You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Browser Copilot — an Automa-style visual workflow builder with an AI agent that builds the flow for you
#2167
I've been building Browser Copilot — an open-source Chrome extension that pairs an AI side-panel agent with an Automa-style visual workflow builder. Since this community understands browser automation best, I wanted to share it here, and especially the part where the AI generates the workflow for you.
👉 [Drop/paste your workflow-canvas screenshot on this line — GitHub uploads it automatically]
The idea: stop hand-wiring every flow
Automa's block canvas is great, but assembling a multi-step flow by hand still takes time. Browser Copilot gives you three ways to get a workflow, and two of them don't involve dragging blocks at all:
Describe it in plain language — the agent builds the flow. Ask in the side panel (e.g. "open my GitHub review queue, summarize each open PR, and give me a digest"), watch the agent read pages and click through the task, then save the executed actions as a workflow. The steps land on the canvas as blocks and re-run deterministically afterwards — no model in the loop, no token cost.
Record it once. Press Record and do the task yourself; clicks, form inputs (text/select/checkbox/radio), scrolling, tab switches, navigations and SPA route changes are captured as blocks — including the wait-for-element / wait-for-load pauses.
Draw it. The same node-graph canvas you already know, with 56 executable blocks ported from Automa: click/hover/scroll, form fill, get text, element-exists, loops over data or elements, while/repeat, condition branches, variables and {{token}} interpolation, JavaScript, cookies, webhook, clipboard, screenshot, downloads, new/switch/reload/close tab, delay, and more. Every block has its own edit form, can be disabled individually, and supports per-block error handling — retry or fall back to another branch on failure.
An AI agent block inside the canvas
There's a dedicated AI agent block: drop it anywhere in a graph and that step is handed to the same agent loop the chat uses. In read-only mode it reads the page (or a specific element) and answers without acting; in full-auto it can click and navigate. Its answer is stored in an output variable that later blocks can interpolate — so you can mix deterministic blocks with LLM judgment in a single flow (e.g. scrape a list → let the agent classify each item → branch on the result).
A few more details:
🧠 Bring your own model — any OpenAI-compatible endpoint: DeepSeek, OpenAI, OpenRouter, Moonshot, DashScope, SiliconFlow, or a local Ollama / LM Studio. Function calling is required for the agent to act.
⏰ Triggers beyond manual — run a flow on a schedule, at browser startup, via a keyboard shortcut, from the context menu, or when a matching page is opened.
🎯 Precision targeting — a built-in element picker generates CSS or XPath selectors (hover to highlight, click to lock, walk the DOM), and reaches elements inside closed shadow DOM through the Chrome DevTools Protocol. One click auto-lays-out the graph.
🔒 Private by construction — no accounts, no telemetry, no cloud server; your keys and data stay in your browser.
The workflow engine is explicitly modeled on Automa, so if you like Automa's canvas but want an LLM that can generate and step through flows — and then save them as reusable, deterministic automations — give it a try. Feedback, issues and ideas are very welcome 🙌
reacted with thumbs up emoji reacted with thumbs down emoji reacted with laugh emoji reacted with hooray emoji reacted with confused emoji reacted with heart emoji reacted with rocket emoji reacted with eyes emoji
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Hi Automa community 👋
I've been building Browser Copilot — an open-source Chrome extension that pairs an AI side-panel agent with an Automa-style visual workflow builder. Since this community understands browser automation best, I wanted to share it here, and especially the part where the AI generates the workflow for you.
👉 [Drop/paste your workflow-canvas screenshot on this line — GitHub uploads it automatically]
The idea: stop hand-wiring every flow
Automa's block canvas is great, but assembling a multi-step flow by hand still takes time. Browser Copilot gives you three ways to get a workflow, and two of them don't involve dragging blocks at all:
{{token}}interpolation, JavaScript, cookies, webhook, clipboard, screenshot, downloads, new/switch/reload/close tab, delay, and more. Every block has its own edit form, can be disabled individually, and supports per-block error handling — retry or fall back to another branch on failure.An AI agent block inside the canvas
There's a dedicated AI agent block: drop it anywhere in a graph and that step is handed to the same agent loop the chat uses. In read-only mode it reads the page (or a specific element) and answers without acting; in full-auto it can click and navigate. Its answer is stored in an output variable that later blocks can interpolate — so you can mix deterministic blocks with LLM judgment in a single flow (e.g. scrape a list → let the agent classify each item → branch on the result).
A few more details:
The workflow engine is explicitly modeled on Automa, so if you like Automa's canvas but want an LLM that can generate and step through flows — and then save them as reusable, deterministic automations — give it a try. Feedback, issues and ideas are very welcome 🙌
All reactions