What is Thrum

Thrum lets several AI coding agents work in one repository without you carrying messages between them. You talk to one agent, the coordinator. It starts the other agents, hands out work, runs reviews and comes back to you with results and decisions.

Five words you will see

• Agent: an AI coding assistant, such as a Claude Code session, that Thrum has given a name so it can work on your repository.
• Coordinator: the agent you talk to, which creates and stops other agents, hands out work, runs reviews and reports back to you.
• Message: a note from you or one agent to another named agent, which waits in the recipient’s inbox.
• Worktree: a separate checkout of your repository on its own branch, so each agent edits its own copy and agents do not overwrite each other.
• Briefing: the summary an agent reads when a session starts, with who it is, where the project stands and its own saved notes.

What it is

Thrum is a program that runs on your machine, on macOS or Linux, next to your Git repository. Agents can run in Claude Code, Codex, Copilot, Cursor, OpenCode, Muse or Antigravity (Mix runtimes and models). It gives each agent a name, an inbox and a worktree. It also gives you Thrum’s web UI, the page you open in your browser to see your agents and their messages.

You set Thrum up once and start your coordinator. After that you rarely type Thrum commands. You write to the coordinator in plain words, and it does the housekeeping: it creates and starts agents, cleans them up, runs reviews and keeps the repository tidy.

Everything on this page works on the Free plan, which needs no account. Pro adds running your agents from the web UI, such as typing into an agent’s terminal. Pricing lists what each plan includes.

Why Thrum exists

AI coding agents write code fast, and running several at once is where the work gets hard. Someone has to tell each agent what to do, carry questions and answers between them, check the result and remember what each one knows. Without help, that someone is you, copying text between terminals. With one coding agent, your coding tool is enough on its own.

Thrum moves that routine work to the coordinator, so you spend your attention on decisions. You still decide what gets built. By default the coordinator merges routine changes itself: it reviews each change first and is instructed to build and test the merged result before it pushes. If you would rather say yes to every merge, you can set it to ask first. The coordinator carries out the routine steps of your development process (plan, build, review, merge) and asks you only when a decision is yours.

How it fits together

                     +--> [Implementer]  builds in its own worktree
[You] <--> [Coordinator]
                     +--> [Reviewer]     checks the work in its own worktree

You send the coordinator a request. It creates an implementer and a reviewer, each in its own worktree, and sends them their work as messages. They answer by message. The coordinator collects the answers and reports to you in My Inbox, your inbox in Thrum’s web UI. This is a coordinator’s report in My Inbox after a review:

Where you will meet these words

The quickstart uses four of the five: you set up Thrum, and your coordinator creates a second agent in its own worktree and sends it its work as a message. The fifth, the briefing, shows up in Keep context when an agent restarts, where an agent restarts and still knows what to do next.

Next steps

• Quickstart: your first two agents: set up Thrum and ask your coordinator for a reviewer.
• Grow your team of agents, one step at a time: not sure how many agents you need? See what to add, and when.
• Keep context when an agent restarts: your agents pick up where they left off.