Add openai codex support

This commit is contained in:
Luis Pater
2025-08-08 15:44:00 +08:00
parent d2394b0be9
commit de7b8501cc
57 changed files with 7488 additions and 2478 deletions

View File

@@ -0,0 +1,114 @@
// Package code provides request translation functionality for Claude API.
// It handles parsing and transforming Claude API requests into the internal client format,
// extracting model information, system instructions, message contents, and tool declarations.
// The package also performs JSON data cleaning and transformation to ensure compatibility
// between Claude API format and the internal client's expected format.
package code
import (
"fmt"
"github.com/luispater/CLIProxyAPI/internal/misc"
"github.com/tidwall/gjson"
"github.com/tidwall/sjson"
)
// PrepareClaudeRequest parses and transforms a Claude API request into internal client format.
// It extracts the model name, system instruction, message contents, and tool declarations
// from the raw JSON request and returns them in the format expected by the internal client.
func ConvertClaudeCodeRequestToCodex(rawJSON []byte) string {
template := `{"model":"","instructions":"","input":[]}`
instructions := misc.CodexInstructions
template, _ = sjson.SetRaw(template, "instructions", instructions)
rootResult := gjson.ParseBytes(rawJSON)
modelResult := rootResult.Get("model")
template, _ = sjson.Set(template, "model", modelResult.String())
systemsResult := rootResult.Get("system")
if systemsResult.IsArray() {
systemResults := systemsResult.Array()
message := `{"type":"message","role":"user","content":[]}`
for i := 0; i < len(systemResults); i++ {
systemResult := systemResults[i]
systemTypeResult := systemResult.Get("type")
if systemTypeResult.String() == "text" {
message, _ = sjson.Set(message, fmt.Sprintf("content.%d.type", i), "input_text")
message, _ = sjson.Set(message, fmt.Sprintf("content.%d.text", i), systemResult.Get("text").String())
}
}
template, _ = sjson.SetRaw(template, "input.-1", message)
}
messagesResult := rootResult.Get("messages")
if messagesResult.IsArray() {
messageResults := messagesResult.Array()
for i := 0; i < len(messageResults); i++ {
messageResult := messageResults[i]
messageContentsResult := messageResult.Get("content")
if messageContentsResult.IsArray() {
messageContentResults := messageContentsResult.Array()
for j := 0; j < len(messageContentResults); j++ {
messageContentResult := messageContentResults[j]
messageContentTypeResult := messageContentResult.Get("type")
if messageContentTypeResult.String() == "text" {
message := `{"type": "message","role":"","content":[]}`
messageRole := messageResult.Get("role").String()
message, _ = sjson.Set(message, "role", messageRole)
partType := "input_text"
if messageRole == "assistant" {
partType = "output_text"
}
currentIndex := len(gjson.Get(message, "content").Array())
message, _ = sjson.Set(message, fmt.Sprintf("content.%d.type", currentIndex), partType)
message, _ = sjson.Set(message, fmt.Sprintf("content.%d.text", currentIndex), messageContentResult.Get("text").String())
template, _ = sjson.SetRaw(template, "input.-1", message)
} else if messageContentTypeResult.String() == "tool_use" {
functionCallMessage := `{"type":"function_call"}`
functionCallMessage, _ = sjson.Set(functionCallMessage, "call_id", messageContentResult.Get("id").String())
functionCallMessage, _ = sjson.Set(functionCallMessage, "name", messageContentResult.Get("name").String())
functionCallMessage, _ = sjson.Set(functionCallMessage, "arguments", messageContentResult.Get("input").Raw)
template, _ = sjson.SetRaw(template, "input.-1", functionCallMessage)
} else if messageContentTypeResult.String() == "tool_result" {
functionCallOutputMessage := `{"type":"function_call_output"}`
functionCallOutputMessage, _ = sjson.Set(functionCallOutputMessage, "call_id", messageContentResult.Get("tool_use_id").String())
functionCallOutputMessage, _ = sjson.Set(functionCallOutputMessage, "output", messageContentResult.Get("content").String())
template, _ = sjson.SetRaw(template, "input.-1", functionCallOutputMessage)
}
}
}
}
}
toolsResult := rootResult.Get("tools")
if toolsResult.IsArray() {
template, _ = sjson.SetRaw(template, "tools", `[]`)
template, _ = sjson.Set(template, "tool_choice", `auto`)
toolResults := toolsResult.Array()
for i := 0; i < len(toolResults); i++ {
toolResult := toolResults[i]
tool := toolResult.Raw
tool, _ = sjson.Set(tool, "type", "function")
tool, _ = sjson.SetRaw(tool, "parameters", toolResult.Get("input_schema").Raw)
tool, _ = sjson.Delete(tool, "input_schema")
tool, _ = sjson.Delete(tool, "parameters.$schema")
tool, _ = sjson.Set(tool, "strict", false)
template, _ = sjson.SetRaw(template, "tools.-1", tool)
}
}
template, _ = sjson.Set(template, "parallel_tool_calls", true)
template, _ = sjson.Set(template, "reasoning.effort", "low")
template, _ = sjson.Set(template, "reasoning.summary", "auto")
template, _ = sjson.Set(template, "stream", true)
template, _ = sjson.Set(template, "store", false)
template, _ = sjson.Set(template, "include", []string{"reasoning.encrypted_content"})
return template
}

View File

@@ -0,0 +1,129 @@
// Package code provides response translation functionality for Claude API.
// This package handles the conversion of backend client responses into Claude-compatible
// Server-Sent Events (SSE) format, implementing a sophisticated state machine that manages
// different response types including text content, thinking processes, and function calls.
