ChatGPT2API

The system includes advanced management for account pools, including:

• Automated Maintenance: Refreshes account email, type, quota, and recovery time; automatically removes invalid tokens; and performs periodic checks on rate-limited accounts.
• Polling: Automatically rotates through available accounts for image generation and editing.
• Import Methods: Supports importing via local CPA JSON files, remote CPA servers, sub2api servers, and access_token imports.
• Proxy Support: Configurable global HTTP/HTTPS/SOCKS5/SOCKS5H proxies and WARP/FlareSolverr runtime.

The Conversation SSE (Server-Sent Events) protocol is a streaming protocol used for upstream dialogue. Clients must consume JSON payloads sequentially to maintain session state, text content, tool call status, and image result pointers.

Payload Types:

• "v1": Protocol version marker.
• [DONE]: End of the SSE stream.
• JSON object: Events, messages, or patches used to update session state.
• JSON string: Short text patches or protocol markers.
• Non-JSON content: Raw content to be preserved as raw events.

Text output is typically delivered via multiple patch events. Follow these rules to reconstruct the text:

• Append: If p == "/message/content/parts/0" and o == "append", append the value v to the current text.
• Replace: If o == "replace", replace the target field with the new value v.
• Batch Patch: If o == "patch" and v is an array, process the patches in the array sequentially.
• Implicit Text: If the payload only contains v as a string, treat it as a text increment for the current stream.

{"p":"/message/content/parts/0","o":"append","v":"Hello"}
{"v":" world"}
{"p":"","o":"patch","v":[
  {"p":"/message/content/parts/0","o":"append","v":"!”},
  {"p":"/message/status","o":"replace","v":"finished_successfully"},
  {"p":"/message/end_turn","o":"replace","v":true}
]}

Use this method if upstream requests are frequently blocked by Cloudflare. This deployment includes warp-proxy, privoxy, flaresolverr, init-config, and the app.

1. Clone the repository and enter the directory.
2. Copy the environment template: cp .env.example .env.
3. Set CHATGPT2API_AUTH_KEY in the .env file.
4. Start the deployment using the specific WARP compose file: docker compose -f docker-compose.warp.yml up -d --build

FlareSolverr configurations can be viewed and tested in the FlareSolverr tab of the admin settings page.

cp .env.example .env
# Edit .env to set CHATGPT2API_AUTH_KEY
docker compose -f docker-compose.warp.yml up -d --build

Follow these steps to upgrade your deployment. Always perform a backup first.

Standard Docker Upgrade:

1. git pull
2. docker compose pull
3. docker compose up -d

WARP / FlareSolverr Upgrade:

1. git pull
2. docker compose -f docker-compose.warp.yml up -d --build

Source Code Upgrade:

1. git pull
2. uv sync
3. (Optional) Rebuild frontend: cd web && bun install && bun run build
4. Restart your backend process manager.

The upstream service typically starts by returning a recovery or session token. Use the conversation_id to identify the session. Note that the token should be kept internal and not exposed to downstream users.

{
  "type": "resume_conversation_token",
  "kind": "topic",
  "token": "...",
  "conversation_id": "..."
}

When a request is refused by policy or moderation, the behavior is as follows:

Policy Refusal:

• Often returns a standard assistant text refusal (e.g., "I can't assist with that request...").
• May include a title_generation event with a title like Request Denied.
• metadata.tool_invoked will be false.

Moderation:

• Returns a moderation event where moderation_response.blocked == true.
• If blocked, prioritize returning the assistant text if available; otherwise, return an appropriate error message.

{
  "type": "moderation",
  "moderation_response": {
    "blocked": true
  },
  "conversation_id": "..."
}

If you encounter Cloudflare interceptions during image generation, use the WARP + Privoxy + FlareSolverr deployment scheme. This setup includes:

• warp-proxy: Provides a WARP SOCKS5 exit.
• privoxy: Converts WARP SOCKS5 to an HTTP proxy.
• flaresolverr: Refreshes Cloudflare clearance.
• init-config: Writes default proxy_runtime configuration.
• app: The main ChatGPT2API service.

By default, only upstream OpenAI/ChatGPT requests use the stable proxy. Account email and CPA auxiliary links are not forced through it. Proxy priority: Account-specific proxy > Stable proxy runtime > Explicit proxy > Old global proxy. You can adjust ports and parameters in .env or via the 'Stable Proxy Runtime' panel in the web settings.

cp .env.example .env
docker compose -f docker-compose.warp.yml up -d --build

Use this method if you do not require WARP or FlareSolverr to bypass Cloudflare protections.

1. Clone the repository: git clone git@github.com:basketikun/chatgpt2api.git cd chatgpt2api
2. Configure your authentication key. You can set auth-key in config.json or via the CHATGPT2API_AUTH_KEY environment variable in docker-compose.yml.
3. Start the service: docker compose up -d

Access the UI at http://localhost:3000 and the API at http://localhost:3000/v1.

git clone git@github.com:basketikun/chatgpt2api.git
cd chatgpt2api
# Configure CHATGPT2API_AUTH_KEY in docker-compose.yml
docker compose up -d

Suitable for local development or temporary debugging.

Backend Setup: Requires uv for dependency management.

git clone git@github.com:basketikun/chatgpt2api.git
cd chatgpt2api
uv sync
uv run main.py

Frontend Development: Requires bun.

cd web
bun install
bun run dev

When running from source, the backend reads config.json and the data/ directory from the project root.

# Backend
uv sync
uv run main.py

# Frontend
cd web
bun install
bun run dev

To avoid misidentifying input attachments as output images, only treat an asset_pointer as a successful image output if all the following conditions are met:

1. message.author.role == "tool" (The source must be a tool message).
2. metadata.async_task_type == "image_gen" (The task must be image generation).
3. The asset_pointer is either file-service://... or sediment://....

{
  "v": {
    "message": {
      "author": {"role": "tool"},
      "content": {
        "content_type": "multimodal_text",
        "parts": [
          {"asset_pointer": "file-service://file_result"},
          {"asset_pointer": "sediment://file_result"}
        ]
      },
      "metadata": {"async_task_type": "image_gen"}
    }
  },
  "conversation_id": "..."
}