Grounded reasoning
Every answer cites its evidence. Ungrounded claims are refused, not smoothed over.
Enterprise agentic platform · macOS · local-first
Turn a firehose into a signal.
Inference Engine ingests, normalizes, and reasons over content from your connected sources. It grounds every answer in a cited span, argues with itself before speaking, and refuses to make things up.
Apple Silicon · macOS 12+ · signed & notarized · ~250 MB
What it is
Deep-think mode
Four-pass loop: draft → self-critique → evidence repair → counter-argument. Toggleable per request.
Local-first
Fine-tuned Llama-3.1 runs on your Mac. Frontier calls (OpenAI / Anthropic) are opt-in and go out through your keys, not ours.
Daily digest
Auto-surfaces signals — bugs, complaints, security concerns, praise, competitor moves — from your connected sources.
Persona memory
Learns your style, tone, and priorities over time so responses fit your voice.
Auditablephase6
Every retry, every fallback, every refusal is a structured log line. No silent swallowing.
Requirements
- Apple Silicon (M1 / M2 / M3 / M4)
- macOS 12.0 or later
- At least 16 GB of unified memory (32 GB+ recommended)
- ~2 GB free disk (backend + optional model download)
- Optional: OpenAI or Anthropic API key for frontier reasoning calls