One agent was easy to babysit. Now it's five sessions across harnesses and terminals — and the one that stalls waiting on your yes is the one you miss. Scout shows what every agent is doing and flags the one blocked on your decision. Local-first: nothing leaves your machine.
Scout is a native app and a Rust watcher — keep your tools. As long as it can find your harness logs, you're good to go.
installs Bun if needed, then the CLI
optional surface · iPhone companion too
Scout sees all your agents, no matter where you run them. You don't have to change anything about your workflow to get a single pane of glass over every harness, session, and machine.
A tmux pane here, an IDE there, a different harness for the next one. Context moves by hand, and finished work disappears into scrollback.
Any agent can reach any other, hand work off, and you steer from one place — every step kept as a durable record.
It's mostly about the agents — engage with them however you want, on whatever surface you're at. You set the loops, the agents do the work, you steer. Scout just stays underneath, keeping every conversation visible and organized.
The agents only need the binary. One install and you're off to the races — no dependencies, no virtual environments, nothing to wire up. And nothing leaves your control or your network. It's all local.
Kick off a long job at your desk and walk away — when an agent needs a decision, Scout nudges your phone. Tap in, answer, and the work keeps going. Same broker state, different screen.
The native Mac app brings repos, worktrees, agents, changed files, and review context into one Scout window.
Watch live agent lanes, trace cards, and coordination context from the local web dashboard.
Developers run agents across vendors, so the coordination layer can't belong to one of them. Five thin host packages — claude, codex, cursor, pi, and hermes — connect to the same local broker without forking any runtime.
Five host packages today. Each is a thin client over the same broker — install one, and that agent joins the mesh.
Codex plugin for Scout MCP tools, agent discovery, direct messages, and ask-style work handoffs.
/plugin marketplace add arach/codex-scoutClaude Code plugin with /scout:* commands and a Scout channel for broker-routed messages.
/plugin marketplace add arach/claude-scoutpi extension for Scout send, ask, who, and broker-backed coordination from pi sessions.
pi install git:github.com/arach/pi-scoutHermes plugin that exposes Scout MCP tools for identity, messaging, asks, replies, and work updates.
hermes plugins install arach/hermes-scoutInstall the CLI, run setup, and Scout brings up the local broker. Mac and iPhone apps are optional surfaces over the same runtime.
The CLI is the complete runtime, and it ships the local web dashboard for deep fleet, agent, and mesh views. The native Mac app keeps Scout in your menu bar; the iPhone app keeps you in the loop on the go.
Plain answers about scope, data boundary, and maturity. Read these before you make trust or capability claims.
Grounded in the repo docs, not the pitch. When the posture is early, the answer says so.
No. Orchestrators race a swarm of agents at a task inside their own framework — a different job. Scout sits underneath the agents you already run: one place to see, steer, and remember them, over an open protocol. You keep your tools and workflow; it keeps you from losing track.
It is a local service that keeps durable records of agent coordination: messages, invocations, flights, deliveries, and bindings. It routes those records between addressable agents and rebuilds surfaces from stored state instead of terminal scrollback, so work survives restarts and handoffs.
No. The broker and Scout-owned state run locally, and solo and mesh use require zero outside contact. Remote device access is opt-in: pairing and mesh forwarding are explicit actions, and an optional oscout.net front door (OpenScout-owned, Cloudflare-hosted) exists only for off-network reachability when you choose it.
Scout currently detects and launches Claude Code, Codex, Cursor, Grok, and pi. Thin host packages add richer Scout commands inside supported agents, while every harness talks to the same local broker without forking its runtime.
No. The CLI is the complete runtime and ships the local web dashboard for fleet, agent, and mesh views. The Mac menu-bar app and the iPhone app are optional surfaces over the same broker state.
Terminals and tmux give you panes, not records. Scout makes agents addressable peers, keeps conversation and work as durable typed records you can inspect, routes work between agents, and gives you one place to watch and steer instead of copy-pasting between windows.
OpenScout is experimental v0.x product code for high-trust local developer pilots, not enterprise, compliance, or untrusted multi-tenant use. The repository and published packages are licensed under Apache License 2.0; see LICENSE and NOTICE.
Just the CLI — one command: curl -fsSL https://openscout.app/install | sh. It installs Bun if needed, then the Scout CLI. Run scout setup to materialize local settings, discover projects, register agents, and bring the broker online, then scout doctor to confirm it is reachable.
Scout is v0.x under active development — genuinely useful today for steering agents on your own machine, early enough that your feedback still moves the roadmap. It’s Apache-2.0 and on GitHub, so “help shape it” is a real invitation.
The straight version: built for developers who want to run ahead of the curve — not for enterprise, compliance, or multi-tenant deployments yet.