Termic is a desktop app for running AI coding-agent CLIs in parallel. It spawns the real claude, codex, agy (Antigravity), copilot, grok, pi, opencode, muse (Meta’s Muse Code) and devin binaries inside terminals, each in its own git worktree, in one window. It’s free, open source (AGPL-3.0), and runs entirely on your machine.
If you’ve used Conductor, the layout is familiar: a project on the left, agents working in isolated copies of your repo, a diff view to check what they did. Termic keeps that workflow and changes a few things, covered below.
No SDK, just the CLI
Termic doesn’t use the vendor SDKs. It spawns the same interactive CLI you’d run in iTerm, the literal claude binary, and talks to it over a PTY (a pseudo-terminal).
Anthropic’s June 15, 2026 Claude Agent SDK credit change put the SDK path (and claude -p, the GitHub Actions integration, and any third-party app that authenticates through the SDK) on a separate per-user monthly credit: $20 on Pro, $100 on Max 5x, $200 on Max 20x. When that credit runs out, SDK requests stop unless you opt in to overflow at standard API rates.
Tools built on the SDK draw from that credit. Termic doesn’t touch the SDK, so it draws from your regular subscription quota, the same pool you spend when you type claude in a terminal. Termic adds no per-token markup and no second bill.
Because Termic runs the CLI and nothing else, the day Anthropic, OpenAI or Google ships a new model, slash command or tool, you have it after a brew upgrade. There’s no wrapper to catch up. As long as the claude CLI exists, Termic works.
What you get
- Parallel worktrees. Every task is a real
git worktree, so N agents can work on N branches at once without stepping on each other. See Tasks & worktrees. - Main checkout mode. Attach an agent straight to your main checkout for a quick question or a README fix. See Tasks & worktrees.
- Multi-repo tasks. Put your backend, frontend and infra repos under one task with a shared
CLAUDE.md, and let the agent work across all of them. See Multi-repo tasks. - A per-task sandbox. A macOS Seatbelt cage plus a network allowlist, so an agent can run unattended with permissions skipped and still not be able to read
~/.sshor phone home. See Sandbox overview. - Bring your own agent. The agent list is an editable registry. Drop in
aider,opencode,ollama, or a shell script. See Agents & the registry. - An editor and diff view, fast search, and Spotlight. See Working in Termic.
Who it’s for
Termic is for developers who already pay for Claude Pro or Max (or use Codex / Grok) and want to run several agents at once without paying twice. It’s a developer tool, built by someone who uses it every day, on macOS first, with Linux and an experimental Windows build shipping every release.
Where to go next
If you just want it running, go to Install and then Your first task. If you’re coming from Conductor, the migration guide maps every concept across and points out where Termic does things differently.