See it in action

Follow the five steps below in order. Each step opens the workspace page it references with sample inputs pre-filled.

1. 1

   Provision a demo tenant

   Run this once on the host running the API — it creates a fresh tenant, an admin key, four starter spaces, and prints the credentials.

   # From the repo root
   ANVION_API_URL=http://localhost:8000 python scripts/seed_dev.py

   Save the printed tenant_id and api_key, then open Settings and paste them.
2. 2

   Ingest a Sharia standard (ACE)

   Go to Documents and upload scripts/fixtures/sample-murabaha.txt into the AAOIFI Standards space. Watch the status flip pending → ingesting → ready — that's the full fetch → extract → chunk → embed → index pipeline, live.
3. 3

   Draft a citation-grounded dossier (ACE)

   Open Compose dossier (the ?demo=1 query pre-fills a Murabaha auto-financing product summary). Pick the AAOIFI Standards space, click Draft dossier (stream).
   Watch — citations appear first, then the answer streams token by token. Every factual sentence ends with a [[c:N]] tag mapping to a citation card. Ungrounded generation is architecturally refused.
4. 4

   Run an approval workflow (Wasla)

   Open Workflow runs — ?demo=1 pre-fills the sharia_dossier workflow with the same product summary. Click Start run. The workflow drafts a dossier (calls ACE), then pauses on an approval node.
   Go to Approvals, add a decision note, click Approve. The workflow resumes, writes the ruling back to ACE as a semantic memory, and finishes.
5. 5

   Prove the Trust Ledger (Wasla)

   Return to the run detail page (link visible after approval). Every step of that workflow — the LLM call, the human approval, the memory write — is a row in an append-only, hash-chained Trust Ledger. The badge at the top says chain verified. Try changing any step's output field in the database directly, then reload — the badge flips to chain BROKEN and pinpoints the tampered row.
   That's what public accountability actually looks like.