Watchfire is free, open source, and self-hosted.

• Line item 1

  Agent API tokens.

  This is the only line item with a per-task dollar figure on it. Watchfire is a control surface; the agent backend you point it at — Claude, Codex, opencode, Gemini, Copilot, Cursor — is the thing actually spending money. A short, well-scoped task on a workhorse model costs cents. A long refactor on a frontier model reading a large codebase costs dollars. For per-backend ranges and the three things that actually move the bill, read what it costs to run Watchfire.

• Line item 2

  Compute on your machine.

  Watchfire runs as a local daemon on your laptop, dev box, or CI runner. Each agent session is a PTY-managed subprocess in its own sandboxed git worktree, so “compute” here is mostly CPU for the agent itself, disk for the worktree copies, and a few hundred megabytes of RAM for the daemon and the GUI. On a normal development machine, the marginal cost is effectively zero. On a rented runner, it is whatever your runner already costs — Watchfire does not add a tier on top.

• Line item 3

  Your time reviewing the diff.

  The unglamorous one, and the one that quietly dominates the others. Every agent run produces a branch and a diff that needs a human to decide whether it ships. Watchfire is designed to make that review fast — the worktree, the transcript, the YAML task spec are all right there — but it does not remove the step. The operator is the loop. Budget reviewer time the same way you would budget API tokens.

• No per-seat fees.

  There is no "team" SKU. Five engineers on one project pay the same as one engineer on five projects — nothing.

• No per-task or per-run fees.

  Wildfire mode can churn through dozens of sessions overnight. Watchfire charges nothing for any of them. The agent backend bills you directly; we are not in that loop.

• No telemetry surcharge.

  There is no telemetry that turns into an invoice. No usage you have to opt out of to avoid a bill.

• No "enterprise" gate on features.

  There is one edition. Every mode, every integration, every backend, every release is in the same open-source repo under the same license.

The pricing questions we get asked.

Three honest answers. More questions live on the FAQ page.

• Is there a hosted version of Watchfire?

  Not today. Watchfire is a daemon that runs on your machine, spawns coding agents inside sandboxed git worktrees, and stores all state on local disk. There is no hosted Watchfire Cloud, no managed control plane, and no waiting list to sign up for. If a hosted offering ever lands, it will be announced on the changelog and the blog — and the self-hosted edition will remain free and open source.

• Will Watchfire ever charge for the open-source product?

  The self-hosted daemon, CLI/TUI, and GUI are released under the Apache License 2.0 and will keep being released under that license. Charging for what is already in the repo would mean re-licensing code that has shipped to users, which is not on the table. If we ever build paid services on top — hosted runners, managed teams, support contracts — those would be additions, not replacements.

• What about commercial or enterprise use?

  Apache 2.0 permits commercial use, modification, redistribution, and embedding Watchfire inside your own tools, with no separate "commercial" license to negotiate. There is also an explicit patent grant. Companies are using it the same way a single developer would — clone the repo, install the binaries, point it at an agent backend. Read the LICENSE for the binding terms.