Last verified
Run AI coding CLIs in parallel git worktrees on a local desktop app; no SDK middleman, no per-token billing, on your existing plan.
Open four claude tabs, each in its own worktree, and work on separate features or bug fixes simultaneously. The broadcast command and work-done notifications reduce the overhead of monitoring multiple agent sessions.
You only ever run one agent at a time, or you prefer managing worktrees manually from the terminal with git worktree add.
Conductor stops working on 2026-11-20. Termic covers the same core workflow (parallel agents in worktrees, work-done detection, diff review) with an AGPL-3.0 license, local execution, and support for Codex, Antigravity, and custom CLI agents beyond Claude.
Your team depends on Conductor's cloud-hosted session storage or managed remote execution. Termic runs locally and does not replicate those features.
Enable the per-task macOS Seatbelt cage to let an agent run with --dangerously-skip-permissions without exposing ~/.ssh, ~/.aws, ~/.gnupg, or Keychains. The in-process network proxy limits outbound connections to per-CLI vendor APIs, GitHub, and package registries by default.
You are on Linux or Windows: the Seatbelt cage is macOS-only. Docker mode is the isolation alternative on those platforms.
Fire one prompt at claude, codex, and agy simultaneously using broadcast or agent races, each in its own worktree. Compare their diffs when they finish and adopt the winner, all from one app.
You only use one AI coding CLI. The parallel-agent value is specific to developers with subscriptions to more than one provider.
Running multiple AI coding agents in parallel against the same codebase creates three concrete problems: file conflicts when two agents edit the same path, billing surprises when a third-party orchestrator routes through the vendor SDK rather than the interactive CLI, and no reliable signal for when an agent actually finishes a turn.
The billing problem is specific to the Anthropic ecosystem since June 2026, when Agent SDK usage moved to a separate credit pool from the interactive Claude CLI. Any orchestrator that calls the SDK silently bills against that pool, while a developer who opens claude in a terminal does not. The file-conflict problem is common to every multi-agent workflow: without worktree isolation, two agents writing to the same project produce merge conflicts and lost work.
Each task creates a git worktree under ~/termic/tasks/<project>/<name>/, so multiple agents work on separate on-disk copies of the same repository without file conflicts. You can run N agents against the same branch across tabs, attach to repo root for one-off questions, or duplicate a worktree to spin up a parallel attempt from the same commit.
Termic spawns the real claude, codex, agy, copilot, and grok binaries in PTYs. Because it launches the interactive CLI rather than the vendor SDK, inference rides on your existing Pro or Max subscription rather than the separate Agent SDK credit meter Anthropic introduced in June 2026. No per-token markup is added.
An optional per-task cage pairs macOS Seatbelt (filesystem default-deny) with an in-process HTTPS CONNECT proxy (hostname allowlist). ~/.ssh, ~/.aws, ~/.gnupg, and browser profiles are blocked by default. The proxy restricts outbound network to per-CLI vendor APIs plus GitHub and package registries; add your own hosts per project. No daemon to install.
Shift+Cmd+B sends one prompt to every agent in a task simultaneously. For races, each agent gets its own fresh worktree and you compare their diffs N-up when they finish, then adopt the winner. These features make multi-agent code review and architectural exploration practical without switching between terminal windows.
Spotlight mirrors one task's changes (committed, uncommitted, and untracked) into the main checkout as a detached HEAD in real time, so your editor, dev server, and browser see the agent's work without you switching branches. It never commits to your branch; disabling it cleanly restores the checkout.
The agent registry in Settings is editable. Claude, Codex, Antigravity, Copilot, Grok, opencode, pi, Muse Code, and Devin ship as built-ins. Add aider, an ollama model, or any PTY-compatible tool in roughly 30 seconds by filling in the launch command.
Conductor is a commercial agent orchestrator that runs Claude in parallel worktrees with a checkpoint-commit approach. Conductor stops working on 2026-11-20.
| Feature | Termic | Conductor |
|---|---|---|
| License | AGPL-3.0 | Proprietary |
| Platforms | macOS, Linux, Windows | macOS only |
| Agents | Claude, Codex, Antigravity, Grok, Copilot, custom | Claude only |
| Deployment | Local desktop app, no backend | Cloud-backed sessions |
| Sandbox | macOS Seatbelt + network proxy | None documented |
| Cost | Free | Paid |
Termic spawns the interactive CLI rather than the Agent SDK, which keeps usage on a developer's existing Pro or Max subscription. Conductor's checkpoint approach commits to your branch; Termic's Spotlight feature syncs changes to the main checkout as a detached HEAD and leaves no commits behind.