// The translation ensures proper sequencing of SSE events and maintains state across
// multiple response chunks to provide a seamless streaming experience.
package code
import (
"fmt"
"github.com/tidwall/gjson"
"github.com/tidwall/sjson"
)
// ConvertCliToClaude performs sophisticated streaming response format conversion.
// This function implements a complex state machine that translates backend client responses
// into Claude-compatible Server-Sent Events (SSE) format. It manages different response types
// and handles state transitions between content blocks, thinking processes, and function calls.
//
// Response type states: 0=none, 1=content, 2=thinking, 3=function
// The function maintains state across multiple calls to ensure proper SSE event sequencing.
func ConvertCodexResponseToClaude(rawJSON []byte, hasToolCall bool) (string, bool) {
// log.Debugf("rawJSON: %s", string(rawJSON))
output := ""
rootResult := gjson.ParseBytes(rawJSON)
typeResult := rootResult.Get("type")
typeStr := typeResult.String()
template := ""
if typeStr == "response.created" {
template = `{"type":"message_start","message":{"id":"","type":"message","role":"assistant","model":"claude-opus-4-1-20250805","stop_sequence":null,"usage":{"input_tokens":0,"output_tokens":0},"content":[],"stop_reason":null}}`
template, _ = sjson.Set(template, "message.model", rootResult.Get("response.model").String())
template, _ = sjson.Set(template, "message.id", rootResult.Get("response.id").String())
output = "event: message_start\n"
output += fmt.Sprintf("data: %s\n", template)
} else if typeStr == "response.reasoning_summary_part.added" {
template = `{"type":"content_block_start","index":0,"content_block":{"type":"thinking","thinking":""}}`
template, _ = sjson.Set(template, "index", rootResult.Get("output_index").Int())
output = "event: content_block_start\n"
output += fmt.Sprintf("data: %s\n", template)
} else if typeStr == "response.reasoning_summary_text.delta" {
template = `{"type":"content_block_delta","index":0,"delta":{"type":"thinking_delta","thinking":""}}`
template, _ = sjson.Set(template, "index", rootResult.Get("output_index").Int())
template, _ = sjson.Set(template, "delta.thinking", rootResult.Get("delta").String())
output = "event: content_block_delta\n"
output += fmt.Sprintf("data: %s\n", template)
} else if typeStr == "response.reasoning_summary_part.done" {
template = `{"type":"content_block_stop","index":0}`
template, _ = sjson.Set(template, "index", rootResult.Get("output_index").Int())
output = "event: content_block_stop\n"
output += fmt.Sprintf("data: %s\n", template)
} else if typeStr == "response.content_part.added" {
template = `{"type":"content_block_start","index":0,"content_block":{"type":"text","text":""}}`
template, _ = sjson.Set(template, "index", rootResult.Get("output_index").Int())
output = "event: content_block_start\n"
output += fmt.Sprintf("data: %s\n", template)
} else if typeStr == "response.output_text.delta" {
template = `{"type":"content_block_delta","index":0,"delta":{"type":"text_delta","text":""}}`
template, _ = sjson.Set(template, "index", rootResult.Get("output_index").Int())
template, _ = sjson.Set(template, "delta.text", rootResult.Get("delta").String())
output = "event: content_block_delta\n"
output += fmt.Sprintf("data: %s\n", template)
} else if typeStr == "response.content_part.done" {
template = `{"type":"content_block_stop","index":0}`
template, _ = sjson.Set(template, "index", rootResult.Get("output_index").Int())
output = "event: content_block_stop\n"
output += fmt.Sprintf("data: %s\n", template)
} else if typeStr == "response.completed" {
template = `{"type":"message_delta","delta":{"stop_reason":"tool_use","stop_sequence":null},"usage":{"input_tokens":0,"output_tokens":0}}`
if hasToolCall {
template, _ = sjson.Set(template, "delta.stop_reason", "tool_use")
} else {
template, _ = sjson.Set(template, "delta.stop_reason", "end_turn")
}
template, _ = sjson.Set(template, "usage.input_tokens", rootResult.Get("response.usage.input_tokens").Int())
template, _ = sjson.Set(template, "usage.output_tokens", rootResult.Get("response.usage.output_tokens").Int())
output = "event: message_delta\n"
output += fmt.Sprintf("data: %s\n\n", template)
output += "event: message_stop\n"
output += `data: {"type":"message_stop"}`
output += "\n\n"
} else if typeStr == "response.output_item.added" {
itemResult := rootResult.Get("item")
itemType := itemResult.Get("type").String()
if itemType == "function_call" {
hasToolCall = true
template = `{"type":"content_block_start","index":0,"content_block":{"type":"tool_use","id":"","name":"","input":{}}}`
template, _ = sjson.Set(template, "index", rootResult.Get("output_index").Int())
template, _ = sjson.Set(template, "content_block.id", itemResult.Get("call_id").String())
template, _ = sjson.Set(template, "content_block.name", itemResult.Get("name").String())
output = "event: content_block_start\n"
output += fmt.Sprintf("data: %s\n\n", template)
template = `{"type":"content_block_delta","index":0,"delta":{"type":"input_json_delta","partial_json":""}}`
template, _ = sjson.Set(template, "index", rootResult.Get("output_index").Int())
output += "event: content_block_delta\n"
output += fmt.Sprintf("data: %s\n", template)
}
} else if typeStr == "response.output_item.done" {
itemResult := rootResult.Get("item")
itemType := itemResult.Get("type").String()
if itemType == "function_call" {
template = `{"type":"content_block_stop","index":0}`
template, _ = sjson.Set(template, "index", rootResult.Get("output_index").Int())
output = "event: content_block_stop\n"
output += fmt.Sprintf("data: %s\n", template)
}
} else if typeStr == "response.function_call_arguments.delta" {
template = `{"type":"content_block_delta","index":0,"delta":{"type":"input_json_delta","partial_json":""}}`
template, _ = sjson.Set(template, "index", rootResult.Get("output_index").Int())
template, _ = sjson.Set(template, "delta.partial_json", rootResult.Get("delta").String())
output += "event: content_block_delta\n"
output += fmt.Sprintf("data: %s\n", template)
}
return output, hasToolCall
}

View File

@@ -0,0 +1,199 @@
// Package code provides request translation functionality for Claude API.
