Feed it, staff it, sell it, sheet 10 of 13
Permission-aware retrieval across the company's systems, and memory that outlives the engineer who built it up.
Prove the permissions with a few planted test files before you sign again, and commit what's worth keeping next to the code.
← → move to the next sheet over, Esc zooms out.
An agent is only as useful as what it can read: the docs, the tickets, the chat history, the customer calls. The context layer is the plumbing that lets it search all of that, and the hard part isn't search. It's making sure the agent only sees what the person asking is allowed to see.
It's a library where every book has its own lock, and the librarian has to check your badge against each one before reading you a sentence. A good librarian is fast. A careless one reads you the HR file.
Then there's memory: what an agent learns while working with you. Today most of it lives in one person's chat history on one laptop. Turn in the laptop and the knowledge goes with it, unless someone wrote it down where the whole team can read it.
Every link opened on 11 Nov 2026. The numbers match the squares on the map.
Product
An enterprise AI platform connected to your company's data, with search, an assistant and agents on top.
The catch. Built for the whole company at once. Ask what search alone costs before you buy the seats.
Open source, paid plans
A self-hostable AI platform for chat and search across your company's sources, with an open-source Community Edition.
The catch. Group-based document access needs a Business or Enterprise plan. In the Community Edition, a document is visible to its connector's creator, and to everyone when the connector is public.
Product
Turns your systems into governed MCP servers over what it says are a thousand-plus connectors, with permissions inherited and synced from each source.
The catch. Inherited permissions are only as good as the sources. Fix the oversharing first.
Product
Connects company knowledge, tools and teams so people can build and share agents, with permissions on data, tools and agents and an audit log.
The catch. Easy for anyone to build an agent means you need to know who built which one.
Open source
Keeps your email, chats, meetings and calendars on your own computer or server, searchable, with an MCP server so an assistant can use the archive.
The catch. Built for one person's archive, not a company's permissions.
Open format
A simple, open format for guiding coding agents: a README for agents, checked into the repo, used by over 60,000 open-source projects, by its own count.
The catch. A format, not a process. Somebody still has to review what goes in it.
Open source and product
Persistent memory for AI apps and agents, as a hosted platform or self-hosted open source.
The catch. Memory that outlives a session also outlives a person's access. Decide what it may keep, and for how long.