EdgeClaw

OpenClaw on Cloudflare Workers. Persistent personal AI assistant running on CF Durable Objects + Workers AI — no server, no machine, no SQLite file to babysit.

What this is

OpenClaw runs as a local daemon on your machine: SQLite for state, a long-running process for channel connections, your own hardware for LLM inference. EdgeClaw takes the same model and runs it on Cloudflare's edge:

OpenClaw (local)	EdgeClaw (CF Workers)
SQLite per agent	Durable Object per agent (SQLite-backed)
Local daemon process	DO hibernation — always available, no idle cost
LanceDB vector memory	Vectorize (coming)
LLM API calls	Workers AI (free inference)
Channel socket listeners	Incoming webhooks
~/.openclaw/ filesystem	KV + R2

Same idea. 300 edges. No machine required.

Deploy in 5 minutes

git clone https://github.com/Stackbilt-dev/edgeclaw
cd edgeclaw
npm install

# Create KV namespace
npx wrangler kv:namespace create edgeclaw-skills
# Paste the returned id into wrangler.toml → kv_namespaces[0].id

# Deploy
npx wrangler deploy

# Set your channel secrets
npx wrangler secret put TELEGRAM_BOT_TOKEN
npx wrangler secret put TELEGRAM_SECRET

# Wire Telegram webhook
curl "https://api.telegram.org/bot<TOKEN>/setWebhook" \
  -d "url=https://edgeclaw.<your-subdomain>.workers.dev/channels/telegram&secret_token=<SECRET>"

That's it. Message your bot.

Channels

Channel	Status	Setup
Telegram	✅	wrangler secret put TELEGRAM_BOT_TOKEN + setWebhook
Slack	✅	wrangler secret put SLACK_SIGNING_SECRET SLACK_BOT_TOKEN
HTTP REST	✅	POST /chat — for testing and integrations
WhatsApp	🔜	Coming
Discord	🔜	Coming

Architecture

Channel webhook → Hono router → AgentSession DO (per user)
                                      ↓
                               Workers AI (Llama 4 Scout)
                                      ↓
                          SQLite history + KV memory

Each user gets their own AgentSession Durable Object — persistent conversation history, isolated state, hibernation when idle (no compute cost). Workers AI handles inference (free on Cloudflare's network).

Adding skills

Skills are functions registered on the AgentSession. Drop a file in src/skills/ and import it in agent-session.ts. Skills can read/write KV, call external APIs, or query D1. The model calls them via tool use.

See src/skills/ for examples (coming).

Relation to AEGIS

EdgeClaw uses the same Cloudflare primitives as AEGIS (the production cognitive kernel), but packaged as a clean deployable template for personal use. If you want the full thing — memory layers, autonomous goals, scheduled tasks, multi-agent governance — run AEGIS. If you want a personal assistant on CF in 5 minutes, start here.

License

MIT — same as OpenClaw.