// It handles parsing and transforming Claude API requests into the internal client format,
// extracting model information, system instructions, message contents, and tool declarations.
// The package also performs JSON data cleaning and transformation to ensure compatibility
// between Claude API format and the internal client's expected format.
package code
import (
"crypto/rand"
"math/big"
"strings"
"github.com/luispater/CLIProxyAPI/internal/misc"
"github.com/tidwall/gjson"
"github.com/tidwall/sjson"
)
// PrepareClaudeRequest parses and transforms a Claude API request into internal client format.
// It extracts the model name, system instruction, message contents, and tool declarations
// from the raw JSON request and returns them in the format expected by the internal client.
func ConvertGeminiRequestToCodex(rawJSON []byte) string {
// Base template
out := `{"model":"","instructions":"","input":[]}`
// Inject standard Codex instructions
instructions := misc.CodexInstructions
out, _ = sjson.SetRaw(out, "instructions", instructions)
root := gjson.ParseBytes(rawJSON)
// helper for generating paired call IDs in the form: call_<alphanum>
// Gemini uses sequential pairing across possibly multiple in-flight
// functionCalls, so we keep a FIFO queue of generated call IDs and
// consume them in order when functionResponses arrive.
var pendingCallIDs []string
// genCallID creates a random call id like: call_<8chars>
genCallID := func() string {
const letters = "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789"
var b strings.Builder
// 8 chars random suffix
for i := 0; i < 24; i++ {
n, _ := rand.Int(rand.Reader, big.NewInt(int64(len(letters))))
b.WriteByte(letters[n.Int64()])
}
return "call_" + b.String()
}
// Model
if v := root.Get("model"); v.Exists() {
out, _ = sjson.Set(out, "model", v.Value())
}
// System instruction -> as a user message with input_text parts
sysParts := root.Get("system_instruction.parts")
if sysParts.IsArray() {
msg := `{"type":"message","role":"user","content":[]}`
arr := sysParts.Array()
for i := 0; i < len(arr); i++ {
p := arr[i]
if t := p.Get("text"); t.Exists() {
part := `{}`
part, _ = sjson.Set(part, "type", "input_text")
part, _ = sjson.Set(part, "text", t.String())
msg, _ = sjson.SetRaw(msg, "content.-1", part)
}
}
if len(gjson.Get(msg, "content").Array()) > 0 {
out, _ = sjson.SetRaw(out, "input.-1", msg)
}
}
// Contents -> messages and function calls/results
contents := root.Get("contents")
if contents.IsArray() {
items := contents.Array()
for i := 0; i < len(items); i++ {
item := items[i]
role := item.Get("role").String()
if role == "model" {
role = "assistant"
}
parts := item.Get("parts")
if !parts.IsArray() {
continue
}
parr := parts.Array()
for j := 0; j < len(parr); j++ {
p := parr[j]
// text part
if t := p.Get("text"); t.Exists() {
msg := `{"type":"message","role":"","content":[]}`
msg, _ = sjson.Set(msg, "role", role)
partType := "input_text"
if role == "assistant" {
partType = "output_text"
}
part := `{}`
part, _ = sjson.Set(part, "type", partType)
part, _ = sjson.Set(part, "text", t.String())
msg, _ = sjson.SetRaw(msg, "content.-1", part)
out, _ = sjson.SetRaw(out, "input.-1", msg)
continue
}
// function call from model
if fc := p.Get("functionCall"); fc.Exists() {
fn := `{"type":"function_call"}`
if name := fc.Get("name"); name.Exists() {
fn, _ = sjson.Set(fn, "name", name.String())
}
if args := fc.Get("args"); args.Exists() {
fn, _ = sjson.Set(fn, "arguments", args.Raw)
}
// generate a paired random call_id and enqueue it so the
// corresponding functionResponse can pop the earliest id
// to preserve ordering when multiple calls are present.
id := genCallID()
fn, _ = sjson.Set(fn, "call_id", id)
pendingCallIDs = append(pendingCallIDs, id)
out, _ = sjson.SetRaw(out, "input.-1", fn)
continue
}
// function response from user
if fr := p.Get("functionResponse"); fr.Exists() {
fno := `{"type":"function_call_output"}`
// Prefer a string result if present; otherwise embed the raw response as a string
if res := fr.Get("response.result"); res.Exists() {
fno, _ = sjson.Set(fno, "output", res.String())
} else if resp := fr.Get("response"); resp.Exists() {
fno, _ = sjson.Set(fno, "output", resp.Raw)
}
// fno, _ = sjson.Set(fno, "call_id", "call_W6nRJzFXyPM2LFBbfo98qAbq")
// attach the oldest queued call_id to pair the response
// with its call. If the queue is empty, generate a new id.
