ai-zero-token 2.0.3 → 2.0.5

package/dist/server/index.js

@@ -41,7 +41,8 @@ async function startServer(params) {
     const app = createApp({
       corsOrigin: resolveCorsOrigin(),
       bodyLimit,
-      onRestart: params?.onRestart
+      onRestart: params?.onRestart,
+      onRestartCodex: params?.onRestartCodex
     });
     try {
       await app.listen({

package/docs/API_USAGE.md

@@ -41,6 +41,41 @@ Image model: gpt-image-2
 
 Use `GET /v1/models` to see the models available through the current local Codex cache.
 
+## OpenClaw Settings
+
+Use the OpenAI-compatible provider mode in OpenClaw:
+
+```text
+Provider: OpenAI compatible
+Base URL: http://127.0.0.1:8787/v1
+API Key: local
+Model: gpt-5.4
+Chat endpoint: /chat/completions
+Streaming: enabled
+Tools / function calling: enabled
+```
+
+The gateway accepts OpenClaw-style `chat.completions` requests with `tools`, `tool_choice`, `parallel_tool_calls`, `reasoning_effort`, assistant `tool_calls`, and tool-role result messages. It translates those fields to the upstream Codex Responses shape and returns OpenAI-style chat responses.
+
+OpenClaw requests are visible in the management console request log when the client sends an OpenClaw user agent. The log keeps safe summaries only; it does not store full access tokens.
+
+## Codex Custom Provider
+
+Codex CLI/Desktop can route model traffic through AI Zero Token by using a custom Responses provider in `~/.codex/config.toml`. The management console Settings page can write this automatically with "接管 Codex 请求" and remove it with "解除接管":
+
+```toml
+model = "gpt-5.4"
+model_provider = "ai-zero-token"
+
+[model_providers.ai-zero-token]
+name = "AI Zero Token"
+base_url = "http://127.0.0.1:8787/codex/v1"
+wire_api = "responses"
+supports_websockets = false
+```
+
+Codex sends `POST /codex/v1/responses` with `Accept: text/event-stream`; the gateway forwards that request to the active Codex OAuth account and streams upstream Responses SSE events back to Codex. The regular `/v1/*` routes remain OpenAI-compatible API routes for non-Codex clients.
+
 ## Models
 
 ```bash
@@ -74,6 +109,50 @@ curl http://127.0.0.1:8787/v1/chat/completions \
     "messages": [
       { "role": "user", "content": "Reply with OK only." }
     ]
+}'
+```
+
+Streaming chat completions:
+
+```bash
+curl http://127.0.0.1:8787/v1/chat/completions \
+  -H "Content-Type: application/json" \
+  -d '{
+    "model": "gpt-5.4",
+    "stream": true,
+    "messages": [
+      { "role": "user", "content": "Reply with OK only." }
+    ]
+  }'
+```
+
+Tool-call compatible request:
+
+```bash
+curl http://127.0.0.1:8787/v1/chat/completions \
+  -H "Content-Type: application/json" \
+  -d '{
+    "model": "gpt-5.4",
+    "messages": [
+      { "role": "user", "content": "What is the weather tool argument for Shanghai?" }
+    ],
+    "tools": [
+      {
+        "type": "function",
+        "function": {
+          "name": "get_weather",
+          "description": "Get weather for a city.",
+          "parameters": {
+            "type": "object",
+            "properties": {
+              "city": { "type": "string" }
+            },
+            "required": ["city"]
+          }
+        }
+      }
+    ],
+    "tool_choice": "auto"
   }'
 ```
 
@@ -134,5 +213,7 @@ console.log(response.choices[0]?.message?.content);
 
 - Login first through the management page or `azt login`.
 - A model appearing in `/v1/models` means the local Codex cache lists it. Final availability still depends on the active account.
-- `stream=true` is not supported yet.
+- `stream=true` is supported for `/v1/chat/completions` through OpenAI-style SSE chunks. Codex passthrough streaming is isolated under `/codex/v1/responses`.
+- `n > 1` is not supported for `/v1/chat/completions`.
+- Tool/function calling is supported for common OpenAI-compatible clients, including OpenClaw, but exact upstream behavior still depends on the active Codex model and account.
 - The default listener is `0.0.0.0:8787`, so local-network clients can call the gateway by using the machine IP.

package/docs/DESKTOP_RELEASE.md

@@ -2,6 +2,27 @@
 
 This project ships the desktop app with Electron. The desktop main process starts the existing local Fastify gateway and loads the React management UI served by that gateway.
 
+## 2.0.5 Release Notes
+
+Version `2.0.5` adds Codex custom provider routing and finer account rotation controls:
+
+- Settings-page Codex provider setup for writing or removing the AI Zero Token managed `~/.codex/config.toml` provider.
+- Local and remote Codex gateway URL modes, including automatic normalization to `/codex/v1`.
+- Dedicated `POST /codex/v1/responses` passthrough route for Codex CLI/Desktop Responses SSE traffic.
+- Provider status reporting in the management console, including active provider, base URL, and config path.
+- Auto-switch exclusion list for accounts that should not participate in automatic quota rotation.
+- Safer settings persistence through normalized loads, deduplicated profile IDs, queued saves, and atomic writes.
+
+## 2.0.4 Release Notes
+
+Version `2.0.4` adds the macOS menu-bar account panel and OpenClaw compatibility work:
+
+- Menu-bar quick account panel for switching gateway/Codex accounts.
+- Menu actions for quota refresh, Base URL copy, console open, gateway restart, and quit.
+- Desktop Codex restart hook after applying an account to local Codex.
+- OpenClaw-compatible `chat.completions` streaming and tool-call fields.
+- Real gateway request log entries for recent API traffic.
+
 ## 2.0.0 Release Notes
 
 Version `2.0.0` is the first desktop-focused major release. It includes:
@@ -93,6 +114,8 @@ AI Zero Token Setup {version}.exe
 AI Zero Token-{version}-win.zip
 ```
 
+For `2.0.5`, replace `{version}` with `2.0.5`.
+
 Artifact purpose:
 
 - `AI Zero Token-{version}-mac-arm64.dmg`: macOS Apple Silicon builds for M1/M2/M3/M4 devices.
@@ -131,6 +154,7 @@ App icon files live in:
 build/icon.png
 build/icon.icns
 build/icon.ico
+build/tray-icon-template.png
 ```
 
 They are included in Electron packaging and npm packing.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "ai-zero-token",
-  "version": "2.0.3",
+  "version": "2.0.5",
   "description": "Local-first OpenAI-compatible AI CLI and gateway with Codex OAuth, multi-account management, and gpt-image-2 image generation/editing.",
   "license": "MIT",
   "author": "AI Zero Token Contributors",
@@ -90,6 +90,7 @@
       "dist/**/*",
       "admin-ui/dist/**/*",
       "build/icon.*",
+      "build/tray-icon-template.png",
       "package.json",
       "README.md",
       "README.zh-CN.md",
@@ -153,6 +154,7 @@
     "build/icon.png",
     "build/icon.icns",
     "build/icon.ico",
+    "build/tray-icon-template.png",
     "build/mac-install-guide.txt",
     "CHANGELOG.md",
     "docs/API_USAGE.md",