Your deep research, re-injected.
You spent hours feeding prompts to Gemini and Claude. The dense, well-structured output landed in Google Docs — paid for in cognitive labor, then shelved. Relational Intelligence turns that Drive folder into a queryable, copy-pasteable corpus your future selves and future models can build on.
Thought Molecules
We call those research docs thought molecules: the cognitive building blocks of deep research. Reusable. Recombinable. Citable. Each one cost real attention to make, and most of them will never be read again — not because they aren't valuable, but because there's no way to find the exact passage you need at the exact moment you need it.
What it does
Point it at a Google Drive folder. It chunks every doc by heading, embeds the chunks locally, and indexes them for both lexical and semantic search. Then, when you ask in natural language —
"every passage where I argue Demis Hassabis's scene construction theory shaped DeepMind's architecture"
— an agent loop plans the search, retrieves candidates, reranks them, decides whether it needs to look further, and assembles the final answer as verbatim passages from your own research, each citing the source doc and section. Paste it into Claude, Gemini, or any other context window and keep building.
The Promise
Doctrinal-grade quote fidelity. Every passage shipped is programmatically verified as a substring of the actual source chunk before it leaves the process. No paraphrase. No bridging prose. No editorial commentary. If a phrase isn't in your docs, it doesn't appear in the output. One fabricated quote breaks the trust forever, so the trust mechanism is in code, not in vibes.
How it ships
- CLI —
cj query "…". Local-first. Your corpus, your machine, your keys. - MCP server — wire it into Claude Desktop or Claude Code as a tool.
- Web UI — minimal local front-end with streaming traces of the agent's reasoning.
Stack
SQLite + sqlite-vec + FTS5 (zero infrastructure, one file on disk). Voyage voyage-3-large embeddings. Voyage rerank-2. Claude Sonnet 4.6 for query planning, the agent loop, and synthesis. Open source, MIT.