httpIdleTimeoutMs for all providers", async () => {
 		const options = await captureStreamOptions("openai-completions", { httpIdleTimeoutMs: 1234 });
 
-		expect(options?.timeoutMs).toBeUndefined();
+		expect(options?.timeoutMs).toBe(1234);
 	});
 
 	it("lets request timeoutMs override httpIdleTimeoutMs for OpenAI Codex", async () => {

From 7e72ca47c8bff19bcec96a98745ca4a3a7624051 Mon Sep 17 00:00:00 2001
From: Mario Zechner <badlogicgames@gmail.com>
Date: Tue, 2 Jun 2026 16:00:29 +0200
Subject: [PATCH 06/10] Add MiniMax-M3 to direct minimax providers

closes #5313
---
 packages/ai/CHANGELOG.md               |  1 +
 packages/ai/scripts/generate-models.ts | 12 +-------
 packages/ai/src/models.generated.ts    | 40 ++++++++++++++++++++++++--
 3 files changed, 39 insertions(+), 14 deletions(-)

diff --git a/packages/ai/CHANGELOG.md b/packages/ai/CHANGELOG.md
index e927bdcaf..3814e2b2f 100644
--- a/packages/ai/CHANGELOG.md
+++ b/packages/ai/CHANGELOG.md
@@ -4,6 +4,7 @@
 
 ### Added
 
+- Added MiniMax-M3 model to the `minimax` and `minimax-cn` direct providers, and removed the hardcoded context-window override that was masking models.dev values ([#5313](https://github.com/earendil-works/pi/issues/5313)).
 - Added NVIDIA NIM as a built-in OpenAI-compatible provider, exposing public NIM models that support tool use.
 
 ### Fixed
diff --git a/packages/ai/scripts/generate-models.ts b/packages/ai/scripts/generate-models.ts
index b0f495711..93a664e23 100644
--- a/packages/ai/scripts/generate-models.ts
+++ b/packages/ai/scripts/generate-models.ts
@@ -1657,17 +1657,7 @@ async function generateModels() {
 		}
 	}
 
-	const minimaxDirectSupportedIds = new Set(["MiniMax-M2.7", "MiniMax-M2.7-highspeed"]);
-
-	for (const candidate of allModels) {
-		if (
-			(candidate.provider === "minimax" || candidate.provider === "minimax-cn") &&
-			minimaxDirectSupportedIds.has(candidate.id)
-		) {
-			candidate.contextWindow = 204800;
-			candidate.maxTokens = 131072;
-		}
-	}
+	const minimaxDirectSupportedIds = new Set(["MiniMax-M2.7", "MiniMax-M2.7-highspeed", "MiniMax-M3"]);
 
 	for (let i = allModels.length - 1; i >= 0; i--) {
 		const candidate = allModels[i];
diff --git a/packages/ai/src/models.generated.ts b/packages/ai/src/models.generated.ts
index b3ff24a97..97001d3a3 100644
--- a/packages/ai/src/models.generated.ts
+++ b/packages/ai/src/models.generated.ts
@@ -5564,6 +5564,23 @@ export const MODELS = {
 			contextWindow: 204800,
 			maxTokens: 131072,
 		} satisfies Model<"anthropic-messages">,
+		"MiniMax-M3": {
+			id: "MiniMax-M3",
+			name: "MiniMax-M3",
+			api: "anthropic-messages",
+			provider: "minimax",
+			baseUrl: "https://api.minimax.io/anthropic",
+			reasoning: true,
+			input: ["text", "image"],
+			cost: {
+				input: 0.3,
+				output: 1.2,
+				cacheRead: 0.06,
+				cacheWrite: 0.375,
+			},
+			contextWindow: 512000,
+			maxTokens: 131072,
+		} satisfies Model<"anthropic-messages">,
 	},
 	"minimax-cn": {
 		"MiniMax-M2.7": {
@@ -5600,6 +5617,23 @@ export const MODELS = {
 			contextWindow: 204800,
 			maxTokens: 131072,
 		} satisfies Model<"anthropic-messages">,
+		"MiniMax-M3": {
+			id: "MiniMax-M3",
+			name: "MiniMax-M3",
+			api: "anthropic-messages",
+			provider: "minimax-cn",
+			baseUrl: "https://api.minimaxi.com/anthropic",
+			reasoning: true,
+			input: ["text", "image"],
+			cost: {
+				input: 0.3,
+				output: 1.2,
+				cacheRead: 0.06,
+				cacheWrite: 0.375,
+			},
+			contextWindow: 512000,
+			maxTokens: 131072,
+		} satisfies Model<"anthropic-messages">,
 	},
 	"mistral": {
 		"codestral-latest": {
@@ -9689,9 +9723,9 @@ export const MODELS = {
 			reasoning: true,
 			input: ["text"],
 			cost: {
-				input: 0.3,
-				output: 2.5,
-				cacheRead: 0.06,
+				input: 0.075,
+				output: 0.625,
+				cacheRead: 0.015,
 				cacheWrite: 0,
 			},
 			contextWindow: 262144,

From 6cb23f9b5d5b6d1747672f535b167d0d809ac010 Mon Sep 17 00:00:00 2001
From: Mario Zechner <badlogicgames@gmail.com>
Date: Tue, 2 Jun 2026 16:30:14 +0200
Subject: [PATCH 07/10] Fix HTML export URL sanitization

---
 packages/coding-agent/CHANGELOG.md            |  1 +
 .../src/core/export-html/template.js          | 25 ++++++++++++++-----
 .../coding-agent/test/export-html-xss.test.ts | 15 +++++++----
 3 files changed, 30 insertions(+), 11 deletions(-)

diff --git a/packages/coding-agent/CHANGELOG.md b/packages/coding-agent/CHANGELOG.md
index a9f3c9268..82c509e41 100644
--- a/packages/coding-agent/CHANGELOG.md
+++ b/packages/coding-agent/CHANGELOG.md
@@ -10,6 +10,7 @@
 
