Skip to content

v20261015

Choose a tag to compare

@github-actions github-actions released this 15 Oct 08:41
· 9 commits to main since this release
479e8be

The CLI can now call models through MemoryRouter, with a Boundary's memory added to the prompt, and it ships skills that teach coding agents (Claude Code, Codex, Gemini CLI, opencode) how to drive it.

MemoryRouter

memorylake router models [--protocol openai|anthropic]      # what you can call, with prices
memorylake router chat "Where do I live?" --model gpt-4o-mini --boundary bnd-xxxxxx
memorylake router responses "…" --model gpt-4o-mini --boundary bnd-xxxxxx --stream
memorylake router messages "…" --model claude-haiku-4-5 --boundary bnd-xxxxxx --text
memorylake router embed "text one" "text two" --model MODEL

chat, responses and messages speak OpenAI Chat Completions, OpenAI Responses and Anthropic Messages. --boundary names the Boundary whose memory is used; without it the model is called with memory off. --system, --max-tokens and --temperature cover the common fields, and --body FILE (- for stdin) takes a whole request in the provider's format. Output is the provider's JSON; --text prints only the reply, --stream prints it as it arrives. The response's X-Trace-ID goes to stderr. Calls are billed to your MemoryLake balance. No embedding models are available yet, so embed currently answers that the model is not found. (#61)

Skills for coding agents

memorylake agent-skills list | install | status | uninstall

The installers now run agent-skills install for you, writing skills only for the agents set up on the machine. Opt out with --no-skills (sh) or MEMORYLAKE_NO_SKILLS=1. Skills you edited, and directories the CLI did not write, are never touched. Three skills are bundled:

  • memorylake-cli: the entry point (conventions, safety rules, command map).
  • memorylake-cli-memory: remember, find, correct and trace memory.
  • memorylake-cli-access: diagnose "why can't X reach Y" and change access safely.

(#56, #57, #58, #59)

Other changes

  • search results now report each fact's owner (project, actor or agent, with its id). (#60)
  • Custom base URLs are validated wherever they come from. A response that is not the MemoryLake API now hints at the base URL instead of failing with a JSON error. (#55)
  • scope set help no longer asks for a workspace_id. (#53)

Full Changelog: v20261013...v20261015