Quill

Quill is the team's Notetaker: part archivist, part investigative editor, part continuity engine. Give him the raw record and he files the durable delta. Ask what changed or why, and he reconstructs the thread from evidence. His standard is exacting: one clean canonical record, with its provenance intact.

Who Quill is#

Quill is a Notetaker built for an engineering manager's reality — a firehose of meetings, decisions, names and deadlines that would otherwise live in your head or scatter across a dozen docs. His job is to capture, reconcile, trace, and recall so the company record stays more useful than folklore.

He's concise and reliable, and he has a few strong habits:

• He classifies the intent first. A durable statement or meeting dump is filed; a question, document lookup, or rewrite request is answered without silently becoming memory.
• He writes fields that stand alone. Summaries and date labels show up on boards with no surrounding context, so each reads like a headline a stranger could act on.
• He keeps one memory per item. When something changes, he updates the existing note rather than creating a second one — and preserves the history ("was Nov 15, moved to Dec 1").
• He's honest about recall. If he doesn't actually have something on file, he says so instead of inventing it.

Giving Quill notes#

There's nothing to structure. Dump whatever you've got and Quill pulls out what matters — the decision, the date, who's involved, what needs doing. For durable notes he files first and confirms the result in a line: "Filed — Aurora review due Nov 15; Sarah is the lead." If your message is actually a question or a request to inspect a document, he treats it that way instead of silently storing it.

Capture anything#

Quill accepts any input format — meeting notes, pasted docs, quick thoughts, bullet points — and extracts the structure himself. From one dump he'll separate out decisions, deadlines (with dates in a real calendar format), action items (owner, the ask, the counterparty, and why), and the people and systems/projects involved, tagging each so it surfaces in the right views later. Dates and names are pulled automatically.

One item, no accidental duplicates#

The discipline that makes the memory trustworthy: each deadline, task, decision or fact should exist as exactly one canonical memory. Every new save runs duplicate checks before persistence: Quill compares the wording and dates, then checks semantic similarity. A likely restatement is stopped before it can become a second record, and the existing matches come back so he can update the right item instead. Writing a separate record requires an explicit override. When work is finished he marks it done (so it drops off deadline boards but stays searchable as history), and he can still find and merge older accidental duplicates on request.

He writes every field to stand alone on a board:

Saving links#

When a message is essentially a URL or a bare domain, Quill files it onto your Links board — it auto-fetches the page title and categorizes it, and you can add a note on why it matters. You don't need to type https; a bare host is enough.

Projects#

When several dated items clearly belong to one initiative — or you tell him something is (or isn't) a project — Quill organizes them on your timeline board. Naming a project lands it as a lane you can hang items and deadlines from, so a stream of notes becomes a tracked initiative.

Recall & search#

Ask Quill about the past and he searches memory before answering — "what did we decide about the ledger", "what's due this week", "what do we know about the budget freeze". He replies directly, leading with the note: "From your notes: …". Dated items also roll up into upcoming-deadline and overdue views.

Recall by meaning#

Quill's recall goes well beyond keyword search. He can match your question against the meaning of every note, so a memory surfaces even when it shares no words with what you asked. When you want the full picture of a topic he pulls a diverse set that covers distinct facts rather than paraphrases of the single best hit, and his deepest recall follows connections through a shared person or project to assemble an answer from evidence that isn't worded like the query at all.

Crucially, he checks a calibrated "do I actually have this?" signal before answering — so low confidence means he tells you he doesn't have it, instead of guessing.

Trace the thread#

For a topic that has evolved across meetings and notes, Quill's trace_thread assembles one evidence-backed brief: the chronology, recorded decisions, open actions, dated commitments, people and tags involved, and the bridge memories that connect the pieces. He calls out gaps and reports recall confidence rather than filling in a missing owner, date, or rationale.

History & what changed#

Quill can inspect the record as a record, not just quote its latest version. memory_history shows the revision trail for one memory. recall_as_of searches what the local event history can reconstruct at a chosen date. what_changed compares two dates and separates added, removed, and changed items, including the fields that moved. If you ask for a date older than the retained history can support, Quill warns about that horizon instead of presenting a false snapshot.

Your live lists#

Quill maintains ready-made views over everything he's captured:

• Open action items — the team's live to-do list, newest first.
• Decisions log — every decision on record, for a "what did we decide about X" review.
• Overdue items — open items whose date has passed, good candidates to chase or close.
• People & tag indexes — everyone who appears across the notes (most-mentioned first) and every tag in use, plus the "everything about Dana" view.
• Timeline & digest — a chronological walk through recent notes, and a "state of the notes" digest: totals, category mix, open items, overdue count, upcoming dates, top people and tags.

The knowledge graph#

Every note lives in a typed knowledge graph — linked to the people, projects and other notes it relates to, with the relationship named (an action that mitigates a risk, a note that follows another). That's what lets Quill answer "how are these connected" with a real chain, and lets Sage trace connections across your whole world.

Beyond the links you imply, Quill also notices ties nobody drew — memories that keep getting recalled together. His "surprising connections" sweep surfaces the genuine huh, these two keep coming up together pairs, and he can roll the whole corpus up into its natural themes.

Consolidation & health#

Like a memory that sleeps, Quill runs a consolidation pass — abstracting recurring episodes into durable schemas and pruning weak, noisy links. It normally runs overnight, but you can trigger it after a big import. He can also tell you which memories are going cold (at risk of slipping out of recall) so you can reinforce, update, or prune the ones that are truly obsolete.

Audit the archive#

Quill's audit_memory runs a read-only health check over the corpus and returns a transparent score with the deductions behind it. It samples likely duplicate clusters, overdue commitments, action items with no owner or date, orphaned links, stale memories, and gaps in semantic-vector coverage. The audit diagnoses; it does not silently rewrite or delete the archive.

Google Drive#

With Google connected, Quill reaches your Drive — read-only. He searches by keyword across file names and full text, browses folders like you would, reads a single file (Docs come back as text, Sheets as CSV, Word and PDFs are extracted), and can download a file to your machine when you need it locally.

OneDrive & SharePoint#

The Microsoft 365 mirror of his Drive powers, on one "Sign in with Microsoft": Quill searches across your OneDrive and every SharePoint site you can read — file names and content — and reads a result's text (Word documents converted, PDFs extracted, text formats read directly). Read-only, as you: he sees exactly what you can see, nothing tenant-wide, and there's no upload path.

Company knowledge sources#

Quill can search a connected Vertex AI Search datastore for the company corpus, or go to the source directly: Confluence, Notion, Box, and Atlassian Rovo, alongside the native Google Drive and Microsoft sources above. That is how "what does the company already know about X" gets answered from the real material rather than guesswork. Retrieval is read-only by default; when you deliberately ask to edit or publish a Confluence page, Quill uses a human-controlled editor or confirmation handoff rather than silently changing the wiki.

Calendar & deadlines#

Quill separates two kinds of time: the deadlines you saved to memory (the migration date, "follow up with Maya") and your real calendar synced from your Mac (actual meetings). He reads both, so "what's due this week" pulls from your notes and "am I free Thursday" checks your lived schedule.

Watches, studios & lessons#

Quill rounds out the team's shared toolkit: put a data pull on the Night Shift watchlist so it's re-checked and you're alerted on a material change; jump into the right studio with a one-click chip; and correct him — "always keep the ticket key", "file people under their full name" — and he records it as a durable lesson that changes how the team works from then on.