 ### Fixed
 
+- Fixed stored XSS in HTML session exports by sanitizing Markdown link and image URLs with a scheme allow-list after stripping control characters.
 - Fixed SDK embedding in bundled Node apps failing with `ENOENT` when `package.json` is not present next to the bundle entrypoint. The package metadata reader now gracefully handles missing `package.json` by using defaults, enabling `createAgentSession()` without requiring package-adjacent files at runtime ([#5226](https://github.com/earendil-works/pi/issues/5226)).
 - Fixed HTTP timeout setting not being respected for non-Codex providers (e.g., llama.cpp via OpenAI-compatible API). The `httpIdleTimeoutMs` setting (set via `/settings` HTTP timeout) now applies as the default SDK request timeout for all providers that support it, not just OpenAI Codex Responses. Disabling the timeout (HTTP timeout = false) now correctly disables SDK timeouts for all supported providers by sending a maximum int32 value (effectively infinite) instead of 0, since SDKs treat timeout=0 as an immediate timeout ([#5294](https://github.com/earendil-works/pi/issues/5294)).
 - Fixed opening and listing very large JSONL session files by reading session entries line-by-line instead of materializing the full file as one string ([#5231](https://github.com/earendil-works/pi/issues/5231)).
diff --git a/packages/coding-agent/src/core/export-html/template.js b/packages/coding-agent/src/core/export-html/template.js
index fe5c6d91c..cfdd16120 100644
--- a/packages/coding-agent/src/core/export-html/template.js
+++ b/packages/coding-agent/src/core/export-html/template.js
@@ -613,6 +613,18 @@
           .replace(/'/g, '&#39;');
       }
 
+      function sanitizeMarkdownUrl(value) {
+        const href = String(value || '').trim().replace(/[\x00-\x1f\x7f]/g, '');
+        if (!href) return href;
+
+        const scheme = href.match(/^([A-Za-z][A-Za-z0-9+.-]*):/);
+        if (scheme && !/^(https?|mailto|tel|ftp)$/i.test(scheme[1])) {
+          return null;
+        }
+
+        return href;
+      }
+
       /**
        * Truncate string to maxLen chars, append "..." if truncated.
        */
@@ -1569,10 +1581,11 @@
           }
         },
         renderer: {
-          // Sanitize link URLs to prevent javascript:/vbscript:/data: XSS
+          // Sanitize link URLs with a scheme allow-list. Browsers strip C0
+          // controls from schemes, so strip them before checking and emitting.
           link(token) {
-            const href = (token.href || '').trim();
-            if (/^\s*(javascript|vbscript|data):/i.test(href)) {
+            const href = sanitizeMarkdownUrl(token.href);
+            if (href === null) {
               return this.parser.parseInline(token.tokens);
             }
             let out = '<a href="' + escapeHtml(href) + '"';
@@ -1582,10 +1595,10 @@
             out += '>' + this.parser.parseInline(token.tokens) + '</a>';
             return out;
           },
-          // Sanitize image src URLs
+          // Sanitize image src URLs with the same scheme allow-list.
           image(token) {
-            const href = (token.href || '').trim();
-            if (/^\s*(javascript|vbscript|data):/i.test(href)) {
+            const href = sanitizeMarkdownUrl(token.href);
+            if (href === null) {
               return escapeHtml(token.text || '');
             }
             let out = '<img src="' + escapeHtml(href) + '" alt="' + escapeHtml(token.text || '') + '"';
diff --git a/packages/coding-agent/test/export-html-xss.test.ts b/packages/coding-agent/test/export-html-xss.test.ts
index 5c6ffe872..da3c19f38 100644
--- a/packages/coding-agent/test/export-html-xss.test.ts
+++ b/packages/coding-agent/test/export-html-xss.test.ts
@@ -4,15 +4,20 @@ import { describe, expect, it } from "vitest";
 describe("export HTML markdown link sanitization", () => {
 	const templateJs = readFileSync(new URL("../src/core/export-html/template.js", import.meta.url), "utf-8");
 
-	it("overrides the marked link renderer to block javascript: protocol", () => {
-		// The custom link renderer must check for dangerous protocols
+	it("overrides the marked link renderer to use scheme allow-list sanitization", () => {
 		expect(templateJs).toMatch(/link\s*\(\s*token\s*\)/);
-		expect(templateJs).toMatch(/javascript/i);
-		expect(templateJs).toMatch(/vbscript/i);
+		expect(templateJs).toMatch(/sanitizeMarkdownUrl\(token\.href\)/);
+		expect(templateJs).toMatch(/\^\(https\?\|mailto\|tel\|ftp\)/);
 	});
 
-	it("overrides the marked image renderer to block javascript: protocol", () => {
+	it("overrides the marked image renderer to use scheme allow-list sanitization", () => {
 		expect(templateJs).toMatch(/image\s*\(\s*token\s*\)/);
+		expect(templateJs).toMatch(/sanitizeMarkdownUrl\(token\.href\)/);
+	});
+
+	it("strips C0 controls before checking and emitting markdown URLs", () => {
+		expect(templateJs).toContain("replace(/[\\x00-\\x1f\\x7f]/g, '')");
+		expect(templateJs).not.toMatch(/\^\\s\*\(javascript\|vbscript\|data\):/i);
 	});
 
 	it("escapes href attributes in the custom link renderer", () => {

From 25a4a8ed1e16c28fa7eceefdff78cbb00e18867e Mon Sep 17 00:00:00 2001
From: Mario Zechner <badlogicgames@gmail.com>
Date: Tue, 2 Jun 2026 16:37:44 +0200
Subject: [PATCH 08/10] Add Ant Ling provider

---
 packages/ai/CHANGELOG.md                      |  1 +
 packages/ai/README.md                         |  4 +-
 packages/ai/scripts/generate-models.ts        | 63 +++++++++++++
 packages/ai/src/env-api-keys.ts               |  1 +
 packages/ai/src/models.generated.ts           | 57 ++++++++++++
 .../ai/src/providers/openai-completions.ts    | 31 +++++--
 packages/ai/src/types.ts                      |  6 +-
 .../openai-completions-tool-choice.test.ts    | 93 ++++++++++++++++++-
 packages/ai/test/stream.test.ts               | 21 +++++
 packages/coding-agent/CHANGELOG.md            |  1 +
 packages/coding-agent/README.md               |  1 +
 packages/coding-agent/docs/providers.md       |  2 +
 packages/coding-agent/src/cli/args.ts         |  1 +
 .../coding-agent/src/core/model-resolver.ts   |  1 +
 .../src/core/provider-display-names.ts        |  1 +
 .../coding-agent/test/model-resolver.test.ts  |  3 +-
 16 files changed, 275 insertions(+), 12 deletions(-)

diff --git a/packages/ai/CHANGELOG.md b/packages/ai/CHANGELOG.md
index 3814e2b2f..6e6769be3 100644
--- a/packages/ai/CHANGELOG.md
+++ b/packages/ai/CHANGELOG.md
@@ -4,6 +4,7 @@
 
