open-deepthink 0.1.11 adds qualitative self-attention and portable agent skills
causality-ai · reddit · 2026-07-24
open-deepthink 0.1.11 adds qualitative self-attention and portable agent skills
The open-source multi-agent system open-deepthink has released 0.1.11 with two notable additions:
- Qualitative self-attention inside multi-agent brainstorming, letting agents read from non-neighbor and past-layer neurons instead of only the previous layer.
- Portable skills that can be copied into tools like Grok Build, Claude Code, Cursor, and Codex without running the full server.
The new brainstorming mode treats LLMs like neurons in a network, with layered forward passes, Mirror Descent on personas, and epoch reframing. The qualitative self-attention flow is described as:
- build a pool of non-local past neurons
- score pairs using qualitative buckets instead of matrix multiplication
- inject the top-k into the prompt as attended values
- log edges in attentionedges
The repo also ships an App Slot Machine mode, which turns vague app ideas into a structured build prompt via a diffusion-like corruption and critique process. Two skills are bundled:
- /qnn for strategy/deadlock exploration
- /qdad for turning a rough app vibe into a buildable prompt
More from coding & agent
- How should an MCP app be designed for non-technical users? — sn0wquake · 2026-07-24
- Greg Kamradt highlights disposable URLs as a new primitive for agent workspaces — GregKamradt · 2026-07-24
- When should you use a dedicated memory framework for agents? — markotkid · 2026-07-24
- Sentry’s Seer agent answered a Slack question with 30 days of Notion MCP usage data — zeeg · 2026-07-24
- Greg Kamradt says vibe coding raises both the floor and the ceiling — GregKamradt · 2026-07-24
- This AI brief writes your actions back so tomorrow’s summary gets smarter — Spirited_Ad_3886 · 2026-07-24