Choose an agent

One sentence

Clicking around a website and pulling data? Use browser-use. Also need to write code, run a terminal, or process files over many steps? Use browsercode. If you omit the agent, requests route to browser-use (cheaper, covers the common case).

The two agents

	`browser-use`	`browsercode`
Best for	One-off or short browser tasks: navigate, click, fill forms, extract data. Do it and done.	Anything needing code, a terminal, or long-horizon multi-step work — with or without a browser.
Capabilities	`browser`, `files`	`browser`, `code`, `terminal`, `files`
Session	disposable (task-scoped)	durable (extend over hours, return later)
Cost / speed	fast, cheap, open-source	more powerful, slower, pricier
Run	`POST /api/v1/agents/browser-use/runs`	`POST /api/v1/agents/browsercode/runs`

Let the registry route for you

GET /api/v1/agents returns a machine-readable description of every agent — the routing truth. An LLM (or your code) reads it and picks, no hardcoding.

from browser_use import BrowserUse

client = BrowserUse()
agents = client.agents.list()
for a in agents:
    print(a.name, a.capabilities, "—", a.best_for)

import { BrowserUse } from "browser-use";

const client = new BrowserUse();
const agents = await client.agents.list();
for (const a of agents) console.log(a.name, a.capabilities, "—", a.bestFor);

curl https://api.browser-use.com/api/v1/agents \
  -H "X-Browser-Use-API-Key: $BROWSER_USE_API_KEY"

Response:

{
  "agents": [
    {
      "name": "browser-use",
      "capabilities": ["browser", "files"],
      "session": { "kind": "disposable" },
      "bestFor": "One-off or short browser tasks: navigating sites, clicking, filling forms, extracting data. Do it and done.",
      "notFor": "Writing or running code, using a terminal, or long-horizon multi-step engineering work.",
      "exampleTasks": ["Book a flight on united.com", "Scrape all prices from this catalog"],
      "run": "POST /api/v1/agents/browser-use/runs"
    },
    {
      "name": "browsercode",
      "capabilities": ["browser", "code", "terminal", "files"],
      "session": { "kind": "durable" },
      "bestFor": "Anything needing code, a terminal, or long-horizon multi-step work, with or without a browser. Extend over many steps or return to it later.",
      "notFor": "Simple one-shot browser navigation (overkill; use browser-use).",
      "exampleTasks": ["Scrape a site then clean the data into a CSV", "Build and run a page monitor"],
      "run": "POST /api/v1/agents/browsercode/runs"
    }
  ]
}

Filter by capability: GET /api/v1/agents?capability=code returns only agents that can run code. Whichever agent you run, you pass shared resources the same way:

browser — the browser settings for the run, including profile_id to start logged-in (profiles).
attached_file_ids — files to give the run (files); workspace_id exists only for browsercode’s advanced cross-conversation sharing.
session_id — continue a previous run’s thread.

An agent never invents its own version of these. Learn one, use both.

Capabilities are enforced — sending a field an agent doesn’t support returns a 422 naming the problem (e.g. workspace_id on browser-use). Errors tell you the correct choice — read them.

Get started

Agents

Browsers

Account

More

One sentence

The two agents

Let the registry route for you

​One sentence

​The two agents

​Let the registry route for you

​Both agents share the same vocabulary

One sentence

The two agents

Let the registry route for you

Both agents share the same vocabulary