 ### Added
 
+- Added Ant Ling as a built-in OpenAI-compatible provider with Ling 2.6 and Ring 2.6 models.
 - Added MiniMax-M3 model to the `minimax` and `minimax-cn` direct providers, and removed the hardcoded context-window override that was masking models.dev values ([#5313](https://github.com/earendil-works/pi/issues/5313)).
 - Added NVIDIA NIM as a built-in OpenAI-compatible provider, exposing public NIM models that support tool use.
 
diff --git a/packages/ai/README.md b/packages/ai/README.md
index 76a987692..8f65bc727 100644
--- a/packages/ai/README.md
+++ b/packages/ai/README.md
@@ -51,6 +51,7 @@ Unified LLM API with automatic model discovery, provider configuration, token an
 ## Supported Providers
 
 - **OpenAI**
+- **Ant Ling**
 - **Azure OpenAI (Responses)**
 - **OpenAI Codex** (ChatGPT Plus/Pro subscription, requires OAuth, see below)
 - **DeepSeek**
@@ -939,7 +940,7 @@ interface OpenAICompletionsCompat {
   requiresAssistantAfterToolResult?: boolean; // Whether tool results must be followed by an assistant message (default: false)
   requiresThinkingAsText?: boolean;  // Whether thinking blocks must be converted to text (default: false)
   requiresReasoningContentOnAssistantMessages?: boolean; // Whether all replayed assistant messages must include empty reasoning_content when reasoning is enabled (default: auto-detected for DeepSeek)
-  thinkingFormat?: 'openai' | 'openrouter' | 'deepseek' | 'together' | 'zai' | 'qwen' | 'qwen-chat-template'; // Format for reasoning param: 'openai' uses reasoning_effort, 'openrouter' uses reasoning: { effort }, 'deepseek' uses thinking: { type } plus reasoning_effort when supported, 'together' uses reasoning: { enabled } plus reasoning_effort when supported, 'zai' uses enable_thinking, 'qwen' uses enable_thinking, 'qwen-chat-template' uses chat_template_kwargs.enable_thinking (default: openai)
+  thinkingFormat?: 'openai' | 'openrouter' | 'deepseek' | 'together' | 'zai' | 'qwen' | 'qwen-chat-template' | 'string-thinking' | 'ant-ling'; // Format for reasoning param: 'openai' uses reasoning_effort, 'openrouter' uses reasoning: { effort }, 'deepseek' uses thinking: { type } plus reasoning_effort when supported, 'together' uses reasoning: { enabled } plus reasoning_effort when supported, 'zai' uses enable_thinking, 'qwen' uses enable_thinking, 'qwen-chat-template' uses chat_template_kwargs.enable_thinking, 'string-thinking' uses top-level thinking, 'ant-ling' uses reasoning: { effort } only for mapped efforts (default: openai)
   cacheControlFormat?: 'anthropic';  // Anthropic-style cache_control on system prompt, last tool, and last user/assistant text content
   openRouterRouting?: OpenRouterRouting; // OpenRouter routing preferences (default: {})
   vercelGatewayRouting?: VercelGatewayRouting; // Vercel AI Gateway routing preferences (default: {})
@@ -1100,6 +1101,7 @@ In Node.js environments, you can set environment variables to avoid passing API
 | Provider | Environment Variable(s) |
 |----------|------------------------|
 | OpenAI | `OPENAI_API_KEY` |
+| Ant Ling | `ANT_LING_API_KEY` |
 | Azure OpenAI | `AZURE_OPENAI_API_KEY` + `AZURE_OPENAI_BASE_URL` (e.g. `https://{resource}.openai.azure.com`) or `AZURE_OPENAI_RESOURCE_NAME`. Supports `*.openai.azure.com` and `*.cognitiveservices.azure.com`; root endpoints auto-normalize to `/openai/v1`. Optional: `AZURE_OPENAI_API_VERSION` (default `v1`), `AZURE_OPENAI_DEPLOYMENT_NAME_MAP`. |
 | Anthropic | `ANTHROPIC_API_KEY` or `ANTHROPIC_OAUTH_TOKEN` |
 | DeepSeek | `DEEPSEEK_API_KEY` |
diff --git a/packages/ai/scripts/generate-models.ts b/packages/ai/scripts/generate-models.ts
index 93a664e23..c98f868cd 100644
--- a/packages/ai/scripts/generate-models.ts
+++ b/packages/ai/scripts/generate-models.ts
@@ -169,6 +169,15 @@ const DEEPSEEK_V4_THINKING_LEVEL_MAP = {
 	xhigh: "max",
 } as const;
 
+const ANT_LING_RING_THINKING_LEVEL_MAP = {
+	off: null,
+	minimal: null,
+	low: null,
+	medium: null,
+	high: "high",
+	xhigh: "xhigh",
+} as const;
+
 const OPENAI_RESPONSES_NONE_REASONING_MODELS = new Set([
 	"gpt-5.1",
 	"gpt-5.2",
@@ -330,6 +339,10 @@ function applyThinkingLevelMetadata(model: Model<any>): void {
 		// OpenCode Zen Grok Build reasons by default but rejects explicit reasoningEffort.
 		mergeThinkingLevelMap(model, { off: null, minimal: null, low: null, medium: null });
 	}
+	if (model.provider === "ant-ling" && model.reasoning) {
+		// Ring reasons by default. Only high/xhigh have documented explicit effort controls.
+		mergeThinkingLevelMap(model, ANT_LING_RING_THINKING_LEVEL_MAP);
+	}
 }
 
 function getAnthropicMessagesCompat(provider: string, modelId: string): AnthropicMessagesCompat | undefined {
@@ -1643,6 +1656,56 @@ async function generateModels() {
 	];
 	allModels.push(...deepseekV4Models);
 