var id string
if len(pendingCallIDs) > 0 {
id = pendingCallIDs[0]
// pop the first element
pendingCallIDs = pendingCallIDs[1:]
} else {
id = genCallID()
}
fno, _ = sjson.Set(fno, "call_id", id)
out, _ = sjson.SetRaw(out, "input.-1", fno)
continue
}
}
}
}
// Tools mapping: Gemini functionDeclarations -> Codex tools
tools := root.Get("tools")
if tools.IsArray() {
out, _ = sjson.SetRaw(out, "tools", `[]`)
out, _ = sjson.Set(out, "tool_choice", "auto")
tarr := tools.Array()
for i := 0; i < len(tarr); i++ {
td := tarr[i]
fns := td.Get("functionDeclarations")
if !fns.IsArray() {
continue
}
farr := fns.Array()
for j := 0; j < len(farr); j++ {
fn := farr[j]
tool := `{}`
tool, _ = sjson.Set(tool, "type", "function")
if v := fn.Get("name"); v.Exists() {
tool, _ = sjson.Set(tool, "name", v.String())
}
if v := fn.Get("description"); v.Exists() {
tool, _ = sjson.Set(tool, "description", v.String())
}
if prm := fn.Get("parameters"); prm.Exists() {
// Remove optional $schema field if present
cleaned := prm.Raw
cleaned, _ = sjson.Delete(cleaned, "$schema")
cleaned, _ = sjson.Set(cleaned, "additionalProperties", false)
tool, _ = sjson.SetRaw(tool, "parameters", cleaned)
}
tool, _ = sjson.Set(tool, "strict", false)
out, _ = sjson.SetRaw(out, "tools.-1", tool)
}
}
}
// Fixed flags aligning with Codex expectations
out, _ = sjson.Set(out, "parallel_tool_calls", true)
out, _ = sjson.Set(out, "reasoning.effort", "low")
out, _ = sjson.Set(out, "reasoning.summary", "auto")
out, _ = sjson.Set(out, "stream", true)
out, _ = sjson.Set(out, "store", false)
out, _ = sjson.Set(out, "include", []string{"reasoning.encrypted_content"})
return out
}

View File

@@ -0,0 +1,251 @@
// Package code provides response translation functionality for Gemini API.
// This package handles the conversion of Codex backend responses into Gemini-compatible
// JSON format, transforming streaming events into single-line JSON responses that include
// thinking content, regular text content, and function calls in the format expected by
// Gemini API clients.
package code
import (
"encoding/json"
"time"
"github.com/tidwall/gjson"
"github.com/tidwall/sjson"
)
type ConvertCodexResponseToGeminiParams struct {
Model string
CreatedAt int64
ResponseID string
LastStorageOutput string
}
// ConvertCodexResponseToGemini converts Codex streaming response format to Gemini single-line JSON format.
// This function processes various Codex event types and transforms them into Gemini-compatible JSON responses.
// It handles thinking content, regular text content, and function calls, outputting single-line JSON
// that matches the Gemini API response format.
// The lastEventType parameter tracks the previous event type to handle consecutive function calls properly.
func ConvertCodexResponseToGemini(rawJSON []byte, param *ConvertCodexResponseToGeminiParams) []string {
rootResult := gjson.ParseBytes(rawJSON)
typeResult := rootResult.Get("type")
typeStr := typeResult.String()
// Base Gemini response template
template := `{"candidates":[{"content":{"role":"model","parts":[]}}],"usageMetadata":{"trafficType":"PROVISIONED_THROUGHPUT"},"modelVersion":"gemini-2.5-pro","createTime":"2025-08-15T02:52:03.884209Z","responseId":"06CeaPH7NaCU48APvNXDyA4"}`
if param.LastStorageOutput != "" && typeStr == "response.output_item.done" {
template = param.LastStorageOutput
} else {
template, _ = sjson.Set(template, "modelVersion", param.Model)
createdAtResult := rootResult.Get("response.created_at")
if createdAtResult.Exists() {
param.CreatedAt = createdAtResult.Int()
template, _ = sjson.Set(template, "createTime", time.Unix(param.CreatedAt, 0).Format(time.RFC3339Nano))
}
template, _ = sjson.Set(template, "responseId", param.ResponseID)
}
// Handle function call completion
if typeStr == "response.output_item.done" {
itemResult := rootResult.Get("item")
itemType := itemResult.Get("type").String()
if itemType == "function_call" {
// Create function call part
functionCall := `{"functionCall":{"name":"","args":{}}}`
functionCall, _ = sjson.Set(functionCall, "functionCall.name", itemResult.Get("name").String())
// Parse and set arguments
argsStr := itemResult.Get("arguments").String()
if argsStr != "" {
argsResult := gjson.Parse(argsStr)
if argsResult.IsObject() {
functionCall, _ = sjson.SetRaw(functionCall, "functionCall.args", argsStr)
}
}
template, _ = sjson.SetRaw(template, "candidates.0.content.parts.-1", functionCall)
template, _ = sjson.Set(template, "candidates.0.finishReason", "STOP")
param.LastStorageOutput = template
// Use this return to storage message
return []string{}
}
}
if typeStr == "response.created" { // Handle response creation - set model and response ID
template, _ = sjson.Set(template, "modelVersion", rootResult.Get("response.model").String())
template, _ = sjson.Set(template, "responseId", rootResult.Get("response.id").String())
param.ResponseID = rootResult.Get("response.id").String()
} else if typeStr == "response.reasoning_summary_text.delta" { // Handle reasoning/thinking content delta
part := `{"thought":true,"text":""}`
part, _ = sjson.Set(part, "text", rootResult.Get("delta").String())
template, _ = sjson.SetRaw(template, "candidates.0.content.parts.-1", part)
} else if typeStr == "response.output_text.delta" { // Handle regular text content delta
part := `{"text":""}`
part, _ = sjson.Set(part, "text", rootResult.Get("delta").String())
template, _ = sjson.SetRaw(template, "candidates.0.content.parts.-1", part)
} else if typeStr == "response.completed" { // Handle response completion with usage metadata
template, _ = sjson.Set(template, "usageMetadata.promptTokenCount", rootResult.Get("response.usage.input_tokens").Int())
template, _ = sjson.Set(template, "usageMetadata.candidatesTokenCount", rootResult.Get("response.usage.output_tokens").Int())
totalTokens := rootResult.Get("response.usage.input_tokens").Int() + rootResult.Get("response.usage.output_tokens").Int()
template, _ = sjson.Set(template, "usageMetadata.totalTokenCount", totalTokens)
} else {
return []string{}
}
if param.LastStorageOutput != "" {
return []string{param.LastStorageOutput, template}
} else {
return []string{template}
}
}
// ConvertCodexResponseToGeminiNonStream converts a completed Codex response to Gemini non-streaming format.
