Make yourself
at home.

One harness. Sherlock and Watson.

We’re planning our own reworked and branded edition of open-source opencode. The RedBeard harness will bring our models, their tools, and your project into one workspace.

Sherlock is intended for more demanding code and reasoning. Watson is intended for everyday work with more generous use. The harness will be tuned for both, rather than asking you to configure everything yourself.

The rebrand and model integration are in development. There is no RedBeard download yet; the website shows an interface concept. These API examples use provisional model names for the planned service.

Your key is your access.

When access opens, you’ll receive a random key beginning with rb_. It connects your tools and opens your usage panel. There’s no email, password, or profile.

Save the key in a password manager. It is shown once, and we don’t keep a copy that can be recovered.

Bring RedBeard into opencode.

Add this provider to your project’s opencode.json, or your global configuration at ~/.config/opencode/opencode.json. Replace YOUR_KEY with your key.

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "redbeard": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "RedBeard",
      "options": {
        "baseURL": "https://redbeard.sbs/v1",
        "apiKey": "YOUR_KEY"
      },
      "models": {
        "sherlock": {
          "name": "Sherlock"
        },
        "watson": {
          "name": "Watson"
        }
      }
    }
  }
}

Select the RedBeard provider and your preferred model in your tool. The current Sherlock prototype accepts text; screenshots and other image input need a vision-capable model.

A familiar API.

RedBeard is designed to work with OpenAI-compatible clients. Set a custom base URL, add your key, and request a streamed response.

curl https://redbeard.sbs/v1/chat/completions \
  -H "Authorization: Bearer YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "sherlock",
    "stream": true,
    "messages": [{ "role": "user", "content": "Hello" }]
  }'

Know how much room is left.

Subscription plans have two allowances: one per 5-hour window and one per week. Reaching either limit pauses access until that window resets. Your panel shows both.

We count tokens, the pieces of text a model reads and writes. Generated text counts fully; input is designed to count for less. Exact metering will be confirmed before the beta.

Prefer occasional use? The prepaid API balance counts down as you work. See the plans and proposed rates.

Keep your work yours.

The service is designed to store usage counters, not your conversations. Your local tools may keep their own history, so check their settings if you want that removed too.

Never put your key in a public repository, share it in a screenshot, or paste it into an untrusted tool. Anyone with your key can use your allowance.