WR

withrobinhq/robinwiki

API development

이 페이지는 AI로 자동 번역되었습니다. 기술적인 세부 사항은 원문을 확인하세요. 원문 보기

11 stars 품질 57 트렌드 22NOASSERTION

Robin knows best

개요

Talk to Robin through Claude Desktop, the web, or any MCP client. It listens, breaks what you say into atomic ideas, and files them into topic-clustered wikis you can search and edit. The point: capture without the friction of categorizing, then trust that whatever you said in March is one search away in November. For step-by-step installation instructions see www.withrobin.org/onboard You think out loud. Robin keeps the thread. You, in Claude Desktop: "I just realized the bouncer-mode flag should be per-wiki, not global. The whole point of strict review on the Decisions wiki is that it's high-stakes. Letting that bleed onto the Drafts wiki where I'm just thinking out loud would kill the value." 1. as an entry in the audit log 2. — atomic ideas, one thought each, fluff stripped 3. into a wiki by topic — "bouncer-mode design" goes to your Decisions wiki; the Drafts mention links a sibling fragment in Drafts 4.

README

원문 README의 일부입니다. 원문 보기.

Robin

A second brain that organizes itself.

Talk to Robin through Claude Desktop, the web, or any MCP client. It listens, breaks what you say into atomic ideas, and files them into topic-clustered wikis you can search and edit. The point: capture without the friction of categorizing, then trust that whatever you said in March is one search away in November.

For step-by-step installation instructions see www.withrobin.org/onboard

Table of Contents

What Robin does

You think out loud. Robin keeps the thread.

A typical capture looks like this:

You, in Claude Desktop: “I just realized the bouncer-mode flag should be per-wiki, not global. The whole point of strict review on the Decisions wiki is that it’s high-stakes. Letting that bleed onto the Drafts wiki where I’m just thinking out loud would kill the value.”

Behind the scenes, Robin:

  1. Captures the raw text as an entry in the audit log
  2. Splits it into fragments — atomic ideas, one thought each, fluff stripped
  3. Classifies each fragment into a wiki by topic — “bouncer-mode design” goes to your Decisions wiki; the Drafts mention links a sibling fragment in Drafts
  4. Regenerates the wiki bodies so the new fragments show up in prose, with citations linking back to the original
  5. Indexes everything for search — both keyword (BM25) and semantic (vector cosine), fused with reciprocal rank fusion

A week later, “what was my position on bouncer mode?” surfaces both fragments, the wiki body that synthesizes them, and the timeline of how the thought evolved.

Architecture

Single-tenant, single-user, Postgres-backed. Everything you write lives as text rows in Postgres — no git-backed markdown store, no filesystem repo. The server owns auth, the API, the MCP endpoint, and the AI pipeline.

Monorepo managed by pnpm workspaces + Turborepo:

core/             @robin/core    — Hono API server, MCP server, AI pipeline, workers
wiki/             @robin/wiki    — Next.js 16 web frontend (shadcn/ui)
packages/agent    @robin/agent   — LLM agent utilities, person resolution
packages/queue    @robin/queue   — BullMQ producer/consumer abstractions
packages/shared   @robin/shared  — Shared types, lookup keys, slug helpers
GitHub에서 전체 README 보기

추천 도구

다른 키워드를 입력하거나 필터를 제거해 보세요.

설치

npx skillfish add withrobinhq/robinwiki
MCP 초보자라면? mcpskill.org 가이드부터 시작하세요 →