+	const antLingCompat: OpenAICompletionsCompat = {
+		supportsStore: false,
+		supportsDeveloperRole: false,
+		supportsReasoningEffort: false,
+		maxTokensField: "max_tokens",
+		supportsLongCacheRetention: false,
+	};
+	const antLingModels: Model<"openai-completions">[] = [
+		{
+			id: "Ling-2.6-flash",
+			name: "Ling 2.6 Flash",
+			api: "openai-completions",
+			baseUrl: "https://api.ant-ling.com/v1",
+			provider: "ant-ling",
+			reasoning: false,
+			input: ["text"],
+			cost: { input: 0.01, output: 0.02, cacheRead: 0, cacheWrite: 0 },
+			contextWindow: 262144,
+			maxTokens: 65536,
+			compat: antLingCompat,
+		},
+		{
+			id: "Ling-2.6-1T",
+			name: "Ling 2.6 1T",
+			api: "openai-completions",
+			baseUrl: "https://api.ant-ling.com/v1",
+			provider: "ant-ling",
+			reasoning: false,
+			input: ["text"],
+			cost: { input: 0.06, output: 0.25, cacheRead: 0, cacheWrite: 0 },
+			contextWindow: 262144,
+			maxTokens: 65536,
+			compat: antLingCompat,
+		},
+		{
+			id: "Ring-2.6-1T",
+			name: "Ring 2.6 1T",
+			api: "openai-completions",
+			baseUrl: "https://api.ant-ling.com/v1",
+			provider: "ant-ling",
+			reasoning: true,
+			input: ["text"],
+			cost: { input: 0.06, output: 0.25, cacheRead: 0, cacheWrite: 0 },
+			contextWindow: 262144,
+			maxTokens: 65536,
+			compat: { ...antLingCompat, thinkingFormat: "ant-ling" },
+		},
+	];
+	allModels.push(...antLingModels);
+
 	for (const candidate of allModels) {
 		if (candidate.api === "openai-completions" && candidate.id.includes("deepseek-v4")) {
 			candidate.compat = {
diff --git a/packages/ai/src/env-api-keys.ts b/packages/ai/src/env-api-keys.ts
index 6321a03a1..510f3b387 100644
--- a/packages/ai/src/env-api-keys.ts
+++ b/packages/ai/src/env-api-keys.ts
@@ -99,6 +99,7 @@ function getApiKeyEnvVars(provider: string): readonly string[] | undefined {
 	}
 
 	const envMap: Record<string, string> = {
+		"ant-ling": "ANT_LING_API_KEY",
 		openai: "OPENAI_API_KEY",
 		"azure-openai-responses": "AZURE_OPENAI_API_KEY",
 		nvidia: "NVIDIA_API_KEY",
diff --git a/packages/ai/src/models.generated.ts b/packages/ai/src/models.generated.ts
index 97001d3a3..54acc9529 100644
--- a/packages/ai/src/models.generated.ts
+++ b/packages/ai/src/models.generated.ts
@@ -1552,6 +1552,63 @@ export const MODELS = {
 			maxTokens: 101376,
 		} satisfies Model<"bedrock-converse-stream">,
 	},
+	"ant-ling": {
+		"Ling-2.6-1T": {
+			id: "Ling-2.6-1T",
+			name: "Ling 2.6 1T",
+			api: "openai-completions",
+			provider: "ant-ling",
+			baseUrl: "https://api.ant-ling.com/v1",
+			compat: {"supportsStore":false,"supportsDeveloperRole":false,"supportsReasoningEffort":false,"maxTokensField":"max_tokens","supportsLongCacheRetention":false},
+			reasoning: false,
+			input: ["text"],
+			cost: {
+				input: 0.06,
+				output: 0.25,
+				cacheRead: 0,
+				cacheWrite: 0,
+			},
+			contextWindow: 262144,
+			maxTokens: 65536,
+		} satisfies Model<"openai-completions">,
+		"Ling-2.6-flash": {
+			id: "Ling-2.6-flash",
+			name: "Ling 2.6 Flash",
+			api: "openai-completions",
+			provider: "ant-ling",
+			baseUrl: "https://api.ant-ling.com/v1",
+			compat: {"supportsStore":false,"supportsDeveloperRole":false,"supportsReasoningEffort":false,"maxTokensField":"max_tokens","supportsLongCacheRetention":false},
+			reasoning: false,
+			input: ["text"],
+			cost: {
+				input: 0.01,
+				output: 0.02,
+				cacheRead: 0,
+				cacheWrite: 0,
+			},
+			contextWindow: 262144,
+			maxTokens: 65536,
+		} satisfies Model<"openai-completions">,
+		"Ring-2.6-1T": {
+			id: "Ring-2.6-1T",
+			name: "Ring 2.6 1T",
+			api: "openai-completions",
+			provider: "ant-ling",
+			baseUrl: "https://api.ant-ling.com/v1",
+			compat: {"supportsStore":false,"supportsDeveloperRole":false,"supportsReasoningEffort":false,"maxTokensField":"max_tokens","supportsLongCacheRetention":false,"thinkingFormat":"ant-ling"},
+			reasoning: true,
+			thinkingLevelMap: {"off":null,"minimal":null,"low":null,"medium":null,"high":"high","xhigh":"xhigh"},
+			input: ["text"],
+			cost: {
+				input: 0.06,
+				output: 0.25,
+				cacheRead: 0,
+				cacheWrite: 0,
+			},
+			contextWindow: 262144,
+			maxTokens: 65536,
+		} satisfies Model<"openai-completions">,
+	},
 	"anthropic": {
 		"claude-3-5-haiku-20241022": {
 			id: "claude-3-5-haiku-20241022",
diff --git a/packages/ai/src/providers/openai-completions.ts b/packages/ai/src/providers/openai-completions.ts
index ae6d583e3..5093d3e88 100644
--- a/packages/ai/src/providers/openai-completions.ts
+++ b/packages/ai/src/providers/openai-completions.ts
@@ -578,6 +578,11 @@ function buildParams(
 		} else if (model.thinkingLevelMap?.off !== null) {
 			openRouterParams.reasoning = { effort: model.thinkingLevelMap?.off ?? "none" };
 		}
+	} else if (compat.thinkingFormat === "ant-ling" && model.reasoning && options?.reasoningEffort) {
+		const effort = model.thinkingLevelMap?.[options.reasoningEffort];
+		if (typeof effort === "string") {
+			(params as typeof params & { reasoning?: { effort: string } }).reasoning = { effort };
+		}
 	} else if (compat.thinkingFormat === "together" && model.reasoning) {
 		const togetherParams = params as Omit<typeof params, "reasoning_effort"> & {
 			reasoning?: { enabled: boolean };
@@ -1079,6 +1084,7 @@ function detectCompat(model: Model<"openai-completions">): ResolvedOpenAIComplet
 	const isCloudflareWorkersAI = provider === "cloudflare-workers-ai" || baseUrl.includes("api.cloudflare.com");
 	const isCloudflareAiGateway = provider === "cloudflare-ai-gateway" || baseUrl.includes("gateway.ai.cloudflare.com");
 	const isNvidia = provider === "nvidia" || baseUrl.includes("integrate.api.nvidia.com");
+	const isAntLing = provider === "ant-ling" || baseUrl.includes("api.ant-ling.com");
 
