OpenClaw Config
Optimized workspace template for OpenClaw pro instances, powered by Fuel.
Clone this repo, add your Fuel virtual key, and you have a production-ready workspace with optimized inference, context management, and cost controls. No config to figure out.
This template combines best practices from the OpenClaw Token Optimization Guide by @mattganzak and the OpenClaw Runbook by @digitalknk. The key difference: this config uses Fuel as the model provider, so you get semantic role routing (worker/reasoning/heartbeat) with automatic cheapest-provider selection instead of managing provider keys and model IDs directly.
Quick Start
# 1. Clone the template
git clone https://github.com/openclaw-rocks/openclaw-config.git my-workspace
cd my-workspace
# 2. Replace the placeholder with your Fuel virtual key
# Get one at https://openclaw.rocks/fuel
sed -i '' 's/<YOUR_FUEL_VK>/vk-your-key-here/' openclaw.json
# 3. Customize your agent
# Edit SOUL.md (personality), USER.md (your context)
# 4. Set this directory as your instance workspace
# In your OpenClaw dashboard, create or edit an instance and
# set the workspace path to this directory. OpenClaw reads
# openclaw.json, SOUL.md, USER.md, and AGENTS.md on startup.
What's in each file
| File | Purpose |
|---|---|
openclaw.json | Main config: Fuel models, concurrency limits, context pruning, compaction, caching |
SOUL.md | Agent personality and principles (template, customize this) |
USER.md | Your context: name, timezone, mission (template, customize this) |
AGENTS.md | Shared agent instructions: session initialization, memory rules, 402 handling |
HEARTBEAT.md | Heartbeat check definitions (system health, task status) |
memory/ | Memory directory for compaction flush (auto-populated) |
.gitignore | Ignores sessions, logs, credentials |
What this config saves you
| Optimization | What it does | Estimated savings |
|---|---|---|
| Multi-provider routing | Routes each role to the cheapest provider | 35-75% on inference costs |
Context pruning (cache-ttl) | Prunes stale messages after 6h | 30-50% fewer input tokens on long sessions |
| Session initialization | Loads 8KB instead of 50KB on session start | 80% fewer tokens per session start |
| Compaction at 40k | Distills context, flushes to memory files | Prevents runaway context that can 5-10x costs |
| Prompt caching | 90% discount on stable system prompts | ~$0.01/session instead of ~$0.10 |
| Cheap heartbeats (1h interval) | Dedicated low-cost heartbeat role | ~24 calls/day at near-zero cost |
| Automatic failover | Worker -> reasoning -> fallback providers | Agent doesn't die on provider errors |
| Concurrency limits (4/8) | Caps parallel calls | Prevents retry loop cost explosions |
| Budget controls (Fuel VK) | Hard spending limit | Agent physically can't overspend |
Typical result: An autonomous agent running 8+ hours/day costs $0.30-1.00/day with Fuel vs $3-5/day with default config.
How Fuel works
Your agent asks for worker, reasoning, or heartbeat. Fuel resolves that to the best real model at the cheapest provider and rewrites both request and response. Your agent never sees a provider-specific model ID.
| Role | Cost (per M tokens) | Use case |
|---|---|---|
fuel/worker | ~$0.67 input / ~$2.02 output | Routine coding, subagents, file operations |
fuel/reasoning | ~$0.72 input / ~$3.60 output | Architecture, complex reasoning, research |
fuel/heartbeat | ~$0.06 input / ~$0.24 output | Heartbeats, status checks |
Costs are approximate. We continuously optimize which providers and models back each role. Current model details are always visible at openclaw.rocks/fuel.
Region filtering
By default, Fuel routes to the best model globally. If you have data sovereignty or compliance requirements, change the baseUrl in openclaw.json:
https://inference.openclaw.rocks/v1 # Default: best globally
https://inference.openclaw.rocks/v1/~eu-eu # GDPR: EU origin + EU providers
https://inference.openclaw.rocks/v1/~all-us # Any origin, US providers only
https://inference.openclaw.rocks/v1/~us-us # US origin + US providers
Two dimensions:
- Provider region (after the
-): Where the API is physically hosted.eu= data only goes to EU-hosted APIs. - Model origin (before the
-): Where the model company is based.eu= only models from EU companies.
Every region filter has full role coverage, no gaps:
| Filter | Worker | Reasoning | Heartbeat |
|---|---|---|---|
~all-all (default) | DeepSeek V3.1 @ Fireworks | Kimi K2.5 @ Fireworks | GPT-OSS 20B @ Fireworks |
~all-us | DeepSeek V3.1 @ Fireworks | Kimi K2.5 @ Fireworks | GPT-OSS 20B @ Fireworks |
~us-us | Llama 4 Maverick @ Together AI | gpt-oss-120b @ Together AI | GPT-OSS 20B @ Fireworks |
~eu-eu | Devstral 2 (123B) @ Mistral | Magistral Medium @ Mistral | Mistral Small 3.2 @ OVHcloud |
Troubleshooting
| Problem | Fix |
|---|---|
401 Unauthorized | Check your virtual key in openclaw.json. It should start with vk-. |
402 Budget Exceeded | Credits exhausted. Top up at openclaw.rocks/fuel. |
429 Too Many Requests | Hit rate limits. Wait a moment or check concurrency settings. |
| Agent not using Fuel | Verify model IDs start with fuel/ in your config. |
| Context growing too fast | Verify contextPruning is set in openclaw.json. Check AGENTS.md session init rules. |
| Still loading full history | Session init rules missing from agent prompt. See AGENTS.md. |
| Heartbeats too expensive | Verify heartbeat.model points to fuel/heartbeat. |
Credits
This template builds on the work of:
- @mattganzak: OpenClaw Token Optimization Guide (session initialization, context pruning, rate limiting, model routing patterns)
- @digitalknk: OpenClaw Runbook (compaction, memory management, heartbeat tuning, security defaults)
Links
- Fuel managed service: get your virtual key
- Fuel repo: proxy source, model matrix, skill
- OpenClaw: your AI agent, live in seconds
License
MIT
还没有评论。