One spec in.
A fleet of agents out.

Jerico is a self-hostable fleet orchestrator for CLI coding agents. You write one Markdown spec; a planner agent decomposes it into a dependency graph of todos; you approve the plan; Jerico dispatches those todos in parallel to the agents already installed on your own machines — Claude Code, Codex, Qwen, opencode, Kimi, Antigravity and more — and streams every terminal panel into a single browser tab.

In tmux you are the scheduler: you decide what runs next, you re-paste context between panes, and you are limited to one machine. Jerico plans the work into a dependency graph, routes each todo to a planner, developer or reviewer agent, and dispatches independent todos in parallel across every machine running a Jerico daemon — while you watch and approve from a browser tab.

Yes. Jerico exposes its orchestration surface over MCP. An MCP-connected orchestrator agent can create dependency-aware todos, spawn and assign workers, inspect their live output, and report work complete while the browser remains your observation and intervention surface.

An approved run returns paused, not erased. Jerico persists its todos, dependencies, transcripts and retry context; after reconnect, you can resume without rebuilding the graph by hand. A watchdog retries timed-out todos and fails work that remains stuck. Plans that have not yet been approved are not included in this guarantee.

Jerico detects and drives the CLI agents already installed on your machine: Claude Code, Codex, Qwen, opencode, Kimi, Antigravity, Aider, Ollama, Forge, GitHub Copilot CLI, and a plain shell. Jerico drives them over a PTY, so it uses your existing subscriptions and API keys — there is no Jerico model and no Jerico API key.

Yes. Each machine runs one bridge-agent daemon that connects out to your server. Todos in the same dependency layer are dispatched in parallel across all connected machines, so a run can span a laptop, a tower and a Linux VM at the same time — all visible in one browser tab.

After a planner produces the execution plan, Jerico pauses the run for your approval. No implementation todo is dispatched until you approve that plan; reviewer agents then gate implementation work during the run.

Jerico scans every todo for high-blast-radius shell before it is dispatched — rm -rf, DROP TABLE, TRUNCATE, terraform destroy, kubectl delete --all, docker system prune, git push -f, dd, mkfs and more. A flagged todo never reaches the agent: it stops in awaiting confirmation until you click through, and its dependents wait behind it. That is a separate gate from the plan approval — it fires even on work you already approved.

The daemon connects outward, over an authenticated connection, to the Jerico server you configure — in the self-hosted setup, that server is yours. It sends terminal output, commands, agent status and supported usage metadata to that server; agents and credentials stay installed locally. The console and the daemon add no third-party analytics, trackers or session recording — these public marketing pages (this page and the guide) use Microsoft Clarity for aggregate funnel analytics. The privacy page is the complete disclosure.

The Jerico daemon (bridge-agent) is MIT-licensed and published on npm, so you can read and audit exactly what runs on your machine. The console is self-hostable — you run the server, you own the data.

There is no native mobile app. Use the web console from a phone to check a run or approve a plan; terminal-heavy work is still better suited to a larger screen.