 	const isNonStandard =
 		isNvidia ||
@@ -1094,9 +1100,11 @@ function detectCompat(model: Model<"openai-completions">): ResolvedOpenAIComplet
 		provider === "opencode" ||
 		baseUrl.includes("opencode.ai") ||
 		isCloudflareWorkersAI ||
-		isCloudflareAiGateway;
+		isCloudflareAiGateway ||
+		isAntLing;
 
-	const useMaxTokens = baseUrl.includes("chutes.ai") || isMoonshot || isCloudflareAiGateway || isTogether || isNvidia;
+	const useMaxTokens =
+		baseUrl.includes("chutes.ai") || isMoonshot || isCloudflareAiGateway || isTogether || isNvidia || isAntLing;
 
 	const isGrok = provider === "xai" || baseUrl.includes("api.x.ai");
 	const isDeepSeek = provider === "deepseek" || baseUrl.includes("deepseek.com");
@@ -1107,7 +1115,8 @@ function detectCompat(model: Model<"openai-completions">): ResolvedOpenAIComplet
 	return {
 		supportsStore: !isNonStandard,
 		supportsDeveloperRole: isOpenRouterDeveloperRoleModel || (!isNonStandard && !isOpenRouter),
-		supportsReasoningEffort: !isGrok && !isZai && !isMoonshot && !isTogether && !isCloudflareAiGateway && !isNvidia,
+		supportsReasoningEffort:
+			!isGrok && !isZai && !isMoonshot && !isTogether && !isCloudflareAiGateway && !isNvidia && !isAntLing,
 		supportsUsageInStreaming: true,
 		maxTokensField: useMaxTokens ? "max_tokens" : "max_completion_tokens",
 		requiresToolResultName: false,
@@ -1120,16 +1129,24 @@ function detectCompat(model: Model<"openai-completions">): ResolvedOpenAIComplet
 				? "zai"
 				: isTogether
 					? "together"
-					: isOpenRouter
-						? "openrouter"
-						: "openai",
+					: isAntLing
+						? "ant-ling"
+						: isOpenRouter
+							? "openrouter"
+							: "openai",
 		openRouterRouting: {},
 		vercelGatewayRouting: {},
 		zaiToolStream: false,
 		supportsStrictMode: !isMoonshot && !isTogether && !isCloudflareAiGateway && !isNvidia,
 		cacheControlFormat,
 		sendSessionAffinityHeaders: false,
-		supportsLongCacheRetention: !(isTogether || isCloudflareWorkersAI || isCloudflareAiGateway || isNvidia),
+		supportsLongCacheRetention: !(
+			isTogether ||
+			isCloudflareWorkersAI ||
+			isCloudflareAiGateway ||
+			isNvidia ||
+			isAntLing
+		),
 	};
 }
 
diff --git a/packages/ai/src/types.ts b/packages/ai/src/types.ts
index 7a4487720..cbf136a63 100644
--- a/packages/ai/src/types.ts
+++ b/packages/ai/src/types.ts
@@ -22,6 +22,7 @@ export type ImagesApi = KnownImagesApi | (string & {});
 
 export type KnownProvider =
 	| "amazon-bedrock"
+	| "ant-ling"
 	| "anthropic"
 	| "google"
 	| "google-vertex"
@@ -390,7 +391,7 @@ export interface OpenAICompletionsCompat {
 	requiresThinkingAsText?: boolean;
 	/** Whether all replayed assistant messages must include an empty reasoning_content field when reasoning is enabled. Default: auto-detected from URL. */
 	requiresReasoningContentOnAssistantMessages?: boolean;
-	/** Format for reasoning/thinking parameter. "openai" uses reasoning_effort, "openrouter" uses reasoning: { effort }, "deepseek" uses thinking: { type } plus reasoning_effort when supported, "together" uses reasoning: { enabled } plus reasoning_effort when supported, "zai" uses top-level enable_thinking: boolean, "qwen" uses top-level enable_thinking: boolean, "qwen-chat-template" uses chat_template_kwargs.enable_thinking, and "string-thinking" uses top-level thinking: string. Default: "openai". */
+	/** Format for reasoning/thinking parameter. "openai" uses reasoning_effort, "openrouter" uses reasoning: { effort }, "deepseek" uses thinking: { type } plus reasoning_effort when supported, "together" uses reasoning: { enabled } plus reasoning_effort when supported, "zai" uses top-level enable_thinking: boolean, "qwen" uses top-level enable_thinking: boolean, "qwen-chat-template" uses chat_template_kwargs.enable_thinking, "string-thinking" uses top-level thinking: string, and "ant-ling" uses reasoning: { effort } only when the mapped effort is non-null. Default: "openai". */
 	thinkingFormat?:
 		| "openai"
 		| "openrouter"
@@ -399,7 +400,8 @@ export interface OpenAICompletionsCompat {
 		| "zai"
 		| "qwen"
 		| "qwen-chat-template"
-		| "string-thinking";
+		| "string-thinking"
+		| "ant-ling";
 	/** OpenRouter-specific routing preferences. Only used when baseUrl points to OpenRouter. */
 	openRouterRouting?: OpenRouterRouting;
 	/** Vercel AI Gateway routing preferences. Only used when baseUrl points to Vercel AI Gateway. */
diff --git a/packages/ai/test/openai-completions-tool-choice.test.ts b/packages/ai/test/openai-completions-tool-choice.test.ts
index c06ac344e..be6419d61 100644
--- a/packages/ai/test/openai-completions-tool-choice.test.ts
+++ b/packages/ai/test/openai-completions-tool-choice.test.ts
@@ -2,7 +2,7 @@ import { Type } from "typebox";
 import { beforeEach, describe, expect, it, vi } from "vitest";
 import { getModel } from "../src/models.ts";
 import { convertMessages } from "../src/providers/openai-completions.ts";
-import { streamSimple } from "../src/stream.ts";
+import { stream, streamSimple } from "../src/stream.ts";
 import type { AssistantMessage, Model, Tool, ToolResultMessage } from "../src/types.ts";
 
