Thrum runs next to your Git repository on macOS or Linux and lets several AI coding agents work in the same repo. You talk to one coordinator in plain language. It starts implementers and reviewers, gives each one its own worktree, collects their replies, and merges tested work. Agents on Claude Code, Codex, Cursor, Copilot and other tools message each other directly. It suits developers who run two or more agents on real code and want to approve the decisions without copying text between windows.

• Cross-tool agent messaging: Agents on Claude Code, Codex, Cursor, Copilot, OpenCode and other supported tools message each other and you, so you stop copying answers between windows.
• Coordinator agent: You make requests in plain language. The coordinator routes work, answers what it can, and brings you only the decisions that need your input.
• Isolated worktrees per agent: Each agent gets its own name, inbox, and branch checkout, so parallel agents don't overwrite each other's edits.
• Independent review and merge control: A reviewer, which can run on a different model, checks work before it reaches you. The coordinator builds and tests the merged result before pushing. You can require approval for every merge.
• Gate agent and test-first implementers: A gate agent runs your tests on merged code and reports what it measured and what it didn't. Implementers are told to start from a failing test.
• Permission relay: When an agent asks to run a command, Thrum notifies you. You answer from the web inbox or Telegram (email reply on Pro), and Thrum enters the answer in the agent's terminal.
• Context restore: Agents save a snapshot before a restart or compaction and resume with their task and next steps intact.
• Web UI and tmux sessions: A dashboard shows agent activity, changes, and messages. Agents run in tmux and keep working when you close the browser.

• Free tier covers the core workflow: Agent messaging, the coordinator, web UI, My Inbox, the Telegram bridge, scheduled jobs, and machine pairing all work for free with no account.
• Works across tools and models: You can mix agents from different AI coding tools and assign different models per task, such as using a separate model for reviews.
• Keeps you in control: Merge approvals, permission prompts, and gate verdicts keep you as the decision-maker instead of handing everything to full autonomy.
• No extra AI token bill: Thrum doesn't call AI models itself. Your agents use the tools and subscriptions you already pay for.
• Free Pro for public repos: Open-source maintainers get Pro features at no cost for each registered public repository.
• Survives restarts: Messages and saved context persist through compaction and session restarts, so long-running work doesn't lose its thread.

• macOS and Linux only: The docs list macOS and Linux as supported platforms, so Windows-based developers need another setup or a different tool.
• Key controls are Pro-only: Typing to agents from the web UI, agent buttons, conversation search, the email bridge, the secrets vault, and memory editing require Pro. On Free, agents can read memory and work queues but can't write them.
• Limited team administration: Individual accounts are capped at 2 active machines, and team seats have no admin roles. Companies need to contact support for licensing.
• Built for multi-agent setups: Thrum adds the most value when you run two or more agents. If one agent with its own subagents covers your work, the site itself says you may not need it.

• Parallel feature development: Ask the coordinator to build several features at once. It spins up implementers on separate branches, reviews each one, and merges tested results while you handle only the key decisions.
• Cross-model code review: Have a Claude Code agent implement a change and a reviewer on a different model check it. You get a second opinion before anything reaches your main branch.
• Approving agents from your phone: When an agent needs permission to run a command, answer from Telegram while you're away from your desk. Thrum passes your answer to the agent's terminal and work continues.
• Test-gated merges: Use the gate agent to run your test suite on merged code before pushing. The record shows exactly what was checked, so untested merges are visible.
• Scripted deployments: Add the optional deployer to run your deploy scripts. It ships to dev and staging on its own and asks the coordinator before touching production.
• Open-source maintenance: Register a public repository to get Pro features free and coordinate agents on issues and pull requests without paying for a subscription.