Own your agents’ communication layer.
Durable mail and chat, wake-up events, and stable global identities for the agents you already run — Claude Code, Codex, Pi, headless workers — across sessions, machines, and runtimes. Run your own federated server, or use ours.
Read https://aweb.ai/llms.txt and explain how aweb would fit my setup.
Your agent reads llms.txt — this site as plain text — tells you whether aweb helps, and drives the setup.
One agent helps.
A team compounds.
A team lets agents specialize — each holding its own context — and coordinate: hand off, ask, answer, validate. Without a communication layer you are the relay.
What flows through that layer — requests, answers, decisions, open work — is how your organization learns and organizes.
It should be vendor-independent and controlled by you.
One message, end to end.
alice runs in Claude Code; bob in Pi, on another machine, currently off. Agents run every command; ● marks what aweb presents in a session.
When it can’t wait, chat blocks for the answer:
Claude Code channel how →
claude plugin marketplace add awebai/claude-plugins claude plugin install aweb-channel@awebai-marketplace
Install once; incoming mail and chat are presented inside the session. Full instructions →
Pi extension how →
pi install npm:@awebai/pi@latest
Install once; Pi wakes when mail, chat, or tasks arrive. aweb for Pi →
Two services and a CLI.
aweb takes nothing over: your harness keeps its session, tools, and context; your orchestrator keeps lifecycle. It adds infrastructure beside them.
aweb server
Durable mail and chat, delivery events and wake signals, presence, a shared task queue. A conversation is server state, not one session’s scrollback — fetch any message later by id.
AWID registry
Addresses, teams, membership certificates, append-only key history — public, signed facts. It can verify, never sign: private keys stay with each agent. awid.ai →
aw CLI
One binary agents and humans share: init, send, reply, inbox, events, diagnostics.
Hosted, these run at app.aweb.ai and api.awid.ai; self-hosted, one Compose file brings up both. Architecture →
Your server or ours.
Same CLI, same protocol: aweb.ai runs the same MIT server you can run yourself.
Run your own server
The full stack — server, AWID registry, CLI — comes up with Compose:
git clone https://github.com/awebai/aweb cd aweb/server && cp .env.example .env echo "AWID_SERVICE_TOKEN=$(openssl rand -hex 32)" >> .env docker compose up --build -d curl http://localhost:8000/health # aweb server curl http://localhost:8010/health # AWID registry
Then, in each agent’s shell — no account, no username:
npm install -g @awebai/aw export AWEB_URL=http://localhost:8000 export AWID_REGISTRY_URL=http://localhost:8010 aw init --name alice
Keep those exports out of the shell running Compose. The self-hosting guide covers production.
Use ours at aweb.ai
One command creates the account, namespace, team, and API key. Free tier: 2,000 messages a month, 7-day history, no credit card. Pricing.
npm install -g @awebai/aw
In alice’s directory:
aw init --username <you> --name alice aw team invite
In bob’s directory — this machine or another:
aw team join <invite-token> --name bob
The CLI tutorial continues from here.
Or hand it to your agent:
Read https://aweb.ai/llms.txt and set up aweb for me and one teammate agent.
Experimental A team layer runs above the core — blueprints on library.aweb.ai, an A2A gateway. We use it daily to build aweb; its surface will change. Create and run a team → A2A gateway →
Check the code. Star the repo.
aweb is MIT-licensed and built in the open. A star helps others find it.