Choose models and tools

Models determine how Maia reasons and responds. Toolsets determine which external actions Maia can attempt. Configure them independently and expose the smallest useful capability set.

Select a provider and model

maia model

Maia supports direct API providers, OAuth-based providers, OpenAI-compatible endpoints, local models, and provider-routing configurations. The selector shows the providers and models currently available to the installed version and connected account.

Protect provider credentials

• Enter API keys through setup, the dashboard Keys page, or the managed environment file.
• Do not place keys in config.yaml, prompts, memory, skills, cron prompts, or company documents.
• Use separate credentials for development and production-like company environments.

Configure toolsets

Toolsets group related actions, such as files, terminal, browser, cron, memory, skills, messaging, or MCP tools. Use Maia's tool configuration interface to enable or disable them for each platform.

maia tools

A messaging user receives the intersection of platform tool configuration and governance authorization. Enabling a toolset does not bypass role, file, terminal, or approval policy.

Connect MCP servers

1. Review the MCP server's publisher, command, URL, and required credentials.
2. Add the server definition without embedding secrets in shared configuration.
3. Limit the exposed MCP tools where filtering is supported.
4. Test with a non-sensitive request.
5. Apply governance and approval rules to the actions the server can reach.

Verify effective capabilities

Ask Maia to list its active tools from the identity and platform being tested. Then attempt one allowed operation and one prohibited operation. Confirm that the audit trail records policy-sensitive denials or approvals.