Integrations
gpt2giga is designed for clients that can set a custom base URL for OpenAI-, Anthropic-, or Gemini-compatible SDKs and CLIs.
Base URLs
OpenAI-compatible clients usually use:
http://localhost:8090/v1
For GigaChat v2 features you can explicitly specify:
http://localhost:8090/v2
The backend selection rule is the same for all compatible clients: a URL with
/v1 is forced into the GigaChat v1 contract, a URL with /v2 into the
GigaChat v2 contract. The root URL without a version (http://localhost:8090)
uses GPT2GIGA_GIGACHAT_API_MODE=v1|v2.
Anthropic-compatible clients usually use:
http://localhost:8090
Gemini-compatible clients usually use the root address:
http://localhost:8090
With this base URL, the official Gemini SDKs/CLIs append a Gemini-style path
themselves, for example /v1beta/models/{model}:generateContent.
If GPT2GIGA_ENABLE_API_KEY_AUTH=True, use GPT2GIGA_API_KEY as the client API key. For Gemini clients, the x-goog-api-key header is also supported.
Runnable examples
- OpenAI examples: examples/openai/
- OpenAI Chat Completions: examples/openai/chat_completions/README.md
- OpenAI Responses: examples/openai/responses/README.md
- Anthropic examples: examples/anthropic/README.md
- Gemini examples: examples/gemini/README.md
- OpenAI Agents SDK: examples/openai/agents/weather_handoff.py
Integration guides
| Tool / client | Guide |
|---|---|
| OpenHands | integrations/openhands/README.md |
| OpenAI Codex | integrations/codex/README.md |
| Grok Build | integrations/grok-build/README.md |
| Aider | integrations/aider/README.md |
| Claude Code | integrations/claude-code/README.md |
| Claude Desktop App | integrations/claude-desktop/README.md |
| Gemini CLI | integrations/gemini/README.md |
| Cursor | integrations/cursor/README.md |
| Qwen Code | integrations/qwen-code/README.md |
| Xcode | integrations/xcode/README.md |
| nginx (reverse proxy) | integrations/nginx/README.md |
Verification records
“Verified” below means the repository contains a dated, versioned manual check for the linked guide. It does not promise compatibility with newer client versions. Re-run the guide after a client upgrade and update its record.
| Client | Recorded version | Date | Protocol and paths | Status |
|---|---|---|---|---|
| OpenAI Codex | codex-cli 0.142.1 | 2026-06-26 | OpenAI Chat, /v1 and /v2 | Verified record |
| Claude Code | 2.1.187 | 2026-06-26 | Anthropic Messages, /v1 and /v2 | Verified record |
| Gemini CLI | gemini 0.46.0 | 2026-06-26 | Gemini content API, /v1 and /v2 | Verified record |
| Claude Desktop | 1.12603.1 with Claude Code 2.1.170 | 2026-06-13 | Anthropic Messages through a 3p gateway | Beta record |
The Grok Build, OpenHands, Aider, Cursor, Qwen Code, Xcode, and nginx pages are maintained setup guides, but they do not yet contain a current dated end-to-end verification record. Treat them as reproducible recipes and report the client version and tested route when confirming or reporting a regression.
Compatibility directory
The following clients and frameworks are plausible integration targets because they expose a custom OpenAI-, Anthropic-, or Gemini-compatible base URL. An entry in this directory is not a verification claim.
| Category | Projects |
|---|---|
| Coding agents and editors | Grok Build, OpenCode, KiloCode, OpenHands, Zed, Cline, Codex, Gemini CLI, Aider, Claude Code, Cursor, Qwen Code, Xcode |
| Agent frameworks | Langflow, DeepAgents, CrewAI, Qwen Agent, PydanticAI, CAMEL, smolagents, OpenAI Agents SDK |
| SDKs and desktop clients | Anthropic Python SDK, Claude Desktop |
For a useful verification report, include the client version, operating system, gateway version, configured base URL, GigaChat backend mode, minimal prompt, whether streaming/tools were used, and the redacted error or response shape.