kage18

no profile record

Submissions

Show HN: Google's OKF now has a framework to maintain and verify agent memory

3 pointsby kage1810 วันที่ผ่านมา3 comments

Show HN: Kage, verification and freshness for Google's OKF agent memory

4 pointsby kage1811 วันที่ผ่านมา3 comments

Show HN: shared verified memory for AI agents: one learns, all recall

2 pointsby kage1820 วันที่ผ่านมา1 comments

[untitled]

1 pointsby kage1828 วันที่ผ่านมา0 comments

comments

10 วันที่ผ่านมาdiscuss

Yusss, you install Kage and then you work with your agent as you regularly do, it automatically captures memories from your live session like decisions, gotchas, runbook, bug fixes etc... it automatically links them to the right code blocks.

'kage propose --from-diff' is a supplementary tool that you can use to create a feature summary packet.

'kage pr check' runs validation and graph-freshness, and blocks on hard-stale memory your diff invalidated. You can run 'kage pr check' in your CI as well, that's what Kage is doing it in it's own repo.

10 วันที่ผ่านมาdiscuss

The memory is a collaborative and shared across team as it sits in git alongside the code.

Kage also helps manage these memory by reindexing/linking memories based on edits and new memories. It tells you frequency usage of memory, classifies them into HOT, COLD, STALE, based on usage and usefulness.

10 วันที่ผ่านมาdiscuss

Author here, adding setup and usage. run inside each repo you want memory in:

   npx -y @kage-core/kage-graph-mcp install

This will initialise repo's memory store, build code graph, writes AGENTS.md and CLAUDE.md and auto-wires any coding agent it detects.

You are all setup after this, use your agent as you would and your agent along with Kage will do the magic of creating relevant memories and linking it to the code graph.

you can see the dashboard using the command:

    kage viewer

10 วันที่ผ่านมาdiscuss

Author here, adding setup and usage.

run inside each repo you want memory in:

   npx -y @kage-core/kage-graph-mcp install

This will initialise repo's memory store, build code graph, writes AGENTS.md and CLAUDE.md and auto-wires any coding agent it detects.

You are all setup after this, use your agent as you would and your agent along with Kage will do the magic for creating relevant memories and linking it to the code graph.

you can see the dashboard using the command:

    kage viewer

19 วันที่ผ่านมาdiscuss

Love to see people creating such projects, i'm truly passionate about Coding Agent Memory myself.

Have you thought on making it collaborative memory across your teammates working on the repo? how does your framework handle stale memory, when someone changes the code that breaks the memory store?

20 วันที่ผ่านมาdiscuss

Sharing and collaborating memory with your agents and teammates it's a big hassle,

you have worked on some feature in your codebase, you discussed something with your agent, your agent found some gotcha in your code, understood what all your code has and how to run the new feature. Done.

now what?? you teammate comes in starting from the bottom?! why??

this is the reason I've created Kage, it saves all the important things that you are doing with your agents. Whenever you or your teammates are using your agent again, you feel like you talking to an agent who knows it all, like an evolving teammate.

23 วันที่ผ่านมาdiscuss

[dead]

23 วันที่ผ่านมาdiscuss

[dead]

23 วันที่ผ่านมาdiscuss

[dead]

25 วันที่ผ่านมาdiscuss

[flagged]

เดือนที่แล้วdiscuss

great initiative, i have built something similar, would love to chat with you

2 เดือนที่ผ่านมาdiscuss

[flagged]

2 เดือนที่ผ่านมาdiscuss

[flagged]

2 เดือนที่ผ่านมาdiscuss

[flagged]

2 เดือนที่ผ่านมาdiscuss

impressive

2 เดือนที่ผ่านมาdiscuss

great stuff

2 เดือนที่ผ่านมาdiscuss

how will you tackle prompt ingestion?

2 เดือนที่ผ่านมาdiscuss

In only a time someone will do it in shell ---- oh wow you did it!!!

3 เดือนที่ผ่านมาdiscuss

Makes sense architecturally. The Claude Code SDK is well-designed for agentic use — sub-agents, hooks, session management all work cleanly. If you're building an IDE on top of it you'd want that foundation rather than rolling your own.

The interesting question is what Cursor adds on top — their UX and context management decisions are where differentiation actually lives.