run your agents wherever the work is. one open-source agent harness.
octomind is one Apache-2.0 agent harness that works with any model. the same runtime runs in your terminal, on a server, behind your app and inside your editor. it is not an ai agent framework you code against: it is an agent runtime you configure in TOML, so moving an agent means moving a file.
run it yourself: one open-source binary
- in your terminal pair with an agent in your shell on any model, and resume any session later.
- self-hosted agent an agent on your own box that takes work from webhooks, timers and your shell.
- agentic workflows multi-step agent pipelines in one TOML file: parallel, loop until it passes, fan-out.
- mcp client plug any stdio or HTTP mcp server into a terminal agent, and add more mid-session.
- agent orchestration a lead agent hands work to specialists at runtime. you get one bill and one verdict.
- self-improving agent correct the agent once. it saves the fix as a file and recalls it next session.
- agent backend your app talks to one WebSocket. octomind runs the agent, its tools and the bill per user.
- in ci pipe a diff in, get JSONL or schema-shaped output back, with a per-request spend stop.
leaving another tool? when octomind fits, and when it doesn't →
or have octomind cloud run it for you
- personal ai agent your own ai agent for each thing you run, reporting back in Telegram, WhatsApp or Slack
- no-code agent builder say the job in one sentence and get an agent with a cloud machine, memory and schedule.
- agent hosting skip the Mac mini: a managed machine that wakes for the job and sleeps for storage cost.
- competitor monitoring checks rivals' pricing and feature pages on your schedule and tells you only what changed
- Telegram ai agent text your octo in Telegram: it does the job on its own computer and replies there
same model. the harness decides.
on octobench, octomind and opencode both ran glm-5.2 through the same endpoint at the same prices, on 25 real merged pull requests. only the harness changed.
| harness | solved | cost |
|---|---|---|
| octomind | 24/25 | $63.43 |
| opencode | 19/25 | $129.54 |
start with one command
Apache-2.0 and any model. install it once, then run it whichever way fits the job.
curl -fsSL https://octomind.run/install.sh | bash