// This function processes the final response.completed event and transforms it into a complete
// Gemini-compatible JSON response that includes all content parts, function calls, and usage metadata.
func ConvertCodexResponseToGeminiNonStream(rawJSON []byte, model string) string {
rootResult := gjson.ParseBytes(rawJSON)
// Verify this is a response.completed event
if rootResult.Get("type").String() != "response.completed" {
return ""
}
// Base Gemini response template for non-streaming
template := `{"candidates":[{"content":{"role":"model","parts":[]},"finishReason":"STOP"}],"usageMetadata":{"trafficType":"PROVISIONED_THROUGHPUT"},"modelVersion":"","createTime":"","responseId":""}`
// Set model version
template, _ = sjson.Set(template, "modelVersion", model)
// Set response metadata from the completed response
responseData := rootResult.Get("response")
if responseData.Exists() {
// Set response ID
if responseId := responseData.Get("id"); responseId.Exists() {
template, _ = sjson.Set(template, "responseId", responseId.String())
}
// Set creation time
if createdAt := responseData.Get("created_at"); createdAt.Exists() {
template, _ = sjson.Set(template, "createTime", time.Unix(createdAt.Int(), 0).Format(time.RFC3339Nano))
}
// Set usage metadata
if usage := responseData.Get("usage"); usage.Exists() {
inputTokens := usage.Get("input_tokens").Int()
outputTokens := usage.Get("output_tokens").Int()
totalTokens := inputTokens + outputTokens
template, _ = sjson.Set(template, "usageMetadata.promptTokenCount", inputTokens)
template, _ = sjson.Set(template, "usageMetadata.candidatesTokenCount", outputTokens)
template, _ = sjson.Set(template, "usageMetadata.totalTokenCount", totalTokens)
}
// Process output content to build parts array
var parts []interface{}
hasToolCall := false
var pendingFunctionCalls []interface{}
flushPendingFunctionCalls := func() {
if len(pendingFunctionCalls) > 0 {
// Add all pending function calls as individual parts
// This maintains the original Gemini API format while ensuring consecutive calls are grouped together
for _, fc := range pendingFunctionCalls {
parts = append(parts, fc)
}
pendingFunctionCalls = nil
}
}
if output := responseData.Get("output"); output.Exists() && output.IsArray() {
output.ForEach(func(key, value gjson.Result) bool {
itemType := value.Get("type").String()
switch itemType {
case "reasoning":
// Flush any pending function calls before adding non-function content
flushPendingFunctionCalls()
// Add thinking content
if content := value.Get("content"); content.Exists() {
part := map[string]interface{}{
"thought": true,
"text": content.String(),
}
parts = append(parts, part)
}
case "message":
// Flush any pending function calls before adding non-function content
flushPendingFunctionCalls()
// Add regular text content
if content := value.Get("content"); content.Exists() && content.IsArray() {
content.ForEach(func(_, contentItem gjson.Result) bool {
if contentItem.Get("type").String() == "output_text" {
if text := contentItem.Get("text"); text.Exists() {
part := map[string]interface{}{
"text": text.String(),
}
parts = append(parts, part)
}
}
return true
})
}
case "function_call":
// Collect function call for potential merging with consecutive ones
hasToolCall = true
functionCall := map[string]interface{}{
"functionCall": map[string]interface{}{
"name": value.Get("name").String(),
"args": map[string]interface{}{},
},
}
// Parse and set arguments
if argsStr := value.Get("arguments").String(); argsStr != "" {
argsResult := gjson.Parse(argsStr)
if argsResult.IsObject() {
var args map[string]interface{}
if err := json.Unmarshal([]byte(argsStr), &args); err == nil {
functionCall["functionCall"].(map[string]interface{})["args"] = args
}
}
}
pendingFunctionCalls = append(pendingFunctionCalls, functionCall)
}
return true
})
// Handle any remaining pending function calls at the end
flushPendingFunctionCalls()
}
// Set the parts array
if len(parts) > 0 {
template, _ = sjson.SetRaw(template, "candidates.0.content.parts", mustMarshalJSON(parts))
}
// Set finish reason based on whether there were tool calls
if hasToolCall {
template, _ = sjson.Set(template, "candidates.0.finishReason", "STOP")
} else {
template, _ = sjson.Set(template, "candidates.0.finishReason", "STOP")
}
}
return template
}
// mustMarshalJSON marshals data to JSON, panicking on error (should not happen with valid data)
func mustMarshalJSON(v interface{}) string {
data, err := json.Marshal(v)
if err != nil {
panic(err)
}
return string(data)
}

View File

@@ -0,0 +1,227 @@
// Package codex provides utilities to translate OpenAI Chat Completions
// request JSON into OpenAI Responses API request JSON using gjson/sjson.
