Build a privilege log

# Decosa Privilege review and privilege log: use the hosted API

You are wiring Decosa's privilege reviewer into this project. It takes emails and memos from a production set and the
client's lawyers, and returns a typed call per document (attorney-client, work product, both, not privileged, or needs
attorney review) with the lawyers involved, a reason checked against the document and a calibrated probability. It
drafts privilege-log entries that pass a leak check, gives exact copies the same call, flags outsiders who could waive
privilege, and ends with a signed record. Each model call has its own signed receipt. Use only what is listed below. If
you need something else, stop and ask me.

- Base URL: `https://api.decosa.ai`
- Health check: `GET https://api.decosa.ai/healthz`.
- The hosted API is for the public and synthetic sample sets and for testing your integration. Privileged documents
  are the most sensitive a client has, and sending them to a third party's service can put privilege at risk: for real
  matters use the self-host prompt instead. Never send client documents here.
- It drafts; a lawyer decides. Show every call as a draft for attorney review, never as a decision.

## Auth: API key (or a demo session)
1. Preferred: an API key (`dk_…`) from "Get an API key" on the tool page. Keep it in an environment variable,
   `DECOSA_API_KEY`, never in code. Send `Authorization: Bearer $DECOSA_API_KEY`.
2. Without a key: `POST https://api.decosa.ai/demo/session` with `{"vertical": "privilege-log"}` returns `{"token", "expires_at", "budget"}`.
   a limited number of sessions per network per hour (the current limits are in `demo_sessions` of GET /healthz); each session has a token allowance (its `budget`) (about 20 documents). Over a limit: HTTP 429 with `Retry-After`.
3. One review at a time per demo token (409 otherwise).

## Endpoints
- `POST /privilege/review` (token). Body:
  `{"documents": [{"id", "date"?, "from"?, "to"?: [...], "cc"?: [...], "bcc"?, "subject"?, "body", "thread_id"?, "family_id"?, "parent_id"?, "bates"?}], "people"?: {"attorneys": [{"name"?, "email"?, "role": "in-house"|"outside", "firm"?}], "client_domains": [...], "counsel_domains": [...], "agents"?: [{"email"}]}, "samples"?: 4, "title"?: "...", "stream"?: true}`
  or `{"sample": "harborline-dispute"}` to run a sample set.
  - Limits: 25 documents per request, 20,000 characters each, 150,000 in all. `samples`: 4 (default), 2, or 0 (one call
    per question, stated confidence only).
  - With `"stream": true` (or `Accept: text/event-stream`) it streams `ready` (people map, flags, groups), a `receipt`
    after each model call, a `document` event per document (replace by `id`), then `consistency`, `log`, `report`,
    `done` and `budget`. With `"stream": false`: one JSON object `{run_id, totals, documents, people, groups, consistency, log, report, budget}`.
  - Each document: `{id, final, model_call, probability, reasons, legal_advice, litigation, flags, basis: {lawyers, reason, grounding}, log?: {description, fallback, attempts}, copy_of?}`.
    `final` is `needs_review` whenever `reasons` is not empty; treat it as withheld until a lawyer decides.
- `POST /privilege/runs/{run_id}/decisions` `{"decisions": [{"doc", "decision": "confirm"|"change", "call"?, "reviewer", "note"?}]}`: the lawyer's calls.
- `GET /privilege/runs/{run_id}/export?format=csv` (the log), `md` (review memo), `record` (signed), `ledger` (signed, with the decisions).
- `POST /record/verify` (no token) `{"record": {...}}` → `{ok, summary, bad}`.
- `GET /privilege/info`, `GET /privilege/samples`, `GET /privilege/samples/{id}`, `GET /attest/signing-key` (no token).

## Example: review a batch and write the draft log (Python, `pip install httpx`)
```python
import httpx, os, pathlib
API = "https://api.decosa.ai"
H = {"Authorization": f"Bearer {os.environ['DECOSA_API_KEY']}"}
s = httpx.get(f"{API}/privilege/samples/harborline-dispute").json()        # your own batch has the same shape
r = httpx.post(f"{API}/privilege/review", headers=H, timeout=900,
               json={"documents": s["documents"], "people": s["people"], "stream": False})
r.raise_for_status()
run = r.json()
for d in run["documents"]:
    print(d["id"], d["final"], d.get("probability"), "; ".join(d["reasons"]))
pathlib.Path("privilege-log.csv").write_text(httpx.get(f"{API}{run['export']['csv']}", headers=H).text)
```

## Honest limits
- The calls are drafts from an open model and code rules. On labelled public email they miss some privileged documents
  and send many to review; see the measured rates on the Stack tab before relying on them.
- It reads what you send: attachments must be sent as their own documents, and the people map is only as good as the
  lawyer list and domains you give it.

On request. The container images and the compose file aren’t public yet. Ask for self-host access and Decosa sends the registry (DECOSA_REGISTRY) and the compose file’s URL (DECOSA_COMPOSE_URL) these steps use. They are the steps we tested end to end on a fresh machine.

