Company knowledge is scattered across docs, repos, chat, and people's heads. CompanyOS turns your raw data into a coherent, self-maintaining wiki — then exposes it to Claude, ChatGPT, and any MCP client so your team can ask open questions and get answers grounded in your own data, with sources.
you › who owns billing, and what did we decide about pricing?
⚙ search_wiki "billing owner · pricing decision"
⬡ oracle Billing is owned by Dana R. (People › Ownership). On pricing: the team moved to seat-based in Q1 [Decisions/pricing.md · 2026-02-11]. A conflicting note proposes usage-based [⚠ uncertain · 2026-04-02] — both kept side-by-side.
Search tools return documents, not answers — and they go stale. Your team already works inside AI clients, but those clients have no grounded, current knowledge of your company. There's no single trustworthy place to ask.
Intelligence lives at ingest time. The read path stays thin and fast.
Ingest-time intelligence
Feed in pasted content, uploads, or connected git repos. An LLM-driven pipeline parses each input, routes it to the right topic page, and reconciles it into the existing wiki — merging facts, keeping conflicting claims side-by-side with sources and dates, and creating new topic pages when nothing fits.
Committed as markdown to a private GitHub repo. Git is the source of truth.
No server-side generation
Connect the CompanyOS MCP server to your AI client over OAuth. The client gets
tools to search, list, and read the wiki.
Your own model reasons over the retrieved pages and answers, citing the wiki's
provenance.
Answers grounded only in your tenant's data — never another company's.
Every fact carries its source and date. Nothing is silently overwritten.
Each fact carries its source and date, so you can trust and audit where knowledge came from — and the Oracle cites it back to you.
Contradicting facts are kept side-by-side and marked uncertain, never silently overwritten. You never lose information or see disagreements hidden.
Your wiki is a private GitHub repo you can browse. Read history, see diffs, recover from any indexing problem. The search index is a disposable, rebuildable cache.
Keyword and semantic ranking fused together, so you get good results whether your wording matches the docs or not — for both reads and merge routing.
Ingestion runs automatically on submit or git push. Re-ingesting an updated doc updates the relevant pages instead of duplicating them.
Per-tenant MCP URL plus OAuth. Tokens carry tenant and per-user identity, so your company's data stays isolated and access ties to real people.
CompanyOS exposes your brain to AI clients as an MCP server. Add it, log in with OAuth, and your team gets a small, stable set of read tools — the client's own model does the reasoning.
list_pagessearch_wikiread_pagePaste, upload, or wire up a git repo. Seed an initial topic ontology your team recognizes.
The pipeline routes each chunk to the right page, merges facts, and stamps provenance.
Clean topic-scoped markdown lands in your private GitHub repo — the durable source of truth.
Query the Oracle from Claude or ChatGPT. Grounded answers, with sources, in seconds.
CompanyOS v1 onboards a single tenant end-to-end. Get in touch to connect yours.