The AI agent that lives
inside your product.
Two script tags. One agent that navigates your app, calls your APIs, and answers from your docs — so users finish the job without leaving the page. Wired up by your coding agent, live in an afternoon.
Two script tags · No backend rewrite · Self-hostable
Ask, navigate, and act inside this app
Add a 30-minute intro call to my event types.
Added “30 Min Intro Call” — your booking link is cal.com/you/30min-intro.
Works in any web app — React, Vue, Rails, or plain HTML.
A small runtime, not a platform to migrate to.
ActBrow embeds with two script tags and configures over a typed API. Self-host it, point it at your own model, and keep your stack.
Two scripts
Embed with actbrow-sdk.js + actbrow-widget.js. No backend rewrite.
API-first
Tools, flows, and knowledge sync over one typed sync API.
Scoped keys
Separate setup and widget keys, scoped per assistant.
Self-host
Run the whole stack with Docker. Bring any OpenAI-compatible model.
Your users are stuck. Your engineers are buried.
Every product accumulates the same tax: workflows that humans have to babysit and features only engineering can build.
Users get lost in the UI
They open support tickets asking where to click instead of finishing the task themselves.
Manual, multi-tab workflows
Refunds, onboarding, config changes — repetitive journeys that nobody automated because it was never worth a sprint.
Every helper is a custom build
In-app guides and copilots get hand-coded, then rot. A permanent maintenance tax on your roadmap.
Chatbots that can't actually do anything
Bolt-on assistants answer questions but can't navigate your app or call your APIs. Users still do the work.
The result: rising support costs, slower execution, and an engineering backlog that never clears.
Stop describing the work. Ship an agent that does it.
ActBrow turns the same workflows into one embed — an agent that acts inside your product instead of pointing at it.
- Weeks of glue code for every in-app helper
- Chatbots that only answer, never act
- Support headcount scales with usage
- Users abandon multi-step flows
- One embed, agent live in an afternoon
- An agent that navigates your app and calls your APIs
- Answers from your docs, retrieved on demand
- Multi-step flows run on a trigger phrase
See the agent work — right in the chat
ActBrow loads as a chat widget on your site. Open it and ask — it navigates, calls your APIs, runs flows, and answers from your docs. Click through what that looks like.
Take me to my upcoming bookings.
Used 2 steps
You're on Bookings — your Thursday calls are highlighted.
Powered by ActBrow
Reads page context and navigates to the right place.
Two minutes, end to end
Create an assistant, sync tools with your coding agent, embed the widget, and watch the agent act inside a real app.
Everything an in-app agent needs
A complete runtime — setup, actions, tools, flows, and knowledge — that matches what actually ships.
Agent-configured setup
Open Connect, paste the setup prompt into Claude Code or Codex, and your agent pushes tools + knowledge through the sync API.
No YAML. Integrated in an afternoon.
In-app navigation & context
The agent navigates to any route and reads live page context, so it always knows where the user is.
Grounded in the real page, not guesses.
HTTP & REST tools
Turn any endpoint into a typed tool. Run same-origin in the browser or server-side.
Reuse the APIs you already have.
Navigation flows
Define multi-step journeys triggered by a phrase to guide users through onboarding, setup, or recovery.
Repetitive journeys run themselves.
Knowledge base
Sync domain docs; the agent retrieves them on demand via knowledge.search instead of stuffing every prompt.
Accurate answers, lower token cost.
Model-flexible
Point ActBrow at any OpenAI-compatible model endpoint and configure it in one place.
No lock-in. Swap endpoints freely.
Integrate with a script, not a sprint
No SDK gymnastics, no backend rewrite. Sign up, let your coding agent push config through the sync API, then paste the embed snippet into your app.
- 1
Create an assistant
Spin one up in the dashboard and open Connect.
- 2
Let your coding agent sync it
Paste the setup prompt into Claude Code or Codex — it pushes tools and knowledge through the typed sync API.
- 3
Drop in two script tags
Add the embed snippet before
</body>and the agent goes live.
<script src="https://actbrow-backend.depak.dev/actbrow-sdk.js"></script> |
<script> |
window.ActbrowWidgetConfig = { |
assistantId: "YOUR_ASSISTANT_ID", |
baseUrl: "https://actbrow-backend.depak.dev", |
apiKey: "wk_...", |
navigate: function (path) { window.location.assign(path); } |
}; |
</script> |
<script src="https://actbrow-backend.depak.dev/actbrow-widget.js"></script> |
Scale operations without scaling headcount
Give every team an agent that executes inside the product — and keep full control of your data, keys, and stack.
Cut operational cost
Deflect repetitive support and manual ops to an agent that works 24/7.
Ship faster
Automate a workflow in minutes instead of a sprint of custom UI work.
Multiply your team
One agent covers journeys that would take a team of operators to staff.
Scale with demand
Handle usage spikes without proportional headcount growth.
What you control
- Scoped widget & setup keys
- Per-user assistant ownership
- Server-side secret storage
- Browser or server tool execution
- Typed sync API
- OpenAPI tool import
- Conversation history
- Self-host with Docker
Self-host to keep data in your own infrastructure.
Connect the APIs you already run on
ActBrow doesn’t lock you into a fixed list of connectors. Turn any REST endpoint into a tool, or import an OpenAPI spec and generate them all at once.
HTTP & REST tools
Point a tool at any endpoint. Typed inputs and outputs, run server-side or same-origin in the browser.
Import an OpenAPI spec
Paste a Swagger / OpenAPI v3 spec and ActBrow generates one typed tool per operation, grouped as an integration.
Incremental re-sync
Re-import any time. ActBrow diffs the spec and adds or removes tools to match — no manual cleanup.
{"key": "orders.refund","type": "SERVER_HTTP","method": "POST","path": "/api/refunds","execution": "server"}
Have an OpenAPI spec? Import it and your whole API becomes agent-ready in one step.
Or hand the sync API to Claude Code / Codex and let your coding agent wire the tools automatically.
Stop shipping help articles.
Ship an agent that does the work.
Create an assistant, let your coding agent sync the tools, and embed the widget. Your users get an agent that acts — in an afternoon.
Two script tags · No backend rewrite · Self-hostable
