Integration
Substation
By OpenClaw plugin — route LLM requests through Claude Max and ChatGPT Pro subscriptions at zero API cost
Install
npm install @anthropic-ai/claude-agent-sdkConfiguration Example
{
"models": {
"providers": {
"indigo": {
"baseUrl": "http://127.0.0.1:8403/v1",
"apiKey": "substation-local",
"api": "openai-completions",
"models": [
{ "id": "opus-4-6", "name": "Claude Opus 4.6 (SubStation)", "contextWindow": 200000, "maxTokens": 128000 },
{ "id": "sonnet-4-6", "name": "Claude Sonnet 4.6 (SubStation)", "contextWindow": 200000, "maxTokens": 64000 },
{ "id": "haiku-4-5", "name": "Claude Haiku 4.5 (SubStation)", "contextWindow": 200000, "maxTokens": 64000 },
{ "id": "gpt-5.4", "name": "GPT 5.4 (SubStation)", "contextWindow": 200000, "maxTokens": 128000 },
{ "id": "gpt-5.4-mini", "name": "GPT 5.4 Mini (SubStation)", "contextWindow": 200000, "maxTokens": 64000 }
]
}
}
}
}README
# SubStation
OpenClaw plugin that routes LLM requests through your existing subscriptions — Claude Max and ChatGPT Pro/Plus — at zero API cost.
## What it does
SubStation sits between OpenClaw and the LLM providers, proxying requests through your subscription accounts instead of paid API keys:
- **Claude** (Opus, Sonnet, Haiku) — Routes through Claude Code's Agent SDK using your Claude Max OAuth tokens. Requests come from a real Claude Code process, indistinguishable from normal usage.
- **ChatGPT** (GPT-5.4, GPT-5.1 Codex) — Routes through the Codex API using your ChatGPT Pro/Plus OAuth tokens.
## Features
- **Multi-account credential pool** — Add multiple OAuth tokens per provider, SubStation rotates between them (LRU) to distribute usage and handle rate limits
- **Persistent sessions** — Claude sessions stay warm after first request, eliminating cold start overhead
- **Real-time streaming** — Token-by-token streaming from both providers piped directly to OpenClaw
- **Auto-rotation on rate limits** — 429s trigger automatic failover to the next token in the pool
- **Token refresh** — ChatGPT OAuth tokens auto-refresh when expired
- **Hardened** — Global error handlers, log rotation, port conflict auto-resolution, request size limits
## Setup
### 1. Install the plugin
```bash
# Copy plugin to OpenClaw extensions
mkdir -p ~/.openclaw/extensions/substation/dist
cp src/index.js ~/.openclaw/extensions/substation/dist/index.js
cp openclaw.plugin.json ~/.openclaw/extensions/substation/
cp package.json ~/.openclaw/extensions/substation/
# Install Agent SDK dependency
cd ~/.openclaw/extensions/substation
npm install @anthropic-ai/claude-agent-sdk
```
### 2. Add Claude Max tokens
SubStation reads Anthropic OAuth tokens from OpenClaw's auth-profiles:
```
~/.openclaw/agents/main/agent/auth-profiles.json
```
Add one or more `anthropic:*` profiles with `"type": "token"` and your OAuth token. SubStation auto-detects and rotates between all available tokens.
### 3. Add ChatGPT tokens (optional)
Run the included auth script:
```bash
node scripts/substation-auth chatgpt
```
This opens your browser for the OAuth PKCE flow. Tokens are saved to `~/.substation/token-pool.json`. Run multiple times for multiple accounts.
Check token status:
```bash
node scripts/substation-auth status
```
### 4. Configure OpenClaw
Add SubStation as a provider in `~/.openclaw/openclaw.json`:
```json
{
"models": {
"providers": {
"indigo": {
"baseUrl": "http://127.0.0.1:8403/v1",
"apiKey": "substation-local",
"api": "openai-completions",
"models": [
{ "id": "opus-4-6", "name": "Claude Opus 4.6 (SubStation)", "contextWindow": 200000, "maxTokens": 128000 },
{ "id": "sonnet-4-6", "name": "Claude Sonnet 4.6 (SubStation)", "contextWindow": 200000, "maxTokens": 64000 },
{ "id": "haiku-4-5", "name": "Claude Haiku 4.5 (SubStation)", "contextWindow": 200000, "maxTokens": 64000 },
{ "id": "gpt-5.4", "name": "GPT 5.4 (SubStation)", "contextWindow": 200000, "maxTokens": 128000 },
{ "id": "gpt-5.4-mini", "name": "GPT 5.4 Mini (SubStation)", "contextWindow": 200000, "maxTokens": 64000 }
]
}
}
}
}
```
Add models to the allowlist under `agents.defaults.models`:
```json
{
"indigo/opus-4-6": { "alias": "indigo-opus" },
"indigo/sonnet-4-6": { "alias": "indigo-sonnet" },
"indigo/haiku-4-5": { "alias": "indigo-haiku" },
"indigo/gpt-5.4": { "alias": "indigo-gpt54" },
"indigo/gpt-5.4-mini": { "alias": "indigo-gpt54-mini" }
}
```
### 5. Restart OpenClaw
Models appear in `/model` picker. Select any `indigo/*` model to route through SubStation.
## Architecture
```
OpenClaw
|
v
SubStation Proxy (:8403)
|
+-- Claude models --> Agent SDK --> Anthropic API
| (real CC binary, persistent sessions)
|
+-- GPT models -----> Codex API --> chatgpt.com
(direct HTTP, SSE streaming)
```
## Endpoints
| Endpoint | Description |
|----------|-------------|
| `POST /v1/chat/completions` | Main proxy — OpenAI-compatible chat completions |
| `GET /v1/models` | List available models |
| `GET /health` | Pool status, version, token counts |
| `GET /pool` | Detailed per-token status |
## Models
### Claude (via Agent SDK)
| Model ID | Name |
|----------|------|
| `opus-4-6` | Claude Opus 4.6 |
| `sonnet-4-6` | Claude Sonnet 4.6 |
| `haiku-4-5` | Claude Haiku 4.5 |
### ChatGPT (via Codex API)
| Model ID | Name |
|----------|------|
| `gpt-5.4` | GPT 5.4 |
| `gpt-5.4-mini` | GPT 5.4 Mini |
| `gpt-5.1-codex` | GPT 5.1 Codex |
| `gpt-5.1-codex-mini` | GPT 5.1 Codex Mini |
| `gpt-5.1-codex-max` | GPT 5.1 Codex Max |
## Token sources
SubStation loads tokens from multiple sources at startup (deduped by value):
1. **`~/.substation/token-pool.json`** — Explicit pool (ChatGPT tokens from auth script)
2. **`auth-profiles.json`** — All `anthropic:*` profiles with a `.token` field
3. **`SUBSTATION_OAUTH_TOKENS`** env var — Comma-separated tokens (assumed Anthropic)
## Privacy
- SubStation **never stores or logs message content**
- Requests are proxied through and discarded
- Token IDs (not values) appear in logs for debugging
- No telemetry, no analytics, no external calls beyond the LLM APIs
- All data stays on your machine
## License
MIT
integration
Comments
Sign in to leave a comment