Model online 90–130 tokens/s
A model for coding agents.
Connected in two minutes.
Brawl 0.1 Beta works in Cursor, Claude Code, Codex, OpenCode and six more agents. OpenAI- and Anthropic-compatible API with a 262K-token context window. Free during the beta.
No password: we’ll email you a code. Your first sign-in creates an account.
- Readworker.py
- Grep"counter +=" · 3 matches
- Editworker.py +4-2
- Bashpytest -q · 18 passed
The counter was updated from two threads without a lock. Added threading.Lock around the increment — all 18 tests pass.
Agents
Works with your agent
Step-by-step guides for ten agents. Any other client with an OpenAI-compatible API works too.
-
Verified
OpenCode
Terminal and app
-
Verified
Claude Code
Terminal
-
Verified
Codex
Terminal
-
Verified
Cursor
Editor
-
Cline
VS Code extension
-
Roo Code
VS Code extension
-
Kilo Code
VS Code extension
-
Continue
VS Code and JetBrains
-
Zed
Editor
-
Aider
Terminal
Windsurf can’t be connected: it doesn’t let you set a custom model endpoint. GitHub Copilot in VS Code and JetBrains AI Assistant connect like any client with an OpenAI-compatible API.
Features
Everything your agent needs
-
262K tokens of context
Room for a large module, its tests and a long history of edits.
-
90–130 tokens per second
Repeated agent steps hit the prompt cache, so only the first reply in a long chat takes a while.
-
Three compatible APIs
OpenAI Chat Completions, OpenAI Responses and Anthropic Messages on one endpoint.
-
Three reasoning levels
low for quick answers, medium by default and high for hard problems.
-
Sees and draws images
Understands screenshots, and when you ask it to draw something, the picture appears right in the agent chat.
-
A key per device
Separate keys for your laptop and your server. See when each key was last used and revoke it in one click.
How to connect
Three steps to the first answer
-
01
Sign in with email
We’ll send a 6-digit code, no password needed. Your first sign-in creates an account.
-
02
Create a key
We show it only once. Save the key to a file, and your agent will read it from there.
-
03
Connect your agent
Pick your agent and OS and follow the steps. The dashboard marks your first request.
API
Works with the OpenAI and Anthropic SDKs
Change the base URL and the model name — the rest of your code stays the same.
- Base URL
- https://llm.brawl.io/v1
- Model
- brawl-0.1-beta
- Anthropic Messages
- https://llm.brawl.io
- Reasoning level
- -low · -high
import os from openai import OpenAI client = OpenAI( base_url="https://llm.brawl.io/v1", api_key=os.environ["BRAWL_API_KEY"], ) stream = client.chat.completions.create( model="brawl-0.1-beta-high", # high reasoning messages=[{"role": "user", "content": "Find the race in worker.py"}], stream=True, ) for chunk in stream: print(chunk.choices[0].delta.content or "", end="")
import OpenAI from "openai"; const client = new OpenAI({ baseURL: "https://llm.brawl.io/v1", apiKey: process.env.BRAWL_API_KEY, }); const stream = await client.chat.completions.create({ model: "brawl-0.1-beta-high", // high reasoning messages: [{ role: "user", content: "Find the race in worker.py" }], stream: true, }); for await (const chunk of stream) { process.stdout.write(chunk.choices[0]?.delta?.content ?? ""); }
curl https://llm.brawl.io/v1/chat/completions \ -H "Authorization: Bearer $BRAWL_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "model": "brawl-0.1-beta-high", "messages": [{"role": "user", "content": "Find the race in worker.py"}] }'
import os from anthropic import Anthropic client = Anthropic( base_url="https://llm.brawl.io", api_key=os.environ["BRAWL_API_KEY"], ) message = client.messages.create( model="brawl-0.1-beta-high", # high reasoning max_tokens=32768, messages=[{"role": "user", "content": "Find the race in worker.py"}], ) print("".join(block.text for block in message.content if block.type == "text"))
Reasoning levels
Depth that fits the task
Pick a level with a model name suffix or your agent’s switcher: in OpenCode it’s Ctrl+T.
-
low
Quick answers, small edits and questions about code.
brawl-0.1-beta-low -
medium default
Everyday agent work: a balance of speed and quality.
brawl-0.1-beta -
high
Hard problems and debugging. Thinks 4–9× longer.
brawl-0.1-beta-high
FAQ
Frequently asked questions
Didn’t find an answer? The dashboard has a guide for every agent and fixes for common errors.
How do I sign in without a password?
Enter your email and we’ll send you a 6-digit code. It’s valid for 10 minutes, and you can request a new one after a minute. If you don’t have an account yet, your first sign-in creates one.
How much does it cost?
Nothing: Brawl is free while it’s in beta.
Where are my requests processed?
On Brawl’s own server. Request logs are kept for 30 days, with keys masked. See the privacy policy for details.
Why does the first reply in a long chat take a while?
Agents send the whole chat history with every request. The model reads the first request in full, at about 3,000 tokens per second, so a 64K-token history takes around 20 seconds. Later steps come from the cache and start almost at once. For a new task, start a new chat.
What should I do if my key leaks?
Revoke it on the API keys page of your dashboard: requests with it stop working within seconds. Then create a new key and update the file on the devices that used the old one.
Can I connect GitHub Copilot or Windsurf?
GitHub Copilot in VS Code takes a custom endpoint on paid plans: in Manage Models, add a Custom Endpoint provider with https://llm.brawl.io/v1. JetBrains AI Assistant takes one through an OpenAI-compatible provider. We haven’t tested either yet. Windsurf doesn’t accept a custom model endpoint, but you can install Cline or Roo Code inside it.
Connect Brawl to your agent
Passwordless email sign-in. Your agent’s setup guide is ready right after you sign in.