OpenClaw Quickstart
What is Doubleword?
Doubleword is a hosted async inference cloud built for high-volume, non-real-time AI workloads. Your OpenClaw agent does a lot of work you don't see — overnight research, scheduled reports, email triage, memory maintenance. Most of this happens in the background, where speed doesn't matter but cost does.
Doubleword gives your agent a cheaper inference tier for exactly those tasks. While you chat in real time on a fast model, the background heavy lifting runs on Doubleword — at up to 4× lower cost.
| Mode | Latency | Cost | Your agent uses it for |
|---|---|---|---|
| Realtime | Immediate | Standard | Answering you in chat |
| Async | ~5 min in practice | Reduced | Background tasks that can wait a few minutes |
| High throughput | Hours | Lowest | Overnight jobs, large document pipelines |
Setup
Step 1 — Install the Doubleword skill
The skill teaches your agent how to use the Doubleword API — submitting jobs, checking status, downloading results, available models and pricing. Install it once and your agent knows everything it needs.
npx skills add https://github.com/doublewordai/batch-skillOr with git:
git clone https://github.com/doublewordai/batch-skillOnce installed, your agent can answer questions like "how do I submit a job?" or "what models are available?" — and more importantly, it can use the API on your behalf without you having to explain anything.
Step 2 — Add your API key
Sign up at app.doubleword.ai — $10 free credits on signup (~20M tokens). Create an API key under API Keys in the sidebar, then store it:
cat >> ~/.openclaw/.doubleword_creds << 'EOF'
DOUBLEWORD_API_KEY={{apiKey}}
DOUBLEWORD_API_BASE=https://api.doubleword.ai/v1
DOUBLEWORD_MODEL_DEFAULT={{selectedModel.id}}
EOFStep 3 — Tell your agent when to use it
Add a routing note to your workspace TOOLS.md:
## Doubleword (Async Inference)
- Credentials: ~/.openclaw/.doubleword_creds
- Use for: overnight research, cron jobs, summarisation,
document processing, synthetic data, model evals —
anything where waiting a few minutes is fine
- Not for: real-time chat, mid-conversation tool callsThat's it. Your agent now knows to reach for Doubleword on background tasks automatically.
What your agent can do with Doubleword
Once set up, just ask — your agent handles the rest:
| Say this | What happens |
|---|---|
| "Research our top competitors and send me a brief in the morning" | Agent submits an async job overnight, emails you results |
| "Summarise all the PDFs in this folder" | Agent processes them in bulk via the API |
| "Triage my unread emails by urgency" | Agent classifies async, reports back in minutes |
| "Generate 500 example support tickets for training data" | Agent runs a high-throughput generation job |
| "Run this eval set against three prompt variations" | Agent submits and compares results |
Further reading
- Doubleword API Skill — full skill reference
- API docs
- Async inference with Autobatcher
- GitHub: batch-skill