Corvana documentation.

What Aria is

Aria is a named, autonomous AI consultant. Not a chatbot, not a pipeline of worker agents — one model call per turn to claude-opus-4-7, conditioned on a system prompt composed server-side from eight composable layers of context. Identity, methodology, and output conventions are cached globally; company, user, and engagement state are composed per request from Postgres.

The conversational surface is at /chat (authenticated, OAuth session required). The public-facing demo at /demo is a separate code path with hardcoded context and no persistence — live and demo contexts never mix.

How an engagement works

An engagement begins at sign-in. A three-step onboarding captures company, role, and goal — nothing more; the rest is captured in conversation with Aria, week after week. Every turn Aria updates her view of the engagement: open hypotheses, unresolved unknowns, findings to date. That memory survives between sessions.

1. Week 1 — Orientation. Aria asks calibrating questions to establish operating model, decision rhythms, and the shape of the real problem.
2. Weeks 2–4 — Hypothesis work. She forms falsifiable hypotheses, asks for evidence, revises confidence, and promotes findings when the evidence lands.
3. Week 5+ — Synthesis and recommendation. Pyramid-structured conclusions, supporting evidence, and explicit uncertainty where it remains.

Explore the docs

Need help?

Reach out at hello@corvanahq.com for scoping, integration review, data handling questions, or procurement documentation.