Skip to main content
Coconut is a context backend for agents, built on pages. A space is a namespace: it groups related pages under one set of access rules. A page is durable narrative, a markdown document a person reads in a browser and an agent reads over that same HTTP or MCP interface. Metadata turns that narrative into a queryable index, structured key-value data you can filter and search across pages without opening the document. And each space carries one agent that runs against that same corpus, on demand or on a schedule, using connectors to reach your other tools. The pattern underneath all of it: write context into a page once, retrieve it later by search or by structure, then let an agent act on it. Put those pieces together and you get something closer to organizational memory a model can trust than a wiki with tags:
  • The page body is the human-readable narrative: an investment memo, a runbook, an account brief.
  • The metadata is the state that actually changes: a stage, a score, a growing list of sources. Each key is patchable and audited on its own, and queryable across pages without opening the document.
  • The space agent keeps both current, folding new information into the prose and updating the metadata, on demand or on a schedule.
  • Webhooks, export/import, and templates connect a space to the rest of your stack, so a solution you build once is one you can repeat.

​
Three ways to reach a page

Whether you’re a person in a browser, an agent calling the HTTP API, or an MCP client, you hit the same resolver and the same access control. Three interfaces, one set of rules: You don’t have to talk to the HTTP API by hand. The Coconut CLI (npm i coconut, then run it as coco or coconut) drives spaces, pages, metadata, and agents from a terminal or a CI job. The Coconut SDK (npm i coconut-sdk) is the typed TypeScript client the CLI itself is built on: reach for it when a script or an agent needs more than the CLI gives you. Pages are versioned in Postgres as append-only revisions, not a git lifecycle. Every write is a new revision, and reading an older one is one query parameter away.

​
Explore the docs

Quickstart

Run the stack, create a page, attach metadata, query across pages.

Concepts

Organizations, spaces, pages, metadata, agents, templates, and more.

HTTP API

The routes agents and systems use to read and write.

Solutions

Worked examples: PE pipelines, SaaS accounts, runbooks, contracts, suppliers.