Integration
Plugins
By CLI-to-API gateway and plugins for OpenClaw
Install
npm install #Configuration Example
{
"port": 4090,
"backends": {
"claude-code": {
"enabled": true,
"command": "claude",
"defaultModel": "sonnet",
"tools": false,
"sessionContinuity": true
},
"codex": {
"enabled": true,
"command": "codex",
"defaultModel": "o4-mini",
"tools": false,
"sessionContinuity": true
},
"gemini": {
"enabled": true,
"command": "gemini",
"defaultModel": "auto",
"tools": false,
"sessionContinuity": true
}
}
}README
# openclaw-plugins
> [!CAUTION]
> ## USE AT YOUR OWN RISK
>
> This software is provided **"AS IS"**, without warranty of any kind, express or implied, including but not limited to the warranties of merchantability, fitness for a particular purpose, and noninfringement.
>
> **This tool spawns CLI subprocesses with the ability to execute arbitrary code on your machine.** If you enable `tools: true`, it passes `--dangerously-skip-permissions` to the CLI, which means **it can read, write, and delete files, run shell commands, and generally do whatever it wants to your system without asking.**
>
> If this software **nukes your system**, **eats your homework**, **maxes out your API bill**, **deletes your production database**, **emails your browser history to your boss**, or **achieves sentience and orders 10,000 rubber ducks to your house** — that is entirely on you. Do not come crying to us. You were warned.
>
> **NO WARRANTY. NO LIABILITY. NO REFUNDS. NO SYMPATHY.**
>
> By using this software you acknowledge that you have read this disclaimer, understood the risks, and have accepted that the consequences of running autonomous AI agents with filesystem access are **your problem and yours alone.**
---
Plugins and tools for [OpenClaw](https://github.com/nicobailon/openclaw).
## cli-gateway
OpenAI-compatible proxy that translates `/v1/chat/completions` requests into CLI subprocess calls. Use your Claude Code, Codex, or Gemini CLI subscription as an LLM endpoint for OpenClaw or any OpenAI-compatible client.
### How It Works
```
OpenClaw / any OpenAI client
|
v
cli-gateway (localhost:4090)
/v1/chat/completions
|
+---------+---------+
| | |
v v v
Claude Codex Gemini
Code CLI CLI
(PTY) (spawn) (spawn)
| | |
v v v
NDJSON -> OpenAI SSE translation
```
The gateway spawns CLI processes, parses their NDJSON output, and translates it to OpenAI-compatible SSE streams. Session state is owned by the CLI tool — only the latest user message is sent on each turn, with `--resume` for continuity.
### Supported Backends
| Backend | Command | Models | Spawn Method |
|---------|---------|--------|-------------|
| **Claude Code** | `claude` | `claude-code/sonnet`, `claude-code/opus`, `claude-code/haiku` | PTY (node-pty) |
| **Codex CLI** | `codex` | `codex/gpt-5.3-codex`, `codex/o3`, `codex/o4-mini` | child_process |
| **Gemini CLI** | `gemini` | `gemini/auto`, `gemini/gemini-2.5-pro`, `gemini/gemini-2.5-flash` | child_process |
Backends are auto-detected at startup. If a CLI tool isn't installed, that backend is silently skipped.
### Quick Start
One line, fresh Ubuntu 22.04+ server:
```bash
curl -fsSL https://raw.githubusercontent.com/jlportman3/openclaw-plugins/main/install.sh | sudo bash
```
This installs all system dependencies, lets you pick which CLI tools to install (Claude Code, Codex, Gemini), walks you through authentication, and deploys cli-gateway as a systemd service.
### Manual Start
```bash
cd cli-gateway
npm install # first time only (installs node-pty)
node --experimental-strip-types src/server.ts
```
### API
- `GET /health` — Backend availability and model counts
- `GET /v1/models` — List all available models across backends
- `POST /v1/chat/completions` — Chat completion (streaming & non-streaming)
### Model Names
Format: `backend/model` — the backend prefix routes to the right CLI tool.
```
claude-code/sonnet -> claude -p --model sonnet
codex/gpt-5.3-codex -> codex exec --json -m gpt-5.3-codex
gemini/auto -> gemini -p --model auto
```
### Configuration
Edit `cli-gateway/config.json`:
```json
{
"port": 4090,
"backends": {
"claude-code": {
"enabled": true,
"command": "claude",
"defaultModel": "sonnet",
"tools": false,
"sessionContinuity": true
},
"codex": {
"enabled": true,
"command": "codex",
"defaultModel": "o4-mini",
"tools": false,
"sessionContinuity": true
},
"gemini": {
"enabled": true,
"command": "gemini",
"defaultModel": "auto",
"tools": false,
"sessionContinuity": true
}
}
}
```
Set `tools: true` to enable tool use. This passes permission-bypass flags to the CLI (`--dangerously-skip-permissions` for Claude, `--yolo` for Codex/Gemini) — **use with extreme caution**.
### Environment Security
The gateway scrubs the environment before spawning subprocesses. Only whitelisted system variables (`PATH`, `HOME`, `TERM`, etc.) and backend-specific auth keys are passed through. No `.env` secrets or stray API keys leak into child processes.
### Session Continuity
Sessions are managed by the CLI tool. The proxy derives a session ID from the conversation's system prompt and first user message (SHA-256 hash). Pass `X-Session-Id` header for explicit session control.
On the first request, a new session is started. Subsequent requests use `--resume` to continue the existing session, sending only the latest user message to conserve tokens.
### Dependencies
- **Node.js 22+** — native TypeScript execution via `--experimental-strip-types`
- **node-pty** — PTY support for Claude Code (installed via `npm install`)
- At least one CLI tool:
- **Claude Code** — `npm install -g @anthropic-ai/claude-code`
- **Codex CLI** — `npm install -g @openai/codex`
- **Gemini CLI** — `npm install -g @google/gemini-cli`
### Adding New Backends
Create a new file in `cli-gateway/src/backends/` implementing the `CliBackend` interface, register it in `backends/index.ts`, and add auth keys to `util/clean-env.ts`. See `codex.ts` for the simplest reference.
### Design
See [docs/plans/2026-02-17-cli-gateway-design.md](docs/plans/2026-02-17-cli-gateway-design.md).
integration
Comments
Sign in to leave a comment