 const mockState = vi.hoisted(() => ({
@@ -1295,4 +1295,95 @@ describe("openai-completions tool_choice", () => {
 		expect(params.reasoning).toEqual({ effort: "high" });
 		expect(params.reasoning_effort).toBeUndefined();
 	});
+
+	it("uses Ant Ling compatibility metadata", async () => {
+		const model = getModel("ant-ling", "Ring-2.6-1T")!;
+		let payload: unknown;
+
+		expect(model.compat).toMatchObject({
+			supportsStore: false,
+			supportsDeveloperRole: false,
+			supportsReasoningEffort: false,
+			maxTokensField: "max_tokens",
+			thinkingFormat: "ant-ling",
+			supportsLongCacheRetention: false,
+		});
+		expect(model.compat?.supportsStrictMode).toBeUndefined();
+		expect(model.compat?.requiresReasoningContentOnAssistantMessages).toBeUndefined();
+
+		await streamSimple(
+			model,
+			{
+				systemPrompt: "Follow instructions.",
+				messages: [{ role: "user", content: "Hi", timestamp: Date.now() }],
+			},
+			{
+				apiKey: "test",
+				maxTokens: 123,
+				reasoning: "high",
+				cacheRetention: "long",
+				sessionId: "ant-ling-session",
+				onPayload: (params: unknown) => {
+					payload = params;
+				},
+			},
+		).result();
+
+		const params = (payload ?? mockState.lastParams) as {
+			max_tokens?: number;
+			max_completion_tokens?: number;
+			messages?: Array<{ role?: string }>;
+			reasoning?: { effort?: string };
+			reasoning_effort?: string;
+			store?: boolean;
+			prompt_cache_key?: string;
+			prompt_cache_retention?: string;
+		};
+		expect(params.max_tokens).toBe(123);
+		expect(params.max_completion_tokens).toBeUndefined();
+		expect(params.messages?.[0]?.role).toBe("system");
+		expect(params.reasoning).toEqual({ effort: "high" });
+		expect(params.reasoning_effort).toBeUndefined();
+		expect(params.store).toBeUndefined();
+		expect(params.prompt_cache_key).toBeUndefined();
+		expect(params.prompt_cache_retention).toBeUndefined();
+	});
+
+	it("omits Ant Ling reasoning for unmapped direct reasoning efforts and non-reasoning models", async () => {
+		const ring = getModel("ant-ling", "Ring-2.6-1T")!;
+		let payload: unknown;
+
+		await stream(
+			ring,
+			{
+				messages: [{ role: "user", content: "Hi", timestamp: Date.now() }],
+			},
+			{
+				apiKey: "test",
+				reasoningEffort: "medium",
+				onPayload: (params: unknown) => {
+					payload = params;
+				},
+			},
+		).result();
+
+		expect((payload ?? mockState.lastParams) as { reasoning?: unknown }).not.toHaveProperty("reasoning");
+
+		const ling = getModel("ant-ling", "Ling-2.6-flash")!;
+		await streamSimple(
+			ling,
+			{
+				messages: [{ role: "user", content: "Hi", timestamp: Date.now() }],
+			},
+			{
+				apiKey: "test",
+				reasoning: "high",
+				onPayload: (params: unknown) => {
+					payload = params;
+				},
+			},
+		).result();
+
+		expect((payload ?? mockState.lastParams) as { reasoning?: unknown }).not.toHaveProperty("reasoning");
+	});
 });
diff --git a/packages/ai/test/stream.test.ts b/packages/ai/test/stream.test.ts
index 1693f4066..9a70b6c28 100644
--- a/packages/ai/test/stream.test.ts
+++ b/packages/ai/test/stream.test.ts
@@ -1169,6 +1169,27 @@ describe("Generate E2E Tests", () => {
 		},
 	);
 
+	describe.skipIf(!process.env.ANT_LING_API_KEY)("Ant Ling Provider (Ling 2.6 Flash via OpenAI Completions)", () => {
+		const llm = getModel("ant-ling", "Ling-2.6-flash");
+
+		it("should complete basic text generation", { retry: 3 }, async () => {
+			await basicTextGeneration(llm);
+		});
+
+		it("should handle tool calling", { retry: 3 }, async () => {
+			await handleToolCall(llm);
+		});
+
+		it("should handle streaming", { retry: 3 }, async () => {
+			await handleStreaming(llm);
+		});
+
+		it("should handle thinking mode", { retry: 3 }, async () => {
+			const ringModel = getModel("ant-ling", "Ring-2.6-1T");
+			await handleThinking(ringModel, { reasoningEffort: "high" });
+		});
+	});
+
 	// =========================================================================
 	// OAuth-based providers (credentials from ~/.pi/agent/oauth.json)
 	// Tokens are resolved at module level (see oauthTokens above)
diff --git a/packages/coding-agent/CHANGELOG.md b/packages/coding-agent/CHANGELOG.md
index 82c509e41..46eebe64b 100644
--- a/packages/coding-agent/CHANGELOG.md
+++ b/packages/coding-agent/CHANGELOG.md
@@ -4,6 +4,7 @@
 
