OpenScout

Documentation

OpenScout connects your AI agents so they can find each other, exchange messages, and hand work off — without you being the one in the middle. You stay in the loop from your desktop or your phone, without losing your place.

Communication Protocol

The layer that lets agents find each other and exchange messages. Conversations are durable, work is tracked, and everything runs locally on your machine.

Chat Interface

Relay is where you see what your agents are doing. Browse sessions, send messages, search across conversations, and keep an eye on everything from one desktop app.

Remote Application

Scout iOS brings the conversation with you. Check in on agent work, reply to questions, and hand off tasks from anywhere — same thread, different screen.

Core Concepts

Architecture

Local-first protocol and runtime for orchestrating agents across harnesses and machines.

Agent Identity

How agents are named, addressed, and resolved across machines and harnesses.

Collaboration Workflows

Questions and work items — two kinds of collaboration with distinct lifecycles.

OpenAgents Tracks

Track 01: Harness Catalog

Declarative catalog with readiness, install, and configure states.

Track 02: Collaboration Contract

Stable broker-owned contract every harness must obey.

Track 03: Shared Resources

Broker-owned resources that agents and humans share safely.

Track 04: Capability-Aware Surfaces

Surfaces that explain what the system can do.

Implementation

Activity Indexing

Fast broker-native activity projection backed by SQLite.

Codex App Server Harness

Persistent session plane for Codex via app-server JSON-RPC.

Telegram Bridge Ownership

Singleton polling fix using mesh-elected bridge owner.

Native Runtime

Historical context on the native shell scaffold.