// It supports tools, multimodal text/image inputs, and Structured Outputs.
package openai
import (
"github.com/luispater/CLIProxyAPI/internal/misc"
"github.com/tidwall/gjson"
"github.com/tidwall/sjson"
)
// ConvertOpenAIChatRequestToCodex converts an OpenAI Chat Completions request JSON
// into an OpenAI Responses API request JSON. The transformation follows the
// examples defined in docs/2.md exactly, including tools, multi-turn dialog,
// multimodal text/image handling, and Structured Outputs mapping.
func ConvertOpenAIChatRequestToCodex(rawJSON []byte) string {
// Start with empty JSON object
out := `{}`
store := false
// Stream must be set to true
if v := gjson.GetBytes(rawJSON, "stream"); v.Exists() {
out, _ = sjson.Set(out, "stream", true)
}
// Codex not support temperature, top_p, top_k, max_output_tokens, so comment them
// if v := gjson.GetBytes(rawJSON, "temperature"); v.Exists() {
// out, _ = sjson.Set(out, "temperature", v.Value())
// }
// if v := gjson.GetBytes(rawJSON, "top_p"); v.Exists() {
// out, _ = sjson.Set(out, "top_p", v.Value())
// }
// if v := gjson.GetBytes(rawJSON, "top_k"); v.Exists() {
// out, _ = sjson.Set(out, "top_k", v.Value())
// }
// Map token limits
// if v := gjson.GetBytes(rawJSON, "max_tokens"); v.Exists() {
// out, _ = sjson.Set(out, "max_output_tokens", v.Value())
// }
// if v := gjson.GetBytes(rawJSON, "max_completion_tokens"); v.Exists() {
// out, _ = sjson.Set(out, "max_output_tokens", v.Value())
// }
// Map reasoning effort
if v := gjson.GetBytes(rawJSON, "reasoning_effort"); v.Exists() {
out, _ = sjson.Set(out, "reasoning.effort", v.Value())
out, _ = sjson.Set(out, "reasoning.summary", "auto")
}
// Model
if v := gjson.GetBytes(rawJSON, "model"); v.Exists() {
out, _ = sjson.Set(out, "model", v.Value())
}
// Extract system instructions from first system message (string or text object)
messages := gjson.GetBytes(rawJSON, "messages")
instructions := misc.CodexInstructions
out, _ = sjson.SetRaw(out, "instructions", instructions)
// if messages.IsArray() {
// arr := messages.Array()
// for i := 0; i < len(arr); i++ {
// m := arr[i]
// if m.Get("role").String() == "system" {
// c := m.Get("content")
// if c.Type == gjson.String {
// out, _ = sjson.Set(out, "instructions", c.String())
// } else if c.IsObject() && c.Get("type").String() == "text" {
// out, _ = sjson.Set(out, "instructions", c.Get("text").String())
// }
// break
// }
// }
// }
// Build input from messages, skipping system/tool roles
out, _ = sjson.SetRaw(out, "input", `[]`)
if messages.IsArray() {
arr := messages.Array()
for i := 0; i < len(arr); i++ {
m := arr[i]
role := m.Get("role").String()
if role == "tool" || role == "function" {
continue
}
// Prepare message object
msg := `{}`
if role == "system" {
msg, _ = sjson.Set(msg, "role", "user")
} else {
msg, _ = sjson.Set(msg, "role", role)
}
msg, _ = sjson.SetRaw(msg, "content", `[]`)
c := m.Get("content")
if c.Type == gjson.String {
// Single string content
partType := "input_text"
if role == "assistant" {
partType = "output_text"
}
part := `{}`
part, _ = sjson.Set(part, "type", partType)
part, _ = sjson.Set(part, "text", c.String())
msg, _ = sjson.SetRaw(msg, "content.-1", part)
} else if c.IsArray() {
items := c.Array()
for j := 0; j < len(items); j++ {
it := items[j]
t := it.Get("type").String()
switch t {
case "text":
partType := "input_text"
if role == "assistant" {
partType = "output_text"
}
part := `{}`
part, _ = sjson.Set(part, "type", partType)
part, _ = sjson.Set(part, "text", it.Get("text").String())
msg, _ = sjson.SetRaw(msg, "content.-1", part)
case "image_url":
// Map image inputs to input_image for Responses API
if role == "user" {
part := `{}`
part, _ = sjson.Set(part, "type", "input_image")
if u := it.Get("image_url.url"); u.Exists() {
part, _ = sjson.Set(part, "image_url", u.String())
}
msg, _ = sjson.SetRaw(msg, "content.-1", part)
}
case "file":
// Files are not specified in examples; skip for now
}
}
}
out, _ = sjson.SetRaw(out, "input.-1", msg)
}
}
// Map response_format and text settings to Responses API text.format
rf := gjson.GetBytes(rawJSON, "response_format")
text := gjson.GetBytes(rawJSON, "text")
if rf.Exists() {
// Always create text object when response_format provided
if !gjson.Get(out, "text").Exists() {
out, _ = sjson.SetRaw(out, "text", `{}`)
}
rft := rf.Get("type").String()
switch rft {
case "text":
out, _ = sjson.Set(out, "text.format.type", "text")
case "json_schema":
js := rf.Get("json_schema")
if js.Exists() {
out, _ = sjson.Set(out, "text.format.type", "json_schema")
if v := js.Get("name"); v.Exists() {
out, _ = sjson.Set(out, "text.format.name", v.Value())
}
if v := js.Get("strict"); v.Exists() {
out, _ = sjson.Set(out, "text.format.strict", v.Value())
}
if v := js.Get("schema"); v.Exists() {
out, _ = sjson.SetRaw(out, "text.format.schema", v.Raw)
}
}
}
// Map verbosity if provided
if text.Exists() {
if v := text.Get("verbosity"); v.Exists() {
out, _ = sjson.Set(out, "text.verbosity", v.Value())
}
}
// The examples include store: true when response_format is provided
store = true
} else if text.Exists() {
// If only text.verbosity present (no response_format), map verbosity
if v := text.Get("verbosity"); v.Exists() {
if !gjson.Get(out, "text").Exists() {
out, _ = sjson.SetRaw(out, "text", `{}`)
}
out, _ = sjson.Set(out, "text.verbosity", v.Value())
}
}
// Map tools (flatten function fields)
tools := gjson.GetBytes(rawJSON, "tools")
if tools.IsArray() {
out, _ = sjson.SetRaw(out, "tools", `[]`)
arr := tools.Array()
for i := 0; i < len(arr); i++ {
t := arr[i]
if t.Get("type").String() == "function" {
item := `{}`
item, _ = sjson.Set(item, "type", "function")
fn := t.Get("function")
if fn.Exists() {
if v := fn.Get("name"); v.Exists() {
item, _ = sjson.Set(item, "name", v.Value())
}
if v := fn.Get("description"); v.Exists() {
item, _ = sjson.Set(item, "description", v.Value())
}
if v := fn.Get("parameters"); v.Exists() {
item, _ = sjson.SetRaw(item, "parameters", v.Raw)
}
if v := fn.Get("strict"); v.Exists() {
item, _ = sjson.Set(item, "strict", v.Value())
}
}
out, _ = sjson.SetRaw(out, "tools.-1", item)
}
}
// The examples include store: true when tools and formatting are used; be conservative
if rf.Exists() {
store = true
}
}
out, _ = sjson.Set(out, "store", store)
return out
}

View File

@@ -0,0 +1,231 @@
// Package codex provides response translation functionality for converting between
// Codex API response formats and OpenAI-compatible formats. It handles both
// streaming and non-streaming responses, transforming backend client responses
// into OpenAI Server-Sent Events (SSE) format and standard JSON response formats.