# Decosa Privilege review and privilege log: run it yourself (containers)

You are setting up the Decosa privilege reviewer on this machine, so privileged documents never leave it. It makes a
typed privilege call per email or memo, drafts privilege-log entries that pass a leak check, reconciles duplicates and
threads, and signs a record. Nothing is sent to Decosa's hosted API.

Status: the container images (${DECOSA_REGISTRY}/decosa-*) and the compose file are on request while self-host is in early access (not on a public registry yet): ask at https://decosa.ai/contact?topic=self-host, and Decosa sends the registry as DECOSA_REGISTRY, the compose file URL as DECOSA_COMPOSE_URL, and pull access. If a pull fails with
"not found", "denied" or "unauthorized", stop and tell me. Do not substitute other images.

Ask me before any command that needs sudo, and show me the command first.

## Step 0: set up with a coding agent, rehearse on mock data, then go private

This prompt is for a coding agent running on the machine that will host the service. We recommend Claude Code with
Claude Opus 5.5; any capable coding agent works. Work in this order:

1. Set up on mock data only. During the whole setup you (the agent) work with the synthetic sample bundle below and
   nothing else. Do not ask me for real data, and do not open, read, list or copy files that hold real data, even to
   "test with something realistic".
2. Rehearse. When the steps below are done and the service is healthy, fetch the mock-data bundle for this tool,
   https://decosa.ai/samples/privilege-log.zip (3 KB, 11 checks, synthetic or openly licensed: see `licence` in expected.json),
   show me what is in it, and run the rehearsal against the local API:
   `docker compose exec api python scripts/rehearse.py privilege-log` (the api image carries the same bundle under /app/rehearsal/privilege-log/;
   with no key set, the script asks the local API for a short demo token). From a decosa-api checkout instead:
   `python scripts/rehearse.py privilege-log --bundle privilege-log.zip --base-url http://127.0.0.1:<PORT>`.
   It sends the mock inputs to the local API and prints PASS or FAIL for each expected property (for example: "the general counsel's legal advice (HL-002) is withheld as privileged or sent to review", "the ops update with a lawyer only copied (HL-021) is produced", "HL-021 is flagged: a lawyer only copied"). Show me
   the full output. Every check must pass. If one fails, fix the install and run it again; never edit `expected.json`
   to make a check pass.
3. Stop there. Once the rehearsal passes, tell me, and I will run my own data against the local API myself, on this
   machine.

For the person running this: a coding agent that runs in the cloud sees everything in its context, including files it
reads, command output and anything pasted into the chat. Keep real data out of the chat and out of anything the agent
can read. Switch to your own data only after the rehearsal has passed and the agent's work is done.

## Steps
1. Docker: if `docker compose version` fails, install Docker Engine and the compose plugin using Docker's official
   instructions for this distribution (docs.docker.com/engine/install). Also install the NVIDIA container toolkit and
   check `docker run --rm --gpus all nvidia/cuda:12.8.0-base-ubuntu24.04 nvidia-smi`.
2. Fetch the compose file:
   `mkdir -p ~/decosa && cd ~/decosa && curl -fsSL "${DECOSA_COMPOSE_URL}" -o compose.yaml`
   Read it. Keep the `llm` service (Qwen3.8-27B on vLLM) and the `api` service. For the `api` service set
   `DECOSA_LLM_ROUTE=direct`, `DECOSA_LLM_URL=http://llm:8000/v1`, `DECOSA_LLM_MODEL=qwen3.8-27b`,
   `DECOSA_PRIVILEGE_MAX_DOCS=200` and bind every port to 127.0.0.1. Nothing in this tool needs the internet after
   the weights are downloaded.
3. Pull and start: `docker compose pull && docker compose up -d`. Wait for the `llm` health check (the first start
   downloads about 20 GB of weights).
4. Check: `curl -fsS http://127.0.0.1:<PORT>/privilege/info` shows `"method": "logprobs (one call per question)"`;
   `GET /attest/signing-key` shows this box's public key. Show me the key: it is what others pin to verify my records.
5. Smoke test: get a token with `POST /demo/session {"vertical":"privilege-log"}` and run
   `POST /privilege/review {"sample": "harborline-dispute", "stream": false}`. Expect HL-004 to be a copy of HL-002 with
   the same call, HL-003 (advice forwarded to an outside consultant) in `needs_review` with a third-party reason, and
   HL-012 and HL-021 not privileged. Then `POST /record/verify` with `report.record`: `ok` must be true.
6. Report back: the public key and key id, the totals, the log entries and how long the run took.

Off, and keep it off on a box that holds privileged documents: joining serves other people's requests on this GPU.
If I ask for it later, on a separate machine, follow the Provide page instead of improvising.

No NVIDIA GPU? This tool also runs entirely on an Apple Silicon Mac (MLX, 32 GB of unified memory or more): use https://decosa.ai/prompts/privilege-log-mac.md instead.