LLMSkillHub
mirroredCopied from ShenSeanChen/waku-agent under MIT. ShenSeanChen has not claimed this page, and is not affiliated with LLM SkillHub.

shenseanchen / waku-memory

Connect Waku Memory, hosted cross-agent memory in Claude Code, Codex, Grok Bot, Hermes. Export Waku skills.

0
SKILL.md
1.9 KB

What Waku Memory is

This app's memory is local. It lives in ~/.waku/ on this machine. Waku also
writes each fact to ~/.waku/memory/, one file per fact, so the Waku Memory
importer can upload it. Waku Memory (waku.one) is a hosted memory that several
agents share over MCP: a fact saved in Claude Code can be recalled in Codex,
Grok Bot or this app. A fact reaches Waku Memory only when it is uploaded, so
never tell the user a fact is there unless a tool's output says so.

Connecting this app

Connecting opens the user's browser, so you cannot do it for them. Tell them to
type /connect waku-memory in the dashboard chat, or to run
waku connect waku-memory in a terminal. It needs the MCP extra:
pip install 'waku-agent[mcp]'. After it connects, a restart of Waku loads the
waku_memory_* tools.

Connecting other agents

  • Claude Code, Codex, Hermes: point them to https://www.waku.one/docs.
  • Grok Bot: Settings → Plugins → add a custom connector with the URL
    https://api.waku.one/mcp and the header Authorization: Bearer <key>,
    using a key from waku.one → Account → Keys. Its connector form takes a URL
    and a header, not a browser sign-in, which is why it needs the key.

  • Muse Code or any other MCP client: add https://api.waku.one/mcp as a
    remote (streamable HTTP) server.

Tell the user to paste a key only into the other agent's own settings, never
into this chat.

Carrying skills

waku skill export --to claude,codex copies Waku's skills into Claude Code's
and Codex's skills folders. --project writes to ./.claude/skills, which
Muse Code also reads. Saving skills into Waku Memory is not available yet.

What not to claim

Don't quote prices, plans or limits; point to https://www.waku.one/docs.