AI agentsAI agents

Build AI agent workflows that run in your browser

Give the agent tools and a goal, and it works out the steps. BNOD's ai_agent block runs an autonomous, multi-step loop — it reasons, calls the tools you give it, reads the results, and keeps going until the job is done. Summarize a page, classify a list of rows, pull one answer out of a document: you describe the outcome, the agent figures out how to get there. It runs as a Chrome extension inside the browser you already use, on the pages you're already logged into.

Most AI agent platforms run in the cloud, on their own infrastructure, with their own model markup baked into a monthly bill. BNOD flips that: the agent runs locally, in your session, and it uses your own API key — so you pay your provider directly at cost, and the agent can act on the browser data in front of you instead of a sanitized cloud copy.

How an agent is built in BNOD

An agent in BNOD is a node with two kinds of sub-nodes wired into it, all in the visual editor.

• A chat-model sub-node. Pick the model that does the reasoning — any of your BYOK providers.
• Tool sub-nodes. Attach the tools the agent is allowed to use — an HTTP request, a sub-workflow, a scrape step — and the agent calls them on its own when it needs them.
• Autonomous multi-step runs. The ai_agent loop plans, calls a tool, reads the result, and decides the next move — repeating until it reaches the goal or you stop it.
• Grounded in real browser data. Because it runs in your tab, the agent can reason over the page you're on, the list you just scraped, or the file you just extracted.

With 40+ blocks in the visual node editor, the agent is one part of a bigger workflow — you can scrape a list first, run each row through the agent, and export the enriched result.

Bring your own key — your model, at cost

The one thing cloud agent platforms can't offer is a bill with no model markup. BNOD is bring-your-own-key across 10 providers: Anthropic, OpenAI, Gemini, OpenRouter, Groq, DeepSeek, Mistral, xAI, Ollama, and any OpenAI-compatible endpoint. There's also a ChatGPT Plus sign-in option if you'd rather use an account you already pay for. You choose the model per agent, and every token is billed by your provider directly — BNOD doesn't sit in the middle.

Ollama in that list means you can run a local model with no API cost at all. And because everything executes in your own browser, the page content, the scraped rows, and the model calls stay on your side — nothing routes through a BNOD server.

What people build with BNOD agents

Because it runs in your own logged-in browser, the agent works on the data and pages you already have access to. Common jobs:

• Enrich scraped rows — run each row of a list through the agent to categorize, tag, or add a summary.
• Classify at scale — sort a page of items, messages, or listings into buckets you define.
• Read a page, extract an answer — point the agent at a page and have it return the one fact you need.
• Summarize long content — condense an article, a thread, or a report into a few lines.
• Multi-step reasoning over browser data — let the agent chain tool calls to work through a task you'd otherwise do by hand.

Start from a working example instead of a blank canvas — the templates gallery ships 14 ready workflows you can adapt, and the AI Workflow Builder can wire an agent from a sentence.

How it works in 3 steps

1. Pick a model. Attach a chat-model sub-node and choose one of your BYOK providers.
2. Give it tools and a goal. Wire in the tool sub-nodes it may use and describe the outcome you want.
3. Run. The agent loops — reason, call a tool, read the result — locally in your browser until the job is done.

FAQ

Is it free? BNOD is free to install. The agent uses your own API key, so you pay your model provider directly at cost — there's no BNOD subscription and no model markup. With Ollama you can run a local model for free.

Which AI providers can I use? Ten: Anthropic, OpenAI, Gemini, OpenRouter, Groq, DeepSeek, Mistral, xAI, Ollama, and any OpenAI-compatible endpoint — plus a ChatGPT Plus sign-in option.

Can the agent call tools? Yes. You attach tool sub-nodes — an HTTP request, a sub-workflow, a scrape step — and the ai_agent block calls them on its own as it works toward the goal.

Does it run in the cloud? No. The agent runs locally in your own Chrome tab, so it can reason over the pages you're logged into, and your data doesn't pass through a BNOD server.

Can it work on data I scraped? Yes. Scrape a list first, then run each row through the agent to summarize, classify, or enrich it in the same workflow.

Do I need to write code? No. You build the agent in the visual node editor, or describe it to the AI Workflow Builder and let it wire the model and tools for you.

Want the agent to enrich real data? See web scraping with BNOD to capture a list first, then run it through your agent, and browse the templates gallery for ready-made workflows. When you're ready, add BNOD to Chrome and build your first agent in minutes.