Each dot is a search a coding agent ran on Lacuna

Each dot is a search a coding agent ran on Lacuna

Each dot in this video is a search a coding agent ran on Lacuna, our ML research map, in the last 7 days. Devs from 34 countries, running through Claude, Codex and local installs.

Video: coding-agent searches on Lacuna, one dot per search, grouped by topic.

You can easily connect Lacuna to Claude Code or Codex: lacuna.tiptreesystems.com/mcp. Your agent can then look up papers, authors and research directions while it works on your code. It's read-only and free, and you don't need an account.

I mostly use it for three things:

  1. Before I build something, I ask if anyone has already done it, and get the closest papers back with links.
  2. When I hit a roadblock while training a model, I have it pull out the latest ideas and wisdom of the ancients.
  3. When I'm reading into a new area, I ask what the main papers are and dig into them.

Lacuna only covers ML research, so a lot of the searches are what you'd guess: reasoning, LM internals, quantization and KV caches, agent memory. Also anomaly detection and spiking networks.

Huge thanks to Miles Qi Li, PhD for helping build Lacuna and the MCP server.

First posted on LinkedIn, November 5, 2026.