Many model services.One unified API endpoint.
Run Buzz on your own server and route every model request through one endpoint: provider keys in one place, a personal API key for each member, and routes, budgets and usage managed from one dashboard.
Docker Compose · Postgres + Redis
Compatibility
Compatible with leading model APIs
Requests pass straight through when the client and provider speak the same protocol. When they differ, Buzz converts what it supports — for example, calling an Anthropic model with the OpenAI SDK.
- OpenAI Chat Completions and Responses
- Anthropic Messages and Gemini GenerateContent
- Conversion covers text, images, basic tool calls and text streaming; see the docs for other capabilities and limits
| Protocol | Endpoint |
|---|---|
| OpenAI Chat Completions | /v1/chat/completions |
| OpenAI Responses | /v1/responses |
| Anthropic Messages | /v1/messages |
| Gemini GenerateContent | /v1beta/models/{model}:generateContent |
| Gemini streaming | /v1beta/models/{model}:streamGenerateContent |
Reliability
Multiple routes with automatic failover
Each model can have several routes. Buzz distributes requests by policy and retries or fails over automatically on errors.
- Distribute requests by round robin or weight, with configurable priorities
- Automatic retries, key cooldowns and failover to the next route
- Invalid keys (400, 401, 403) are automatically disabled, and their status is recorded
- Dashboard
- My API Keys
- Users
- Models
- Providers
claude-sonnet-5.5Policy · Round robinclaude-sonnet-5.510ActiveCooldown until 14:32anthropic/claude-sonnet-5.55ActivePOST /v1/messagesclaude-sonnet-5.5- Anthropic429 Rate limited
- Key on cooldown until 14:32; trying the next route
- OpenRouter200 1.4s
Teams & costs
Member keys and budgets
Provider keys are kept inside Buzz. Members sign in to manage their own API keys, and administrators set a budget for each member.
- Administrator and user roles; users can only view the dashboard and their own API keys
- Weekly credits are spent first, then lifetime credits; requests are refused once both run out
- Set the credit cost per million tokens for each model, with detailed usage including cached input and reasoning tokens
Model catalog
Manage models in one place
Add models in bulk from presets or load the provider’s model list, and sync parameters such as context and output limits.
- Add from presets or load upstream models
- Sync context, max output and capability parameters
- Clients such as Pudding can read this model information directly
One-command install
Run the install script on your server to start Buzz together with Postgres and Redis. To upgrade, run the same command again or use make upgrade.
curl -fsSL https://raw.githubusercontent.com/teatak/buzz/main/install.sh | shThen open http://<server-ip>:9622/admin/
Get started
Quick start
- 01
Deploy
Run the install command to start Buzz and its databases.
- 02
Add providers
Create an admin account the first time you open the dashboard, then add providers, their API keys and the models you want to offer.
- 03
Connect apps
Create an access key for each member, then enter your Buzz address and that key in any compatible client.
Does Buzz include model credits?
No. You need your own provider accounts and API keys, and providers bill you for usage.
Can I limit how much each person uses?
Yes. Set weekly and lifetime credit limits for each user. Requests use credits at the rate configured for each model. New requests are rejected when credits run out; setting both limits to 0 allows unlimited usage.
Do I need Pudding to use it?
No. Buzz works on its own with any compatible client, and it can also serve as Pudding’s model endpoint.
Can I host and modify it myself?
Yes. Buzz is MIT licensed. Installation steps, configuration docs and the full license are on GitHub.
Get started with Buzz
MIT licensed. Runs on your own server with Docker Compose.