// The package supports content translation, function calls, reasoning content,
// usage metadata, and various response attributes while maintaining compatibility
// with OpenAI API specifications.
package openai
import (
"github.com/tidwall/gjson"
"github.com/tidwall/sjson"
)
type ConvertCliToOpenAIParams struct {
ResponseID string
CreatedAt int64
Model string
}
// ConvertCodexResponseToOpenAIChat translates a single chunk of a streaming response from the
// Codex backend client format to the OpenAI Server-Sent Events (SSE) format.
// It returns an empty string if the chunk contains no useful data.
func ConvertCodexResponseToOpenAIChat(rawJSON []byte, params *ConvertCliToOpenAIParams) (*ConvertCliToOpenAIParams, string) {
// Initialize the OpenAI SSE template.
template := `{"id":"","object":"chat.completion.chunk","created":12345,"model":"model","choices":[{"index":0,"delta":{"role":null,"content":null,"reasoning_content":null,"tool_calls":null},"finish_reason":null,"native_finish_reason":null}]}`
rootResult := gjson.ParseBytes(rawJSON)
typeResult := rootResult.Get("type")
dataType := typeResult.String()
if dataType == "response.created" {
return &ConvertCliToOpenAIParams{
ResponseID: rootResult.Get("response.id").String(),
CreatedAt: rootResult.Get("response.created_at").Int(),
Model: rootResult.Get("response.model").String(),
}, ""
}
if params == nil {
return params, ""
}
// Extract and set the model version.
if modelResult := gjson.GetBytes(rawJSON, "model"); modelResult.Exists() {
template, _ = sjson.Set(template, "model", modelResult.String())
}
template, _ = sjson.Set(template, "created", params.CreatedAt)
// Extract and set the response ID.
template, _ = sjson.Set(template, "id", params.ResponseID)
// Extract and set usage metadata (token counts).
if usageResult := gjson.GetBytes(rawJSON, "response.usage"); usageResult.Exists() {
if outputTokensResult := usageResult.Get("output_tokens"); outputTokensResult.Exists() {
template, _ = sjson.Set(template, "usage.completion_tokens", outputTokensResult.Int())
}
if totalTokensResult := usageResult.Get("total_tokens"); totalTokensResult.Exists() {
template, _ = sjson.Set(template, "usage.total_tokens", totalTokensResult.Int())
}
if inputTokensResult := usageResult.Get("input_tokens"); inputTokensResult.Exists() {
template, _ = sjson.Set(template, "usage.prompt_tokens", inputTokensResult.Int())
}
if reasoningTokensResult := usageResult.Get("output_tokens_details.reasoning_tokens"); reasoningTokensResult.Exists() {
template, _ = sjson.Set(template, "usage.completion_tokens_details.reasoning_tokens", reasoningTokensResult.Int())
}
}
if dataType == "response.reasoning_summary_text.delta" {
if deltaResult := rootResult.Get("delta"); deltaResult.Exists() {
template, _ = sjson.Set(template, "choices.0.delta.role", "assistant")
template, _ = sjson.Set(template, "choices.0.delta.reasoning_content", deltaResult.String())
}
} else if dataType == "response.reasoning_summary_text.done" {
template, _ = sjson.Set(template, "choices.0.delta.role", "assistant")
template, _ = sjson.Set(template, "choices.0.delta.reasoning_content", "\n\n")
} else if dataType == "response.output_text.delta" {
if deltaResult := rootResult.Get("delta"); deltaResult.Exists() {
template, _ = sjson.Set(template, "choices.0.delta.role", "assistant")
template, _ = sjson.Set(template, "choices.0.delta.content", deltaResult.String())
}
} else if dataType == "response.completed" {
template, _ = sjson.Set(template, "choices.0.finish_reason", "stop")
template, _ = sjson.Set(template, "choices.0.native_finish_reason", "stop")
} else if dataType == "response.output_item.done" {
functionCallItemTemplate := `{"id": "","type": "function","function": {"name": "","arguments": ""}}`
itemResult := rootResult.Get("item")
if itemResult.Exists() {
if itemResult.Get("type").String() != "function_call" {
return params, ""
}
template, _ = sjson.SetRaw(template, "choices.0.delta.tool_calls", `[]`)
functionCallItemTemplate, _ = sjson.Set(functionCallItemTemplate, "id", itemResult.Get("call_id").String())
functionCallItemTemplate, _ = sjson.Set(functionCallItemTemplate, "function.name", itemResult.Get("name").String())
functionCallItemTemplate, _ = sjson.Set(functionCallItemTemplate, "function.arguments", itemResult.Get("arguments").String())
template, _ = sjson.Set(template, "choices.0.delta.role", "assistant")
template, _ = sjson.SetRaw(template, "choices.0.delta.tool_calls.-1", functionCallItemTemplate)
}
} else {
return params, ""
}
return params, template
}
// ConvertCodexResponseToOpenAIChatNonStream aggregates response from the Codex backend client
// convert a single, non-streaming OpenAI-compatible JSON response.