 ### Added
 
+- Added Ant Ling provider selection and setup documentation.
 - Added NVIDIA NIM provider selection, setup documentation, and direct NIM request attribution headers.
 - Added `ctx.mode` to extension contexts so extensions can distinguish TUI, RPC, JSON, and print mode.
 - Added `ctx.getSystemPromptOptions()` for extension commands to inspect the current base system prompt inputs.
diff --git a/packages/coding-agent/README.md b/packages/coding-agent/README.md
index 585a93a0c..8ef336c79 100644
--- a/packages/coding-agent/README.md
+++ b/packages/coding-agent/README.md
@@ -110,6 +110,7 @@ For each built-in provider, pi maintains a list of tool-capable models, updated
 
 **API keys:**
 - Anthropic
+- Ant Ling
 - OpenAI
 - Azure OpenAI
 - DeepSeek
diff --git a/packages/coding-agent/docs/providers.md b/packages/coding-agent/docs/providers.md
index e7a77c5aa..0ecf439df 100644
--- a/packages/coding-agent/docs/providers.md
+++ b/packages/coding-agent/docs/providers.md
@@ -49,6 +49,7 @@ pi
 | Provider | Environment Variable | `auth.json` key |
 |----------|----------------------|------------------|
 | Anthropic | `ANTHROPIC_API_KEY` | `anthropic` |
+| Ant Ling | `ANT_LING_API_KEY` | `ant-ling` |
 | Azure OpenAI Responses | `AZURE_OPENAI_API_KEY` | `azure-openai-responses` |
 | OpenAI | `OPENAI_API_KEY` | `openai` |
 | DeepSeek | `DEEPSEEK_API_KEY` | `deepseek` |
@@ -85,6 +86,7 @@ Store credentials in `~/.pi/agent/auth.json`:
 ```json
 {
   "anthropic": { "type": "api_key", "key": "sk-ant-..." },
+  "ant-ling": { "type": "api_key", "key": "..." },
   "openai": { "type": "api_key", "key": "sk-..." },
   "deepseek": { "type": "api_key", "key": "sk-..." },
   "nvidia": { "type": "api_key", "key": "nvapi-..." },
diff --git a/packages/coding-agent/src/cli/args.ts b/packages/coding-agent/src/cli/args.ts
index 8156bd03f..458bd43ee 100644
--- a/packages/coding-agent/src/cli/args.ts
+++ b/packages/coding-agent/src/cli/args.ts
@@ -328,6 +328,7 @@ ${chalk.bold("Examples:")}
 ${chalk.bold("Environment Variables:")}
   ANTHROPIC_API_KEY                - Anthropic Claude API key
   ANTHROPIC_OAUTH_TOKEN            - Anthropic OAuth token (alternative to API key)
+  ANT_LING_API_KEY                 - Ant Ling API key
   OPENAI_API_KEY                   - OpenAI GPT API key
   AZURE_OPENAI_API_KEY             - Azure OpenAI API key
   AZURE_OPENAI_BASE_URL            - Azure OpenAI/Cognitive Services base URL (e.g. https://{resource}.openai.azure.com)
diff --git a/packages/coding-agent/src/core/model-resolver.ts b/packages/coding-agent/src/core/model-resolver.ts
index e3af076da..a00471c9c 100644
--- a/packages/coding-agent/src/core/model-resolver.ts
+++ b/packages/coding-agent/src/core/model-resolver.ts
@@ -13,6 +13,7 @@ import type { ModelRegistry } from "./model-registry.ts";
 /** Default model IDs for each known provider */
 export const defaultModelPerProvider: Record<KnownProvider, string> = {
 	"amazon-bedrock": "us.anthropic.claude-opus-4-6-v1",
+	"ant-ling": "Ring-2.6-1T",
 	anthropic: "claude-opus-4-8",
 	openai: "gpt-5.4",
 	"azure-openai-responses": "gpt-5.4",
diff --git a/packages/coding-agent/src/core/provider-display-names.ts b/packages/coding-agent/src/core/provider-display-names.ts
index 789f31530..3481e6aef 100644
--- a/packages/coding-agent/src/core/provider-display-names.ts
+++ b/packages/coding-agent/src/core/provider-display-names.ts
@@ -1,6 +1,7 @@
 export const BUILT_IN_PROVIDER_DISPLAY_NAMES: Record<string, string> = {
 	anthropic: "Anthropic",
 	"amazon-bedrock": "Amazon Bedrock",
+	"ant-ling": "Ant Ling",
 	"azure-openai-responses": "Azure OpenAI Responses",
 	cerebras: "Cerebras",
 	"cloudflare-ai-gateway": "Cloudflare AI Gateway",
diff --git a/packages/coding-agent/test/model-resolver.test.ts b/packages/coding-agent/test/model-resolver.test.ts
index 7339200bc..0ac549394 100644
--- a/packages/coding-agent/test/model-resolver.test.ts
+++ b/packages/coding-agent/test/model-resolver.test.ts
@@ -378,11 +378,12 @@ describe("default model selection", () => {
 		expect(defaultModelPerProvider["openai-codex"]).toBe("gpt-5.5");
 	});
 
-	test("zai, minimax, and cerebras defaults track current models", () => {
+	test("zai, minimax, cerebras, and ant-ling defaults track current models", () => {
 		expect(defaultModelPerProvider.zai).toBe("glm-5.1");
 		expect(defaultModelPerProvider.minimax).toBe("MiniMax-M2.7");
 		expect(defaultModelPerProvider["minimax-cn"]).toBe("MiniMax-M2.7");
 		expect(defaultModelPerProvider.cerebras).toBe("zai-glm-4.7");
+		expect(defaultModelPerProvider["ant-ling"]).toBe("Ring-2.6-1T");
 	});
 
 	test("ai-gateway default tracks current model", () => {

From e30b1b18d0bf1db233c1a4c8407361cd5837b7bd Mon Sep 17 00:00:00 2001
From: Armin Ronacher <armin.ronacher@active-4.com>
Date: Tue, 2 Jun 2026 17:03:40 +0200
Subject: [PATCH 09/10] Added security file

---
 SECURITY.md | 70 +++++++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 70 insertions(+)
 create mode 100644 SECURITY.md

diff --git a/SECURITY.md b/SECURITY.md
new file mode 100644
index 000000000..0a249014b
--- /dev/null
+++ b/SECURITY.md
@@ -0,0 +1,70 @@
+# Security Policy
+
+This document should guide you about understanding the security concept behind
+Pi and also where the boundaries are.
+
+In general Pi is a coding agent that runs locally within the security boundary
+of the user that is running it.  It's the responsibiltiy of the user to monitor
+its operations or to contain it within a container, virtual machine or other
+Sandbox solution.
+
+Pi relies on users installing trustworthy extensions and loading trustworthy
+skills and only to use pi within trusted repositories.  This is because files
+like `AGENTS.md` or instructions in comments can be used to prompt inject the
