Agent Bridge connects an existing OpenWA WhatsApp session to a coding agent on your
own machine or server. Send an @me prompt in your self-chat, receive progress and
a labelled answer, and continue the same harness session with the next message.
The bridge normalizes each CLI’s output into one event contract. Command Code, Codex and Claude Code use the same channel and configuration dashboard; the Hermes OpenWA companion reuses the same gateway through its own adapter. Only one webhook receives new turns. Each harness retains its own history.
Control the bridge directly from the self-chat: @me /models lists the active
harness’s catalog, /model <id> verifies a model before applying it, and
/harness <id> verifies and switches adapters. Named threads use /new <name>,
/threads and /use <name>; each harness keeps its own thread history.
Images and documents can be attached with an @me caption. Use /send <path>
to retrieve a workspace file. Private context can include selected ChatGPT,
Claude and CLI transcript exports, while clarification options use numbered
text replies where a harness exposes a reply transport. See the
chat controls guide
for supported transports and limits.
Codex starts in a read-only sandbox, and Claude Code starts in plan mode with unanswered permission requests denied. Command Code also supports tool approvals through WhatsApp. The new adapters have been checked with real model replies and session resume, alongside deterministic tests for their event streams.
Try it locally
git clone https://github.com/mayurathavale18/agent-bridge.git
cd agent-bridge
npm ci
npm install -g @openai/codex
codex login
node src/index.ts --harness codex "Describe this repository"
No cluster or WhatsApp pairing is needed for the first CLI run. The minimal setup guide covers Windows executable paths, Claude Code, permissions and connecting an existing OpenWA session.
The project runs on my Contabo k3s cluster. Its dashboard is an operator interface protected by authentication. Source, editable technical diagrams and contribution instructions are available in the repository.