One terminal. Every agent you have. Working at once.
Claude Code, Codex, and OpenCode are remarkable at running one task deeply. Medulla is what runs a hundred of them. It decides what work to hand out, places each piece on a harness that can do it, streams back what every one of them is doing, and keeps a live picture of the whole operation in front of you.
Fleets with everyone.
macOS and Linux
curl -fsSL https://raw.githubusercontent.com/tinyhumansai/medulla/main/install.sh | shWindows
irm https://raw.githubusercontent.com/tinyhumansai/medulla/main/install.ps1 | iexEither script downloads the prebuilt binary for your platform, verifies its SHA-256 against the release manifest, and installs to ~/.medulla/bin (%USERPROFILE%\.medulla\bin on Windows). If it updated your PATH, open a new terminal — or run exec $SHELL — so medulla resolves.
Then:
medulla login # browser sign-in
medulla # start the orchestratorNot ready to sign in? medulla --mock runs a full offline demo — no account, no network.
Prebuilt binaries ship for Linux (x86_64, aarch64), macOS (Apple Silicon), and Windows (x86_64). Building from source, pinning a version, and the SDK are covered in Developers → Getting Started.
Your whole fleet, legible. One lane per agent, live. See what each is doing, answer the one that has a question, cancel the one that has gone wrong — without losing your place in everything else.
Any machine, in two steps. Your laptop runs work by default. Add a build box or a server by pasting one line into an SSH session; it prints an address, you paste it back. Now the fleet is bigger.
Repositories it understands. Point Medulla at your projects once. It writes a short profile for each and uses it to route work to the right place, rather than guessing from a directory name.
Plans that actually run. A workflow is a saved, multi-step plan whose steps each run as a real harness session — with parallel branches, and approval gates where a human has to say yes. Ask for one in plain words and an agent will build it for you.
Small surface, low spend. The bulk of your fleet's output never reaches the orchestrator's context. It reasons over a distilled, current picture, so what you pay orchestrator rates on stays small however much is running underneath.
Routing › Harnesses also manages named OpenRouter-backed presets that reuse
Claude Code or Codex as the coding harness. Presets select an OpenRouter model
and fleet host while referring to OPENROUTER_API_KEY by environment-variable
name only; restart the local host after saving one. See
config.example.toml for the complete shape.
Full documentation: tinyhumans.gitbook.io/medulla
- Workers and Sessions — capacity, threads, and what survives.
- Workflows — authored multi-step plans and their runs.
- MEDULLA.md Workspace Profiles — telling the orchestrator what a repo is.
- Orchestrator Routing — cognitive tiers, harness selection, strategies.
- Token Efficiency and Budgets — small surfaces and enforced budgets.
Building on Medulla, or running it yourself? Everything technical — the TUI in depth, the CLI, worker daemons, configuration, architecture, and the SDK — is in Developers.
Medulla is in early alpha, and access is gated. It is rolling out to a small group of OpenHuman subscribers first, alongside gated API access for select teams building agentic systems. Alpha partners get direct access to the team, and their workloads shape what Medulla becomes.
Request access and tell us what you are orchestrating.
Ask a harness to coordinate other harnesses and you hit the same quiet failure mode everywhere: the orchestrator is just another model with a transcript, and every harness it manages writes into that transcript. Accuracy degrades well before the context window fills. An orchestrator that reads raw fleet traffic stops scaling at a handful of agents — long before it runs out of room, it stops being able to think.
Orchestration is becoming the dominant pattern in agentic systems, yet it has been running on architectures designed for chat. A chat model manages one thread. An orchestrator has to hold an operation in its head: agents in flight, work being decomposed and delegated, results streaming back, decisions made continuously. Medulla is built for that.
This repository is the Rust workspace behind the product: the medulla SDK and the medulla-tui crate that ships the binary. The orchestrator is gated; this code is not — read it, build it, and run the whole thing offline against the mock runtime. Start with Contributing.