Agent memory needs a standard — LoreKit ships a local one now

A front-page pitch for an OpenTelemetry-style memory spec across Claude Code, Cursor and Codex; plus Gemini auto-routes Omni 1.1 calls to 1.2 on Dec 3.

Nowline NOV 9 2:00 AM banner

Top AI stories from the last hour

Top AI stories from the last hour

Copy markdown

  • An OpenTelemetry moment for agent memory

    A widely-shared proposal argues agent memory should be a shared open standard — five MCP operations (write/read/list/search/archive), a portable Markdown record envelope, and conformance tests instead of per-tool silos. You can try the direction today: LoreKit (MIT) installs via `npx @lorekit/cli install` and gives Claude Code, Cursor and Codex local, inspectable memory as plain Markdown — with Portable Memory, Open Memory Protocol and Cognee's COGX chasing the same portability.

  • Update: Gemini auto-routes Omni 1.1 to 1.2 on Dec 3

    Google's Gemini API will silently route `gemini-omni-1.1-flash` requests to `gemini-omni-1.2-flash` on Dec 3 and bill them at 1.2 rates — no opt-out, no error, so unpinned video code keeps running at a new price. Audit your tools for 1.1 model IDs, pin `gemini-omni-1.2-flash` after testing, and re-cost per clip (720p ~$0.07/sec, ~30% under 1.1, though 4K upscales stack up).