Every AI coding session starts from zero. Every engineer re-derives the same architectural calls. The reasoning behind your codebase lives in people's heads, merged PRs, and lost Slack threads — and it evaporates. kgai is the shared memory layer your repo and your AI have been missing.
claude plugin marketplace add kgaidev/kgaiThe bottleneck in modern development moved. It's no longer typing the code — it's holding the context: why this was built this way, what was tried and rejected, which decision replaced which. Two forces make that harder every day.
Your assistant is brilliant and forgetful. Each session it re-derives choices you already made, contradicts past decisions, and confidently re-opens debates you closed months ago — because it never saw them.
The why lives in one person's head. When they context-switch, go on holiday, or leave, it's gone. New teammates — and new AI — inherit the code but not the reasoning that shaped it.
The result: the same decisions, re-made, re-litigated, and quietly broken — over and over.
kgai turns the decisions behind your code into a living knowledge graph — durable, searchable, and versioned. Your AI reads it before it touches your code, and writes to it after. So does everyone else.
Every other tool records only what you shipped. kgai keeps the rejected approaches, the abandoned experiments, the alternatives you ruled out — together with the reason they failed. Because a decision is never deleted, the blind alleys stay in the graph forever. So no engineer, and no AI, ever burns a week re-walking a path your team already proved was wrong.
kgai models your app as elements — features, services, business objects — reshaped over time by immutable decisions. Three moves, all automatic.
As you and your AI work, real architectural decisions are recorded — what changed, and the why — as immutable events. No extra ceremony.
Before changing code, your AI pulls the decisions relevant to those files — so it respects prior calls instead of contradicting them.
Decisions change. A new one supersedes the old, and the full lineage stays queryable. Ask “why is this like this?” and get the real history.
Append-only event history is the source of truth. The graph is a rebuildable projection of it.
Two teammates decide differently on the same element? kgai surfaces the branch instead of silently losing one.
Ask in plain language, not exact keywords. Relevance-ranked recall built for how AI actually queries.
Sync over a private git repo or S3 bucket you own. Free, no lock-in.
Reconstruct the exact shape of the graph as of any past date. See what the team believed, then.
The engine and plugin are MIT-licensed. Your memory is yours.
Real codebases accumulate a lot of history. kgai is engineered for it — thousands of interconnected elements and tens of thousands of decisions stay instant to query. A brand-new teammate clones the entire decision history and rebuilds the whole graph in seconds, not minutes.
Bring-your-own git or S3 is free forever. kgai cloud adds a hosted sync plane, an interactive graph you can explore and edit in the browser, and an MCP endpoint to plug the shared memory into any AI — with zero infrastructure to run.
Request beta accessGive your codebase — and your AI — a memory the whole team shares.
claude plugin marketplace add kgaidev/kgai