The Atlas Donna's documentation, bound to its code
147 documents

First hour with Donna

Orientation: what Donna is, the one architectural idea everything follows from, and where to go next. Four short reads plus the docs map.

What's inside

  • Assistant chat with streaming answers and character-verified citation pills — hover for the source quote, click to open the document panel jumped to the exact cited passage. A per-turn receipts drawer shows every retrieval, inference, and skill event behind an answer, including whether anonymization was applied.
  • Matters (projects) — scope chats to a matter with files, linked knowledge bases, attached skills, and free-form context; privileged matters enforce a minimum inference tier in the composer.
  • Knowledge bases — create, link, upload; documents auto-ingest for retrieval (RAG) with live status, hybrid-search tuning, and per-file download.
  • Workflows hub — four kinds of reuse:
    • Skills: reusable instruction blocks with typed inputs; author your own or fork built-ins, attach them per-message (slash aliases supported).
    • Playbooks: negotiation positions applied to a contract → verdict scorecard + consolidated redline view; generate a draft playbook from your own documents.
    • Prompts: saved snippets inserted at the cursor.
    • Automations: runs Donna executes on its own — run-now, cron schedules, and KB-arrival watches — each leaving a transparency receipt (phases, tool calls, cost, terminal reason) plus its results (findings, proposed memories, and recurring precedents), with a notifications inbox. Opted-in runs also produce document-grade artifacts — memos the run writes into its target knowledge base, openable inline or downloadable straight from the receipt.
  • Tabular review — the same questions across many documents → a cited, confidence-scored grid; per-column model-tier floors and ensemble verification; Excel/CSV export.
  • Redlines — consolidated change-set view of a playbook run with severity-ordered margin notes.
  • Settings — account & security, data export / scheduled deletion, preferences (incl. ambient trust pills), a read-only trust matrix, and model management: per-category routing, installed local (Ollama) models, and bring-your-own provider keys (admin, hot-applied, write-only).
  • Prompt enhance on every composer, file attach in chat, and an in-app guide at /about — including interactive playgrounds explaining how the LQ-AI engine works.