Project memory your AI agents can trust
AI coding agents start every session knowing nothing about the last one. Knownbase is the memory layer that fixes it: a searchable workspace of decisions, runbooks, and project context that both people and agents can read and write.
Connects over the Model Context Protocol
Works with Claude Code, Codex, and ChatGPT
Free plan: 2 projects, 100 notes, no card
Revisions, backlinks, tags, and statuses
The problem it solves
Context dies with the session
Everything an agent worked out about your codebase is gone when the window closes, and the next session re-derives it.
Git records what, not why
A diff shows the change. It does not show the two approaches rejected first, or the outage that made the rule necessary.
Chat history is not retrieval
Scrolling back through transcripts is not a search system, and it does not survive a tool change.
Platform
Built for agents and the people working alongside them
A human dashboard for reading and curating, and an agent-facing tool surface for everything else.
MCP-native access
Agents connect over the Model Context Protocol and get real tools such as search, read, write, and link, rather than a pasted wall of context. Works with Claude Code, Codex, ChatGPT, and other MCP-compatible clients.
Search built for retrieval
Notes are searchable by project, tag, status, full text, and last-modified date, with semantic search where the plan supports it. Results stay lean by default so an agent does not burn its context window on a lookup.
Links and backlinks
Notes reference each other by title and resolve into a graph, so following one decision to the incident that caused it does not require knowing where to look.
Revisions and backups
Every note keeps its revision history, so a note an agent rewrote last week can still be read as it was.
Projects as boundaries
Each codebase, product, or client gets its own project. An agent working on one is not wading through another's runbooks.
The knowledge Git does not hold
Why an approach was rejected, what the last production incident actually taught, which decisions are still provisional: none of it lives in a diff, and all of it is expensive to rediscover.
Use Cases
Who Knownbase is for
Anyone paying the cost of context that has to be rebuilt from scratch.
Teams using AI coding agents
Give every session the same starting context instead of re-explaining the architecture, the conventions, and the last three decisions each time a new one opens.
Solo developers running several products
Switching between codebases means paging in context that has gone cold. Keep it written down where both you and your agents can find it.
Agencies managing client context
Handoffs, environment quirks, and client-specific decisions survive the gap between engagements and between the people staffed on them.
DevOps and platform teams
Runbooks, incident history, and operational decisions stay searchable by the humans on call and the agents helping them.
FAQ
Common questions about Knownbase
What is MCP, and why does it matter here?
The Model Context Protocol is an open standard that lets AI applications connect to external tools and data. Because Knownbase is an MCP server, an agent does not just receive a dump of text. It can search the workspace, read specific notes, and write findings back during a session.
How is this different from putting notes in the repository?
It is complementary. Documentation that must version with the code belongs in the repo. Knownbase is for the cross-session layer that does not: current state, why a decision was made, what the last debugging trail found, what the next session should pick up. In practice both are used together.
Which tools can connect to it?
Any MCP-compatible client. That includes Claude Code, Codex, and ChatGPT, alongside IDE agents that speak the protocol. Access uses static API keys or OAuth 2.1.
Is there a free plan?
Yes. There is a free plan with 2 projects and 100 notes, no card required. Paid plans add capacity and come with a 7-day free trial.
Should everything go into it?
No, and that is a deliberate product stance. Knownbase is for durable knowledge, not an archive of raw conversation. Notes should be concise, source-aware, and useful to whoever, or whatever, reads them next.
We use it on our own engagements
Knownbase came out of a real problem in our own delivery work: agents and engineers rediscovering the same context on every project. If you are integrating AI into a development workflow, whether MCP servers, agent tooling, or automation with proper engineering controls, that is work we take on for clients.