Local-LLM mode (Pro)

On the Pro ($24/mo) tier, FastBI narrates over your locally-computed facts using a model running on your machine. The deterministic engine already runs locally in every tier; local-LLM mode moves the narration there too, so your data never leaves the box.

How it works

It is an endpoint swap — the same client code. Point FastBI at any Anthropic Messages API–compatible endpoint with one environment variable:

export FASTBI_BASE_URL="http://localhost:11434"   # your local runtime

FastBI computes the measures and evidence locally, sends only that compact, sanitized summary to your local model, and the model writes the narration. Gated by tier, not by surface — local-LLM unlocks everywhere at $24.

Supported runtimes

  • Ollama ≥ 0.14 — ollama serve, then set FASTBI_BASE_URL to its address.
  • LM Studio — start the local server and point FASTBI_BASE_URL at it.

Requirements

  • Minimum context window: 32K tokens. FastBI’s evidence packs are compact, but headroom keeps long dashboards coherent.
  • A mid-size instruct model. Recommended starting points:
    • Llama 3.1 8B Instruct (or larger)
    • Qwen2.5 14B Instruct
    • Mistral Small / Ministral 8B Instruct

Privacy

In local-LLM mode nothing is sent to the FastBI backend for narration — the computed facts and the model both stay on your machine. This is the mode for regulated or sensitive data.

See pricing → · Back to docs