Summarise PDF attachments arriving in Gmail or Outlook

import base64
import io
from pypdf import PdfReader
from google.oauth2.credentials import Credentials
from googleapiclient.discovery import build
from clickruntime import on_trigger, report_fetched, http, credentials, ai, state, notify

api = lambda name, version: build(name, version, credentials=Credentials(credentials.connection("google", "work")), cache_discovery=False)

def graph(method, path, **kw):
    r = http.request(method, "https://graph.microsoft.com/v1.0" + path, headers={"Authorization": f"Bearer {credentials.connection('microsoft', 'work')}"}, **kw)
    return r.raise_for_status() or (r.json() if r.content else None)

pdf_text = lambda b64: " ".join(p.extract_text() or "" for p in PdfReader(io.BytesIO(base64.b64decode(b64))).pages)[:12000]

@on_trigger
def run(event):
    gmail, seen, since = api("gmail", "v1").users().messages(), set(state.get("seen") or []), state.get("since") or "2026-01-01T00:00:00Z"
    pdfs = []
    for ref in gmail.list(userId="me", q="has:attachment filename:pdf newer_than:1d").execute().get("messages", []):
        if ref["id"] in seen:
            continue
        m = gmail.get(userId="me", id=ref["id"]).execute()
        subject = next(h["value"] for h in m["payload"]["headers"] if h["name"] == "Subject")
        for p in m["payload"].get("parts", []):
            if p.get("filename", "").lower().endswith(".pdf"):
                data = gmail.attachments().get(userId="me", messageId=ref["id"], id=p["body"]["attachmentId"]).execute()["data"]
                pdfs.append((f"Gmail: {subject}", base64.b64encode(base64.urlsafe_b64decode(data)).decode()))
        seen.add(ref["id"])
    out = graph("GET", "/me/messages", params={"$filter": f"hasAttachments eq true and receivedDateTime gt {since}", "$select": "id,subject,receivedDateTime"})["value"]
    for m in out:
        pdfs += [(f"Outlook: {m['subject']}", a["contentBytes"]) for a in graph("GET", f"/me/messages/{m['id']}/attachments")["value"] if a.get("contentType") == "application/pdf"]
    report_fetched(len(pdfs))
    if not pdfs:
        return {"summary": "No new PDFs"}
    parts = [f"{label}\n" + ai.generate(pdf_text(b64), system="Summarise this document in 5 bullet points, noting any deadline.", max_output_tokens=300).text for label, b64 in pdfs]
    notify.email_me(f"{len(pdfs)} PDFs summarised", "\n\n".join(parts))
    state.set("seen", sorted(seen)[-3000:])
    state.set("since", max([since] + [m["receivedDateTime"] for m in out]))
    return {"pdfs": len(pdfs)}