Endpoints
POST /v1/extract
multipart: files, user_id, optional industry, email_subject, email_body, auto_memory.
POST /v1/remember
json: {user_id, text, source?, tags?}
GET /v1/recall
query: user_id, q, limit. Keyword-scored memory search.
POST /v1/chat
json: {user_id, message}. Memory-grounded Q&A (Sonnet).
GET /v1/history
query: user_id. Past extractions (task_id, trust, routing).
POST /v1/feedback
json: {user_id, task_id, kind, payload?}. kind: accept / reject / correct / note.
SDK
from monceai import Extraction, Outlook
ex = Extraction("quote.pdf", user_id="7a3f9b2c", industry="glass")
ex.lines # structured rows
ex.trust # {"score": 82, "routing": "auto_accept"}
ex.insights # Haiku-distilled memory bullets
ol = Outlook(user_id="7a3f9b2c", auto_memory=True)
ol.extract_email(attachments=[pdf_bytes], subject="Devis VIP", body="comme d'hab")
ol.recall("what does this user usually do with VIP files?")
ol.remember("client always wants 44.2 rTherm as intercalaire")
Outlook add-in
/addin/ — installable via manifest.xml, posts attachments to /v1/extract.
Interactive
/ui — drop a PDF, watch memory + insights unfold · /addin/ — Outlook taskpane
Docs
/dashboard — live ops view · /paper — concept and reflex loop · /architecture — diagram, endpoints, layout · /economics — cost and capacity