Skip to main content
A model provider is one place Monad can send an inference request. Monad ships 24 built-in provider types — 8 with a dedicated SDK and 16 presets over the bundled OpenAI-compatible adapter — and any provider can also arrive as an Atom Pack. How to connect Monad to a model provider: pick a provider, add a credential, choose a default model. This is the user-facing companion to model-providers.md, which explains the provider catalog, the two provider strategies, and why the long tail of OpenAI-compatible providers needs no extra packages.

​
What you need

  • A provider account and an API key (a few providers, like a local Ollama server, work without one).
  • For self-hosted or account-scoped providers (Azure OpenAI, Ollama, and similar), the base URL of your endpoint.
The full list of built-in providers comes from the provider catalog — see the internals doc for how it is defined. Every setup surface below reads the same catalog from the daemon, so they always offer the same providers.

​
Three ways to set up

​
1. Guided terminal setup

The interactive wizard lists the provider catalog, prompts for the base URL when the provider needs one, prompts for any provider-specific fields (for example the AWS region for Amazon Bedrock), asks for the API key, tests the connection before saving anything, and then lets you pick a default model from the provider’s model list. If the test fails you can retry with a different key or go back and pick another provider. monad init --no-input (or -y) seeds the home directory without prompting, for scripted setups.

​
2. Explicit CLI commands

The CLI splits the surface into three nouns: provider, credential, and model. All of them accept the global --json flag for machine-readable output. Add or update a provider (the JSON shape is id, label, type, optional baseUrl and extra):
Attach a credential to a provider:
Probe a provider and key without saving either:
On success the daemon returns the provider’s model catalog, so this doubles as a model discovery command.

​
3. Web UI

Run monad on a fresh install and the browser opens the same setup flow. Afterwards, Studio → Models and providers (/studio/models) adds providers, manages each provider’s credentials, browses its models, and edits model profiles. See the web UI.

​
Model profiles and the default model

A model profile is a named recipe that maps roles to concrete models: chat is the required default route, and optional routes cover fast, vision, image, video, speech, transcription, embedding, and memory. A profile can also carry generation params and fallback targets. The fixed default profile is used whenever a request does not name one.
The web UI edits the same profiles under Studio → Models and providers.

​
Where things are stored

  • Providers, their credentials, and profiles live in agents.json under Monad’s config directory. Provider credentials are native settings and do not use Agent Runtime Credentials or ${secret:...} references.
  • Secrets never come back out either. monad credential list shows only a masked token preview, and monad credential add echoes just the new credential id.
  • The user is responsible for protecting the native settings file. The daemon and settings APIs continue to redact secrets from responses and logs.

​
Provider-specific notes

  • Azure OpenAI requires a base URL for your resource (https://{resource}.openai.azure.com/openai/v1), and the model id is your Azure deployment name.
  • Amazon Bedrock requires an AWS region (prompted as an extra field) and a bearer API key (ABSK…).
  • Neither Azure nor Bedrock exposes a standard model-listing route, so when the connection test returns no models, the wizard falls back to manual model-id entry — type the model or deployment id yourself.
  • Ollama and similar local servers treat the API key as optional; just point the base URL at your server.
Details and rationale for all of the above are in the internals doc.

​
Custom providers

A provider that is not in the catalog can still be added as a third-party atom pack: drop a module that exports a ModelProvider into ~/.monad/atoms/providers/ and the daemon picks it up without a restart. See model-providers.md for the contract.

​
For contributors

When developing Monad itself, copy .env.example to .env.local and set OPENROUTER_API_KEY; the dev environment seeds it as a working credential so live runs work out of the box. This is a dev-only path — release builds configure providers only through the flows above. See CONTRIBUTING.md.