Your team's Google Drive has more to say than anyone's reading. Liminal surfaces what's rising, what's stale, what's drifting, and what needs attention — without anyone having to go looking.
Team knowledge lives in Google Drive, but Drive was built for storage — not insight. You can't easily answer: What docs is the team actually relying on? Which specs are outdated but still being linked? Are two teams using different words for the same thing? You find out the hard way, when someone acts on stale information or two workstreams collide.
Liminal connects your documents into a graph and watches how that graph changes over time. It answers the questions Drive doesn't.
- Team leads and PMs who want a weekly pulse on what the team is reading, what's changed, and what to flag — without wading through Drive activity logs.
- Ops leads, TPMs, and Chiefs of Staff who need a durable queue of knowledge health issues to work through: stale hubs, orphaned notes, duplicate specs, and terminology that's diverged across teams.
| Overview with Team Digest | Doc Audit queue |
|---|---|
 |
 |
| Doc Audit detail panel | Document graph |
|---|---|
 |
 |
The Overview page gives anyone on the team an at-a-glance picture of what's happening in the knowledge base this week:
- Worth reading — the docs gaining the most activity, personalized by role
- What changed — which docs are newly rising and why
- Decisions and follow-ups — a plain-language summary of what needs attention, with direct links into the Doc Audit for action
Signal chips at the top show active finding counts by type — click any chip to jump straight to that filter in the Doc Audit.
A durable task queue for whoever owns knowledge health on the team. Each finding has a severity level, a suggested action, and a full review panel:
- Stale hubs — documents with many inbound links that have gone quiet; high blast radius if they're wrong
- Rising docs — newly active documents that may need freshness or owner review
- Went quiet — docs that lost activity after a period of engagement
- Terminology drift — linked document pairs using different words for the same concepts, detected automatically without an LLM
- Possible duplicates — document pairs with overlapping content and similar titles
- Orphaned meeting notes — meeting docs that were never linked into the broader knowledge base
Click any row to open the full review panel: suggested action, status and disposition tracking, assignee, follow-up date, document metrics, and activity timeline — all without leaving the page.
An interactive graph of every document and the links between them. Filter by node type, focus on a single document to see its neighborhood, or enable alignment mode to spot terminology divergence visually. Hub documents (high inbound-link count) are highlighted — these are the load-bearing docs in your knowledge base.
Trigger a fresh index from Settings. Watch active phase, current document, and progress live without leaving the web app. Supports personal Drive and Shared Drives; each Shared Drive gets its own isolated workspace in the selector.
This project is an extension of the Living Documents concept from Agentics Beyond Code: documents aren't static artifacts, they're living signals of organizational activity. By connecting them into a graph and tracking how that graph changes over time, you get a dynamic picture of what the org knows, values, and is actively working on.
Terminology drift detection is inspired by the Getting the Words Right playbook on representation drift in organizations. See docs/ontology-roadmap.md for the phased plan toward LLM-enhanced semantic alignment.
drive-analytics/
├── README.md # This file
├── SPEC.md # Technical specification and data model
├── FUTURE_IDEAS.md # Tabled ideas for later consideration
├── docs/
│ ├── google-setup.md # How to configure Google APIs and OAuth
│ ├── ontology-roadmap.md # Phased plan for terminology/semantic alignment
│ └── screenshots/ # Current web app screenshots
├── src/
│ ├── api.py # FastAPI app and web/API routes
│ ├── auth.py # OAuth flow and Google service clients
│ ├── indexer.py # Drive crawling, activity, and link extraction
│ ├── graph.py # Graph construction and analysis
│ ├── analytics.py # Rising/stale/hub detection
│ ├── ontology.py # Term extraction and alignment scoring
│ └── operations.py # Team Digest and Doc Audit state
├── web/
│ ├── index.html # Local web app shell
│ ├── app.js # Views, drawers, and interactions
│ └── styles.css # Google Drive-inspired interface
├── tests/ # API, indexing, graph, and analytics tests
└── data/ # Local graph storage (gitignored)
Follow docs/google-setup.md to create a Google Cloud project, enable the required APIs, and download credentials.json.
pip3 install -r requirements.txtcp config.example.json config.json
# Edit config.json to add domain-specific settings for your orgOptional environment variables:
export OPENAI_API_KEY=... # enables brief polishing
export DRIVE_ANALYTICS_WRITE_TOKEN=... # protects FastAPI POST/PATCH endpoints
export DRIVE_ANALYTICS_CORS_ORIGINS=http://localhost:3000python3 src/auth.py --verifypython3 src/indexer.py --days 90 --expandTo index a particular Shared Drive into its own selectable workspace:
python3 src/indexer.py --shared-drive "https://drive.google.com/drive/folders/0AP6VPalWOTU-Uk9PVA" --days 365 --expanduvicorn src.api:app --reloadOpen http://127.0.0.1:8000. The app opens in Demo product team mode by default. Switch to Live Drive or an indexed Shared Drive from the workspace selector.
Reset or prepare the demo dataset:
python3 src/demo_data.pypython3 -m pytestInteractive API documentation: http://127.0.0.1:8000/docs
When DRIVE_ANALYTICS_WRITE_TOKEN is configured, operational writes require it in the X-Admin-Token header.
- Local (default) — FastAPI serves the web app, API, and background indexer on your machine. Good for prototyping and personal Drive testing.
- Google Cloud (production path) — Cloud Run for the web app, API, and indexer; Cloud Scheduler for periodic scans; Firestore or BigQuery for graph storage. Designed to migrate cleanly from local.
See SPEC.md for the full technical plan.