PharmAgent

PharmAgent is a clinical note assistant for pharmacists. It helps turn a patient scenario into a concise French intervention note that can be reviewed and copied into the pharmacy file.

The tool is designed for practical pharmacotherapy questions: hypertension, diabetes, anticoagulation, asthma, dyslipidemia, hypothyroidism, migraine, COPD/MPOC, ADHD/TDAH, and other supported domains as the knowledge base grows. PharmAgent retrieves evidence from local professional documents before drafting; if it cannot find relevant local knowledge, it refuses to provide a clinical recommendation.

PharmAgent supports documentation. It does not replace the pharmacist's clinical judgment.

Why It Exists

Pharmacists often need to document a clear intervention quickly while still grounding the recommendation in clinical references. The goal of PharmAgent is to reduce drafting friction without turning the answer into a generic chatbot response.

The expected output is always structured for charting:

• Collecte de données — only the facts provided in the case.
• Analyse — targeted clinical reasoning for this patient.
• Intervention et recommandations — the recommended plan, monitoring, and follow-up.
• Sources — the main references found in the retrieved documents.

How It Works

1. The pharmacist enters a patient scenario and may attach supporting documents.
2. PharmAgent retrieves relevant excerpts from the local clinical knowledge base.
3. If the retrieved evidence is sufficient, an OpenClaw-authenticated model drafts the note.
4. If the evidence is insufficient, PharmAgent returns a refusal note instead of guessing.

Clinical Knowledge Base

The current demo corpus includes parsed French clinical toolboxes covering:

• anticoagulation
• asthme
• dyslipidémie
• diabète
• hypertension artérielle
• hypothyroïdie
• migraine
• MPOC
• TDAH

The PDF-derived content is converted into JSON-LD/RDF source excerpts. Generated clinical source entries are marked for human review before any verified-memory promotion.

Local Development

npm run check
PORT=3088 npm start

The default generation path expects a local OpenClaw installation already authenticated with OAuth.

Knowledge Pipeline

Install Python dependencies:

python3 -m venv kg-pipeline/.venv
kg-pipeline/.venv/bin/pip install -r kg-pipeline/requirements.txt

Validate JSON-LD:

kg-pipeline/.venv/bin/python kg-pipeline/scripts/validate_jsonld.py

Build the keyword retrieval index:

kg-pipeline/.venv/bin/python kg-pipeline/scripts/build_embeddings.py --keyword-only

Stage reviewed knowledge to the local graph memory layer:

DKG_AUTH_TOKEN=<token> node kg-pipeline/scripts/stage_working_memory.mjs

Run readiness checks:

node bounty/verify-readiness.mjs

Safety and Privacy Notes

• PharmAgent is for pharmacist documentation support, not autonomous prescribing.
• Recommendations are limited by the quality and coverage of the local knowledge base.
• Unsupported cases should return a clear refusal rather than an LLM-only answer.
• Do not commit local .dkg, .openclaw, .env, API keys, OAuth tokens, embeddings, or runtime logs.