Remote compact for API-key users (#7835)

codex-rs/core/src/codex.rs

@@ -2175,7 +2175,10 @@ pub(crate) async fn run_task(
 
                 // as long as compaction works well in getting us way below the token limit, we shouldn't worry about being in an infinite loop.
                 if token_limit_reached {
-                    if should_use_remote_compact_task(&sess) {
+                    if should_use_remote_compact_task(
+                        sess.as_ref(),
+                        &turn_context.client.get_provider(),
+                    ) {
                         run_inline_remote_auto_compact_task(sess.clone(), turn_context.clone())
                             .await;
                     } else {

codex-rs/core/src/compact.rs

@@ -1,5 +1,6 @@
 use std::sync::Arc;
 
+use crate::ModelProviderInfo;
 use crate::Prompt;
 use crate::client_common::ResponseEvent;
 use crate::codex::Session;
@@ -18,7 +19,6 @@ use crate::truncate::TruncationPolicy;
 use crate::truncate::approx_token_count;
 use crate::truncate::truncate_text;
 use crate::util::backoff;
-use codex_app_server_protocol::AuthMode;
 use codex_protocol::items::TurnItem;
 use codex_protocol::models::ContentItem;
 use codex_protocol::models::ResponseInputItem;
@@ -32,13 +32,11 @@ pub const SUMMARIZATION_PROMPT: &str = include_str!("../templates/compact/prompt
 pub const SUMMARY_PREFIX: &str = include_str!("../templates/compact/summary_prefix.md");
 const COMPACT_USER_MESSAGE_MAX_TOKENS: usize = 20_000;
 
-pub(crate) fn should_use_remote_compact_task(session: &Session) -> bool {
-    session
-        .services
-        .auth_manager
-        .auth()
-        .is_some_and(|auth| auth.mode == AuthMode::ChatGPT)
-        && session.enabled(Feature::RemoteCompaction)
+pub(crate) fn should_use_remote_compact_task(
+    session: &Session,
+    provider: &ModelProviderInfo,
+) -> bool {
+    provider.is_openai() && session.enabled(Feature::RemoteCompaction)
 }
 
 pub(crate) async fn run_inline_auto_compact_task(

codex-rs/core/src/context_manager/history.rs

@@ -92,7 +92,7 @@ impl ContextManager {
                     encrypted_content: Some(content),
                     ..
                 }
-                | ResponseItem::CompactionSummary {
+                | ResponseItem::Compaction {
                     encrypted_content: content,
                 } => estimate_reasoning_length(content.len()) as i64,
                 item => {
@@ -258,7 +258,7 @@ impl ContextManager {
             | ResponseItem::FunctionCall { .. }
             | ResponseItem::WebSearchCall { .. }
             | ResponseItem::CustomToolCall { .. }
-            | ResponseItem::CompactionSummary { .. }
+            | ResponseItem::Compaction { .. }
             | ResponseItem::GhostSnapshot { .. }
             | ResponseItem::Other => item.clone(),
         }
@@ -277,7 +277,7 @@ fn is_api_message(message: &ResponseItem) -> bool {
         | ResponseItem::LocalShellCall { .. }
         | ResponseItem::Reasoning { .. }
         | ResponseItem::WebSearchCall { .. }
-        | ResponseItem::CompactionSummary { .. } => true,
+        | ResponseItem::Compaction { .. } => true,
         ResponseItem::GhostSnapshot { .. } => false,
         ResponseItem::Other => false,
     }

codex-rs/core/src/model_provider_info.rs

@@ -28,6 +28,8 @@ const MAX_STREAM_MAX_RETRIES: u64 = 100;
 const MAX_REQUEST_MAX_RETRIES: u64 = 100;
 pub const CHAT_WIRE_API_DEPRECATION_SUMMARY: &str = r#"Support for the "chat" wire API is deprecated and will soon be removed. Update your model provider definition in config.toml to use wire_api = "responses"."#;
 
+const OPENAI_PROVIDER_NAME: &str = "OpenAI";
+
 /// Wire protocol that the provider speaks. Most third-party services only
 /// implement the classic OpenAI Chat Completions JSON schema, whereas OpenAI
 /// itself (and a handful of others) additionally expose the more modern
@@ -210,7 +212,7 @@ impl ModelProviderInfo {
     }
     pub fn create_openai_provider() -> ModelProviderInfo {
         ModelProviderInfo {
-            name: "OpenAI".into(),
+            name: OPENAI_PROVIDER_NAME.into(),
             // Allow users to override the default OpenAI endpoint by
             // exporting `OPENAI_BASE_URL`. This is useful when pointing
             // Codex at a proxy, mock server, or Azure-style deployment
@@ -247,6 +249,10 @@ impl ModelProviderInfo {
             requires_openai_auth: true,
         }
     }
+
+    pub fn is_openai(&self) -> bool {
+        self.name == OPENAI_PROVIDER_NAME
+    }
 }
 
 pub const DEFAULT_LMSTUDIO_PORT: u16 = 1234;

codex-rs/core/src/rollout/policy.rs

@@ -28,7 +28,7 @@ pub(crate) fn should_persist_response_item(item: &ResponseItem) -> bool {
         | ResponseItem::CustomToolCallOutput { .. }
         | ResponseItem::WebSearchCall { .. }
         | ResponseItem::GhostSnapshot { .. }
-        | ResponseItem::CompactionSummary { .. } => true,
+        | ResponseItem::Compaction { .. } => true,
         ResponseItem::Other => false,
     }
 }

codex-rs/core/src/tasks/compact.rs

@@ -25,7 +25,10 @@ impl SessionTask for CompactTask {
         _cancellation_token: CancellationToken,
     ) -> Option<String> {
         let session = session.clone_session();
-        if crate::compact::should_use_remote_compact_task(&session) {
+        if crate::compact::should_use_remote_compact_task(
+            session.as_ref(),
+            &ctx.client.get_provider(),
+        ) {
             crate::compact_remote::run_remote_compact_task(session, ctx).await
         } else {
             crate::compact::run_compact_task(session, ctx, input).await