3.13.3 Configure AI models

The AI engine tells Moss which model handles complex work and which handles quick answers. Start by understanding reasoning vs fast defaults.

Reasoning vs fast

Mode	Use for	Typical choice
Reasoning	Main chat, heavy tasks, device triage, code changes	Stronger cloud or team-hosted model
Fast	Short Q&A, summaries, file skim	Cheaper model or local Ollama

If fast mode points to local Ollama but the model is not running, AI Dock prompts you to open 3.12 Local LLMs.

Field reference

Field	What to enter
Display name	A label you recognize, e.g. “Company model” or “Local fast”
Provider	Match your platform or self-hosted stack
Model name	Exactly as the provider documents (case-sensitive)
API key	Usually required for remote models; often empty for local Ollama. Enter in settings—do not paste keys into chat
Base URL	For self-hosted gateways; leave empty for vendor defaults

After saving, pick different models independently for reasoning and fast.

Wrong provider

Provider must match the platform. “OpenAI-compatible” services use the matching compatible type; local Ollama uses Ollama. Wrong provider often shows failed tests, unknown model, or invalid key.

If unsure, share provider docs or internal field specs with Moss—redact keys in screenshots.

Relationship to local LLMs

The local LLMs page installs, starts, downloads, and tests Ollama models; the AI engine assigns them to Moss. Recommended: on local LLMs click “set as fast model config”, then confirm in AI engine.

Relationship to OpenClaw

On-device agents can follow Moss’s reasoning model, but some models cannot run on device. Local Ollama on the PC is not reachable from the board by default—the UI warns when that matters.

Connectivity test

Click test before relying on a profile. If it fails, check:

• Provider choice
• Model name spelling
• Full API key copied (no extra spaces)
• Base URL if your team requires a custom endpoint
• Local Ollama is running and the model is downloaded