+coding agent trivially and this cannot be protected against.
+
+## Reporting a Vulnerability
+
+If you believe you found a security vulnerability in pi or another package in
+this repository, please report it privately by either:
+
+- Emailing `security@earendil.com`, or
+- Opening a private report through GitHub Security Advisories for this repository
+
+Please include:
+
+- A description of the issue and its impact
+- Steps to reproduce, proof of concept, or relevant logs
+- Affected package, version, commit, or configuration
+- Any known mitigations
+
+Do not open a public issue for security-sensitive reports.  We will review
+reports and coordinate disclosure as appropriate.
+
+## Scope
+
+Security issues in the distributed packages, command-line tools, APIs, and
+repository code are in scope as well as earendil operated infrastricture
+on `pi.dev`.
+
+## Out Of Scope
+
+- Local code execution or sandboxing behavior (the Pi coding agent intentionally does not have a sandbox)
+- Behavior of pi extensions or skills installed by the user
+- Risks from working in untrusted repositories
+- Risks from installing untrusted extensions, skills, packages, or tools
+- Isuses caused by non trustworthy MITM proxies
+- Public internet exposure of a Pi installation
+- Prompt injection attacks
+- Exposed secrets that are third-party/user-controlled credentials
+- Reports requiring write access to trusted local state/config (`~/.pi`, workspace
+  files, `AGENTS.md`, skills/extensions config), unless they show how an attacker
+  gets that write access.
+- Issues caused by intentionally weakened user configuration.
+- Resource/DOS claims that require trusted local input/config against the pi coding agent.
+- Reports about malicious model output.
+- User-approved or user-initiated local actions presented as vulnerabilities.
+
+## Notes for Reporters
+
+The most useful reports show a current, reproducible security boundary bypass
+with demonstrated impact.  Reports that only show expected local-agent behavior,
+prompt injection, or a malicious trusted extension/skill are not security
+vulnerabilities under this model.
+
+When possible, include the exact affected path, package version or commit SHA,
+configuration, and a proof of concept against the latest release or latest
+`main`.  For dependency reports, include evidence that the shipped dependency is
+affected and that the issue is reachable through Pi.  For exposed-secret reports,
+include evidence that the credential is owned by Earendil or grants access to
+Earendil-operated infrastructure or services.

From 0462d44f5695b49ef6e27ab3cd225219085a5b2b Mon Sep 17 00:00:00 2001
From: Armin Ronacher <armin.ronacher@active-4.com>
Date: Tue, 2 Jun 2026 17:24:41 +0200
Subject: [PATCH 10/10] doc: clarify wording in SECURITY.md

---
 SECURITY.md | 23 ++++++++++++++++++++---
 1 file changed, 20 insertions(+), 3 deletions(-)

diff --git a/SECURITY.md b/SECURITY.md
index 0a249014b..ddd75e14d 100644
--- a/SECURITY.md
+++ b/SECURITY.md
@@ -8,6 +8,14 @@ of the user that is running it.  It's the responsibiltiy of the user to monitor
 its operations or to contain it within a container, virtual machine or other
 Sandbox solution.
 
+Pi treats the local user account and files writable by that account as inside
+the same trust boundary as the Pi process itself.  If an attacker can modify files
+under the user's home directory, workspace, shell startup files, environment, or
+Pi configuration, they can generally influence Pi or other local developer tools.
+Reports that depend on such prior local write access are not security
+vulnerabilities unless they demonstrate how Pi grants that write access or crosses
+an operating-system privilege boundary.
+
 Pi relies on users installing trustworthy extensions and loading trustworthy
 skills and only to use pi within trusted repositories.  This is because files
 like `AGENTS.md` or instructions in comments can be used to prompt inject the
@@ -47,9 +55,13 @@ on `pi.dev`.
 - Public internet exposure of a Pi installation
 - Prompt injection attacks
 - Exposed secrets that are third-party/user-controlled credentials
-- Reports requiring write access to trusted local state/config (`~/.pi`, workspace
-  files, `AGENTS.md`, skills/extensions config), unless they show how an attacker
-  gets that write access.
+- Reports requiring the ability to create, modify, delete, or replace files,
+  directories, symlinks, environment variables, shell configuration, or other
+  user-controlled local state on the target machine. This includes `~/.pi`,
+  `~/.pi/agent/models.json`, workspace files, `AGENTS.md`, skills, extensions,
+  extension configuration, dotfiles, and files synchronized through NFS, roaming
+  profiles, or dotfile managers, unless the report shows how Pi itself grants
+  that access.
 - Issues caused by intentionally weakened user configuration.
 - Resource/DOS claims that require trusted local input/config against the pi coding agent.
 - Reports about malicious model output.
@@ -62,6 +74,11 @@ with demonstrated impact.  Reports that only show expected local-agent behavior,
 prompt injection, or a malicious trusted extension/skill are not security
 vulnerabilities under this model.
 
+For example, a report showing that malicious contents written to a trusted Pi
+configuration file cause Pi to execute commands, load attacker-controlled tools,
+send credentials to an attacker-controlled endpoint, or otherwise change behavior
+is out of scope.
+
 When possible, include the exact affected path, package version or commit SHA,
 configuration, and a proof of concept against the latest release or latest
 `main`.  For dependency reports, include evidence that the shipped dependency is