RelayPlane + OpenClaw: One Environment Variable
Set ANTHROPIC_BASE_URL to point OpenClaw at RelayPlane and get automatic rate limit management, per-session cost tracking, and intelligent model routing — with zero code changes.
Why OpenClaw users add RelayPlane
- Hitting Max rate limits? RelayPlane auto-fails over to a backup model so your session never hard-crashes on a 429.
- Want cost visibility? Every OpenClaw call is logged to local SQLite with model, tokens, and exact dollar cost — no cloud required.
- Want model routing? Simple tasks route to Haiku automatically. Complex ones escalate to Opus. Saves 40–60% on a typical OpenClaw spend.
Setup in 2 steps
No config files, no patches, no plugin. Just two commands.
Install RelayPlane
Point OpenClaw at RelayPlane
That is it. Every LLM call OpenClaw makes now flows through RelayPlane. Cost, model, and token count are logged locally. OpenClaw does not know or care.
What you get
Cost tracking per session
Every request logged to local SQLite — model, token count, and dollar cost. Queryable via the localhost:4100 dashboard.
Automatic failover on rate limits
Hit a 429? RelayPlane retries on a fallback model automatically. Your OpenClaw session keeps going.
Complexity-based model routing
Simple edits route to Haiku. Large context windows route to Opus. Configured once, applied to every call.
Zero code changes to OpenClaw
OpenClaw respects ANTHROPIC_BASE_URL natively. No fork, no plugin, no patch required.
OpenClaw without vs. with RelayPlane
| Feature | With RelayPlane | Without RelayPlane |
|---|---|---|
| Rate limit management When OpenClaw hits a Max rate limit, RelayPlane automatically retries on a fallback model or provider so your session keeps running. | Auto-failover on 429s | Hard error, session dies |
| Cost visibility Every call OpenClaw makes is logged to local SQLite with the exact model, token count, and dollar cost. Review it in the localhost:4100 dashboard. | Per-request SQLite log | No visibility |
| Model routing Simple tasks route to cheaper models like Haiku automatically. Large-context requests escalate to Opus. OpenClaw does not need to know about this. | Complexity-based routing | Single model always |
| Multi-provider fallback When Anthropic is overloaded, RelayPlane can fall back to OpenAI or Gemini. OpenClaw continues without interruption. | ||
| Zero code changes OpenClaw reads ANTHROPIC_BASE_URL from the environment. Point it at http://localhost:4100 and you are done. | One env var | N/A |
How it works under the hood
OpenClaw reads ANTHROPIC_BASE_URL from your environment and uses it as the Anthropic API endpoint. RelayPlane exposes a fully Anthropic-compatible HTTP server at localhost:4100. When OpenClaw sends a request, RelayPlane intercepts it, applies your routing rules, forwards it to the real provider, logs the metadata locally, and returns the response unchanged.
The entire stack runs on your machine. No data leaves your environment beyond the normal Anthropic API call. RelayPlane is MIT licensed with no telemetry.