Models & providers
Connect a model account, API key, or local endpoint.
Bring your own key (BYOK)
bdibranding talks to models through your credentials. The simplest path is an OpenRouter API key: one key, hundreds of models, transparent per-token pricing. Paste it once in Settings and every agent can use any model you allow.
Supported providers
The provider picker is a first-class roster, not one generic endpoint wearing different labels:
- Account sign-in: ChatGPT (Codex), Grok, and Kimi for Coding. These use the provider's sign-in flow instead of asking you to paste an API key.
- Direct API keys: OpenRouter, OpenAI, Anthropic, Google Gemini, xAI, Groq, Mistral, DeepSeek, Together AI, Fireworks AI, Perplexity, and Cerebras.
- Local and custom: Ollama on your machine, or any custom endpoint that speaks the OpenAI-compatible wire format.
Where keys live
Keys are stored locally in your OS keychain via the desktop shell — never in plain project files, never sent anywhere except the provider they belong to. Provider requests leave your machine only when an agent actually runs.
Service keys (the KEYS tab)
Beyond model providers, agents sometimes need platform keys to do real work — a Stripe test key, a GitHub token, a search API key. The KEYS tab lets you paste any service key; bdibranding exposes it to agent runs as an environment variable and tells the agent only the name of what's available, never the value. The key itself never enters the model prompt.
3D models (Meshy)
Paste a Meshy API key in KEYS and any agent
with a STUDIO in its room can build 3D models from a description or a picture. Each model is
saved to the agent's workspace as a .glb file. Place a HOLO BAY on your station and
click it to turn the models over, zoom in, and switch between solid, hologram and wireframe views.
Every generation spends your Meshy credits: the run reports the exact credits Meshy charged, and
scheduled or Night Shift runs can only use the key once you approve it for unattended use.
Cost honesty
Every run records what it actually cost, from the provider's own usage numbers. The ledger shows real spend per agent and per task — there is no abstraction layer marking up your tokens, and no meter that guesses.