func ConvertCodexResponseToOpenAIChatNonStream(rawJSON string, unixTimestamp int64) string {
template := `{"id":"","object":"chat.completion","created":123456,"model":"model","choices":[{"index":0,"message":{"role":"assistant","content":null,"reasoning_content":null,"tool_calls":null},"finish_reason":null,"native_finish_reason":null}]}`
// Extract and set the model version.
if modelResult := gjson.Get(rawJSON, "model"); modelResult.Exists() {
template, _ = sjson.Set(template, "model", modelResult.String())
}
// Extract and set the creation timestamp.
if createdAtResult := gjson.Get(rawJSON, "created_at"); createdAtResult.Exists() {
template, _ = sjson.Set(template, "created", createdAtResult.Int())
} else {
template, _ = sjson.Set(template, "created", unixTimestamp)
}
// Extract and set the response ID.
if idResult := gjson.Get(rawJSON, "id"); idResult.Exists() {
template, _ = sjson.Set(template, "id", idResult.String())
}
// Extract and set usage metadata (token counts).
if usageResult := gjson.Get(rawJSON, "usage"); usageResult.Exists() {
if outputTokensResult := usageResult.Get("output_tokens"); outputTokensResult.Exists() {
template, _ = sjson.Set(template, "usage.completion_tokens", outputTokensResult.Int())
}
if totalTokensResult := usageResult.Get("total_tokens"); totalTokensResult.Exists() {
template, _ = sjson.Set(template, "usage.total_tokens", totalTokensResult.Int())
}
if inputTokensResult := usageResult.Get("input_tokens"); inputTokensResult.Exists() {
template, _ = sjson.Set(template, "usage.prompt_tokens", inputTokensResult.Int())
}
if reasoningTokensResult := usageResult.Get("output_tokens_details.reasoning_tokens"); reasoningTokensResult.Exists() {
template, _ = sjson.Set(template, "usage.completion_tokens_details.reasoning_tokens", reasoningTokensResult.Int())
}
}
// Process the output array for content and function calls
outputResult := gjson.Get(rawJSON, "output")
if outputResult.IsArray() {
outputArray := outputResult.Array()
var contentText string
var reasoningText string
var toolCalls []string
for _, outputItem := range outputArray {
outputType := outputItem.Get("type").String()
switch outputType {
case "reasoning":
// Extract reasoning content from summary
if summaryResult := outputItem.Get("summary"); summaryResult.IsArray() {
summaryArray := summaryResult.Array()
for _, summaryItem := range summaryArray {
if summaryItem.Get("type").String() == "summary_text" {
reasoningText = summaryItem.Get("text").String()
break
}
}
}
case "message":
// Extract message content
if contentResult := outputItem.Get("content"); contentResult.IsArray() {
contentArray := contentResult.Array()
for _, contentItem := range contentArray {
if contentItem.Get("type").String() == "output_text" {
contentText = contentItem.Get("text").String()
break
}
}
}
case "function_call":
// Handle function call content
functionCallTemplate := `{"id": "","type": "function","function": {"name": "","arguments": ""}}`
if callIdResult := outputItem.Get("call_id"); callIdResult.Exists() {
functionCallTemplate, _ = sjson.Set(functionCallTemplate, "id", callIdResult.String())
}
if nameResult := outputItem.Get("name"); nameResult.Exists() {
functionCallTemplate, _ = sjson.Set(functionCallTemplate, "function.name", nameResult.String())
}
if argsResult := outputItem.Get("arguments"); argsResult.Exists() {
functionCallTemplate, _ = sjson.Set(functionCallTemplate, "function.arguments", argsResult.String())
}
toolCalls = append(toolCalls, functionCallTemplate)
}
}
// Set content and reasoning content if found
if contentText != "" {
template, _ = sjson.Set(template, "choices.0.message.content", contentText)
template, _ = sjson.Set(template, "choices.0.message.role", "assistant")
}
if reasoningText != "" {
template, _ = sjson.Set(template, "choices.0.message.reasoning_content", reasoningText)
template, _ = sjson.Set(template, "choices.0.message.role", "assistant")
}
// Add tool calls if any
if len(toolCalls) > 0 {
template, _ = sjson.SetRaw(template, "choices.0.message.tool_calls", `[]`)
for _, toolCall := range toolCalls {
template, _ = sjson.SetRaw(template, "choices.0.message.tool_calls.-1", toolCall)
}
template, _ = sjson.Set(template, "choices.0.message.role", "assistant")
}
}
// Extract and set the finish reason based on status
if statusResult := gjson.Get(rawJSON, "status"); statusResult.Exists() {
status := statusResult.String()
if status == "completed" {
template, _ = sjson.Set(template, "choices.0.finish_reason", "stop")
template, _ = sjson.Set(template, "choices.0.native_finish_reason", "stop")
}
}
return template
}