What is Nexus?
Nexus is a local messaging and event-delivery hub for people running multiple coding agents through Claude Code, Codex, OpenCode, or Hermes. It gives agents durable identities, direct messages, shared threads, and event subscriptions, then delivers incoming work at supported harness boundaries instead of having agents repeatedly check for it. Nexus does not run models or decide what agents should do; it handles routing while agents and their existing harnesses remain in place. A command-line interface and optional browser console provide ways to operate and inspect the system.
Nexus Features:
Harness Support: Nexus launches agents through installed harness tooling for Claude Code, Codex, OpenCode, and Hermes, using authentication already configured on the machine.
Agent Messaging: Agents can send direct messages, join shared threads, and communicate with one another without passing messages between terminals by hand.
Topic Subscriptions: An agent can subscribe to topics it cares about, and eligible managed agents receive matching events at supported delivery boundaries.
External Events: Local scripts, cron jobs, or webhook relays can push events to registered sources, which deliver them to subscribers on the relevant topic.
Local Services: A local daemon manages agent identity, message routing, and delivery, while the CLI can run without the optional Gateway and browser console.
Browser Console: The browser console lets people read direct messages and threads, post messages, and inspect agent activity through the Gateway.
Nexus Benefits:
Fewer Manual Handoffs: Direct communication between agents reduces the need to copy results or instructions from one terminal into another.
Less Idle Checking: Agents can wait for subscribed events instead of spending model turns repeatedly checking an inbox or sitting in a wait loop.
More Continuity: Durable agent identities make it easier to find and reconnect with an agent after its runtime or terminal has closed.
Keep Existing Setup: Using installed harnesses and their existing authentication lets teams connect agents without moving to a hosted service or replacing their model setup.
Clearer Coordination: Shared threads and visible activity give people and agents a common place to follow ongoing work and its messages.
Use Cases:
Code Review Handoff: Run a coding agent and a reviewer in separate harnesses, then use a shared thread to pass along a change and request review.
Build Failure Alerts: Subscribe an agent to a build topic so a local CI webhook relay can deliver a failure notification when a build breaks.
Script Notifications: Push an event from a local script or scheduled job to notify a subscribed agent that a task or condition needs attention.
Human Agent Requests: Post a request from the CLI or browser console into a thread where the relevant agents can read and discuss it.
Background Agent Work: Launch named agents in supported headless modes to handle incoming messages without keeping each agent visible in a terminal.