Bootspring Browser
Bootspring's own browser-automation layer — agents navigate, read, and verify their own web changes. Pluggable engine, white-labeled, part of Agent Flow early access.
Early access · Engine capability — Bootspring Browser is part of the Agent Flow engine. It is built and tested, but not yet exposed as a standalone CLI or MCP tool. It powers agent self-verification today; direct end-user access is on the roadmap. Request early access →
Bootspring Browser is our own browser-automation layer — built and maintained in-house. It lets a Bootspring agent drive a real browser to verify its own web changes: navigate to a page, read what's there, interact, screenshot, and assert that the change actually landed before reporting success.
Why it exists
An agent that edits a web app should be able to confirm its work — an agent you can't see is an agent you can't improve. Bootspring Browser closes that loop: after an agent makes a change, it opens the page and checks it, then reports a pass/fail verdict back into the fleet. That verdict becomes part of the run's receipt.
Pluggable engines
Bootspring Browser exposes one owned API over interchangeable engines. The engine is an internal detail — you work with the Bootspring surface, not the engine.
| Engine | Use it for | Notes |
|---|---|---|
| Full browser | Interaction — click, type, screenshot, run scripts, verify JS-rendered pages | Uses a Chromium engine under the hood, resolved on demand |
| Fetch | Fast text checks — "is this text on the page?" | No browser needed; ideal for CI, servers, and quick verification |
Because the engine is pluggable, the same agent code runs against a full browser when it needs interaction, or a lightweight fetch engine when a text check is enough — with no change to how you use it. Additional engines can be added behind the same API.
What an agent can do
- Navigate to a URL and wait for the page to settle
- Read the page title and visible text
- Interact — click elements, fill inputs, run scripts (full-browser engine)
- Screenshot a page for evidence
- Verify a page against a set of checks and get a structured verdict — the core self-verification primitive
Self-verification in a fleet
Inside Agent Flow, a browser-QA agent is a node like any other. It verifies a URL against checks, then messages the orchestrator with the result:
- An agent makes a web change.
- A browser-QA agent opens the page and runs its checks.
- It reports a pass/fail verdict to the orchestrator over the two-way bus.
- The verdict is recorded in the run's receipt.
This is the same mechanism that verified Bootspring's own /agent-flow page during development — confirming the page rendered correctly with no stale content.
What is available today
| Piece | Status |
|---|---|
| Browser automation engine (navigate/read/interact/screenshot/verify) | Built and tested |
| Pluggable engines (full browser + fetch) | Built and tested |
| Self-verification inside Agent Flow fleets | Built and tested |
| Standalone CLI / MCP tool for direct use | Roadmap |
Learn more
- Agent Flow — the surface Bootspring Browser is part of
- Request early access