Compiled knowledge base
Your company’s memory should compound.
Hosted LLM wikis that compile email and Slack into a living, linked knowledge base — every day. Export anytime.
The problem
Context dies. Chat resets. Notes rot.
Teams live in Slack and email. AI chat re-answers from scratch. DIY wikis work until security and daily ops don’t.
Threads expire
Decisions hide in channels. New hires inherit oral history.
RAG doesn’t compound
Every query rediscovers. Nothing stays structured for next week.
DIY is fragile
Agent folders + home servers fail on isolation, connectors, and cadence.
The model
Three layers. Not another chatbot.
Same architecture as a disciplined LLM wiki — hosted, secured, and fed every day.
Immutable sources
Email, Slack, uploads — hashed, provenance kept, never silently rewritten.
Linked pages
Entities, concepts, comparisons. Cross-links, index, and an append-only log.
Quality rules
Domain, tags, thresholds, contradiction policy — the agent stays disciplined.
How it works
Connect a stream. Wake up smarter.
Connect
One email label or Slack channel allowlist — not your whole company dump.
Ingest
Daily (or continuous) pull with dedupe, filters, and attachment extract.
Compile
Create and update wiki pages with wikilinks, tags, and provenance.
Ask / digest
Cited answers from compiled pages. Morning note of what changed.
Contrast
Compile vs chat-with-your-data
| Chat-with-your-data | Compound |
|---|---|
| Retrieves chunks each time | Compiles maintained pages |
| Answer dies in the thread | Knowledge accumulates in the vault |
| Hard to audit | Citations → raw sources + log |
| Lock-in formats | Markdown export / Obsidian-friendly |
Trust
Private by default. Leave with your files.
- Workspace isolation — your vault isn’t mixed with anyone else’s context.
- Encryption — TLS in transit; encryption at rest on stored content.
- No training on your data — your sources aren’t a model improvement feed.
- Full export & delete — markdown zip out; hard-delete when you say so.
- Start narrow — one low-risk label or channel before you widen scope.
- Human review hooks — contested claims surface instead of silently hardening.
Who it’s for
Founder-led teams and operators
Founders & PMs
Stop re-litigating decisions. Onboard people against a living wiki, not Slack archaeology.
Investors & researchers
Thesis and deal vaults that update as sources arrive — comparisons included.
Agencies
Per-client isolation, export at offboarding, less context trapped in inboxes.
Design partners — $99/mo
Limited seats. Email + Slack, daily compile, founder access, price lock when we GA. Apply below with intent “Design partner”.
Waitlist
Get in early
Tell us how context dies for you. We’ll write when a spot opens — design partners first.
FAQ
Straight answers
Is this just RAG?
No. RAG retrieves chunks per question. Compound compiles sources into maintained wiki pages with links, an index, and a log — then answers from that compiled layer.
What do you read from Slack or email?
Only what you allowlist (label, address, or channels). Filters exist so bots and noise don’t poison the vault.
Can I export?
Yes. Markdown vault export is a first-class feature — Obsidian-friendly on purpose.
Will you train on my data?
No. Customer content is not used to train models.
Who are design partners?
Early paid users ($99/mo) who connect a real source, give product feedback, and get founder access plus a price lock toward GA.
When do you launch?
We’re onboarding design partners from the waitlist first. Join and tell us your source — priority goes to clear pain + willingness to connect email or Slack.