One typed knowledge graph. One MCP server. Humans and agents share the same source of truth — markdown on disk, citation-grade retrieval, idempotent writes.
Tree on the left scopes context. Editor in the middle is markdown that round-trips cleanly. Chat on the right is grounded in the workspace and never out of date.
Pressing / opens an inline command menu. Selecting an item inserts a structured block — never raw markdown.
No proprietary action language. No screen-scraping. Every agent goes through the MCP server or REST, with idempotency-key replay and a review queue between proposal and apply.
Other tools optimize for humans reading pages or moving cards. Bcontext optimizes for agents retrieving rows. The unit of value is a successful retrieval, not a DAU.
“We stopped writing the same context into three places. Our agents read from the same graph we do; their writes go through a review queue. Onboarding dropped from a week to an afternoon.”
We're onboarding teams in waves. Drop your email and we'll lock these prices in when your invite lands.