Baton is a commercial tool for orchestrating AI coding agents with a managed cloud backend. Termic runs entirely locally with no account or backend server, so there is no service to authenticate with and no data leaving the machine.
| Feature | Termic | Baton |
|---|---|---|
| License | AGPL-3.0 | Proprietary |
| Deployment | Local desktop app | Cloud-managed |
| Agent types | Multiple CLI agents | Cloud-orchestrated |
| Offline use | Yes | No |
Termic is the better choice when data privacy, local execution, or CLI flexibility is a requirement. Baton is worth considering when managed cloud orchestration, team-level session sharing, or a hosted environment is a requirement that cannot be met with a local desktop tool.
Install on macOS via Homebrew; on Linux, the install script downloads the signed AppImage.
```bash
# macOS
brew install --cask simion/termic/termic
# Linux
curl -fsSL https://termic.dev/install.sh | sh
```Termic spawns the interactive claude CLI, the same binary as running claude in a terminal. Anthropic's June 2026 Agent SDK credit change moved the SDK and claude -p onto a separate credit pool; the interactive CLI stays on regular Pro or Max usage limits. Termic adds no markup and routes no inference through its own backend.
Yes. Termic is AGPL-3.0, free to use, and free to build from source. The AGPL means any fork or derivative must also remain open source under AGPL-3.0. The source is on GitHub at github.com/simion/termic. The app itself costs nothing; you pay only for whatever CLI subscriptions you already use.
No. Termic is a local app with no backend. Prompts travel from the terminal emulator through a PTY to the CLI binary to wherever the CLI sends them; Termic does not read or forward them. A debug log in the system temp directory is available but off by default. Because the app is fully open source, you can read exactly what it does and build it yourself to verify.
The per-task sandbox combines macOS Seatbelt (default-deny filesystem) with an in-process HTTPS CONNECT proxy (hostname allowlist). It blocks the agent from reading ~/.ssh, ~/.aws, ~/.gnupg, Keychains, browser profiles, and personal folders. Network is restricted to per-CLI vendor APIs, GitHub, and package registries by default. Enable it for any task where you want to run with --dangerously-skip-permissions without full system exposure. The sandbox is pinned at task creation and cannot be changed on a live task.
Yes, with caveats. Linux ships as a signed x86_64 AppImage; the in-app updater works the same as on macOS. Windows 11 x64 ships as an experimental per-user installer that is not yet code-signed, so SmartScreen asks once on first launch. The macOS Seatbelt sandbox does not exist on Linux or Windows; Docker mode is the isolation alternative. The rest of the feature set, including worktrees, parallel tabs, diff view, and broadcast, works the same on all three platforms.
Self-hosted AI agent management with org charts and budgets
#1 in the Conductor guide
Drop-in Conductor replacement with 27 or more agents on macOS, Windows, and Linux under MIT
#2 in the Conductor guide
MIT-licensed control plane for 6 or more agents with mobile and web clients and a large community
#3 in the Conductor guide
Daemon-based architecture for agents running on a VM or Docker host with full mobile parity
Sponsors
Alternative to
Superhuman
AI email assistant that drafts replies and clears your inbox
Alternative to
Grok Bot
AI teammates you own: your keys, your model, your machine.
Live monitoring on LinkedIn, X, Reddit & more. Agentic social listening for startups.
The projects we add and every new guide. Free, unsubscribe any time.
Claude Code to 
OpenCode
Jira to 
Windshift
Notion to 
AppFlowy
Slack to 
Mattermost
Linear to 
GitLab
Wispr Flow to 
Handy
Arc to 
Zen Browser
CapCut to 
Concat
Grok Bot to 
rakazo
Superhuman to 
Inbox Zero
Airtable to 
AITable
Muse to 
OpenClaw
Railway to 
Coolify
GitHub Copilot to 
Cline
Datadog to 
NetdataConfluence to 
Wiki.js
Microsoft Teams to 
Rocket.Chat
Firebase to 
Supabase
Pinecone to 
Qdrant
Zapier to 
n8n
1Password to 
KeePassXC
Devin AI to 
OpenHands
Asana to 
Leantime
Contentful to 
Strapi
