Skip to content

Docs / AI providers

AI providers.

Ignia separates the AI choice from the app. Three modes, pickable per skill: managed, your own API key, or a model running entirely on your machine. Each commits to a different trust and billing contract.

Three modes.

Managed

Ignia-hosted AI.

A Course Pack covers the AI calls for one course. No API key to manage, no per-call billing on your side. We pay the upstream API costs out of the pack price.

Trust contract: your prompts pass through our infrastructure on the way to the model provider.

BYOK

Your API key.

Plug in a key from Anthropic, Google, OpenAI, or any OpenAI-compatible endpoint. The app talks to the provider directly. Ignia is never in the payment loop.

Trust contract: prompts go from your device to the provider you chose. The key is stored in your OS keychain.

Local

Fully on your machine.

Ollama or any local OpenAI-compatible server (llama.cpp, vLLM, LM Studio). No network call leaves the box. Course quality depends on the model you can run locally.

Trust contract: nothing leaves your device. Best for sensitive subjects.

Per-skill override.

The provider you pick at install is a default. Each skill can override it. A typical setup: managed AI for a fast-moving technical skill where quality matters; local Ollama for case notes or anything you do not want leaving the machine.

To override, open Settings on the skill page and pick a different provider from the dropdown. The override sticks until you change it.

Supported providers.

The providers Ignia recognizes today.

  • Anthropic Claude (Opus, Sonnet, Haiku families)

    Strong on long-form writing judges. Available both ways.

    Managed / BYOK

  • Google Gemini

    BYOK only at this stage. Strong context window.

    BYOK

  • OpenAI

    BYOK. Standard chat-completions endpoint.

    BYOK

  • OpenAI-compatible endpoint

    Any server speaking the OpenAI chat-completions API. Use this for vLLM, Together, Fireworks, llama.cpp server, LM Studio, and similar.

    BYOK

  • Ollama

    Local model runner. Point Ignia at the Ollama daemon, pick a model. All inference on your machine.

    Local