What your company knows
The AI knowledge base your team and your agents both read from.
It's where your company's context lives: brand voice, personas, playbooks, decisions, and the facts your work runs on. Your team reads it like documents. Your agents read the same thing, live, through whatever tool they already run in.
$149/mo after trial · Cancel any time
Documents on top, plain files underneath
Everything in the knowledge base is a living document in one workspace tree: people, companies, projects, decisions, playbooks. Your team edits in a normal document editor — autosave, a slash menu, nothing to learn. Underneath, every document is plain markdown you can export in full, any time. Nothing proprietary holds your knowledge hostage.
Documents link to each other, and the links resolve both ways: open your brand voice guide and see every playbook that depends on it. One search runs across all of it, ranked so approved knowledge surfaces first. Collections group documents into named sets — the approved kit for a campaign, a client, a function — so people pull from the curated shelf, not the whole library.
Seed it with what you already have
Nobody starts from a blank page. Import a repo or a set of URLs and your existing material — CLAUDE.md files, brand docs, SOPs, that one doc — comes in as-is. And the knowledge base can feed itself: connect a Slack channel, an Instantly campaign, or any URL, and what happens there syncs in as source material your team reviews like everything else.
Every door in
Your team uses the app. Your agents pick whichever door fits where they run — and the reading is the same through every one of them:
- MCP, for agents in Claude Code, Claude Desktop, Cursor, or any MCP-compatible tool. Agents discover what's available on their own; connecting is a guided one-time setup.
- A CLI, for agents and scripts living in a terminal.
- A retrieval API that returns the best-fitting context sized to an agent's context window, respecting what your team has approved — not just what matches.
If your team isn't technical, you never touch any of this. The doors are for your agents; your team works in documents and the review screen.
Go deeper
- The MCP server — how agents in Claude Code, Cursor, or Claude Desktop connect and read.
- The prompt library — playbooks: your prompts with inputs, context, and an approval state attached.
The knowledge base is one third of the story
What keeps it current is the AI memory: Patina mines what agents learn and flags what's gone stale. And what makes it trustworthy is how things get in: agents and teammates propose, and your team approves. That's human-in-the-loop review — and every approval joins the audit trail.
$149/mo after trial · Cancel any time