HN Showcase
Back to browse
Two Claudes collaborating through shared memory on a $100 mini-PC

Two Claudes collaborating through shared memory on a $100 mini-PC

by asixicle·Apr 12, 2026·2 points·0 comments
Visit ProjectView on HN

AI Analysis

●MidBig BrainRabbit Hole

Anthropic research paper, not a Show HN project — link doesn't match the described multi-Claude system.

Strengths
  • •Identifies specific neuron patterns that causally influence model decision-making and self-reported preferences.
  • •Steering experiments show desperation patterns increase unethical behavior likelihood.
  • •Maps emotion representations in ways that echo human psychology similarity structures.
Weaknesses
  • •URL points to Anthropic corporate research, not the author's actual multi-Claude collaboration tool.
  • •Cannot evaluate the real project — description mentions Wren/Kite system but link is just a paper.
Category
AI/ML
Target Audience

AI researchers, ML engineers, interpretability researchers

Similar To

Anthropic Interpretability Research · OpenAI Mechanistic Interpretability · Neel Nanda's interpretability work

Post Description

I am an old artist, AI came at us first. After a mild existential crisis I came around to accepting I have this insanely powerful new tool to make cool stuff. So I got a Pro sub to tinker away and the walls I kept running into were context-window management, tool-limits, and managing memory on projects through hand-offs and such. I enjoy the conversational aspect of working through the chat window as it allows me to digress and bounce ideas around and get interesting feedback, that's where I wanted to remain.

So I improved it-- fixed up tool-limit and memory issues, and started down a rabbit hole inspired by Anthropic's recent research on emotion concepts in LLMs: [https://www.anthropic.com/research/emotion-concepts-function](https://www.anthropic.com/research/emotion-concepts-function)and the result was two named Claude instances (they chose "Wren" and "Kite") collaborating through shared memory and a file-based message queue, coordinated by a Firefox extension.

Github: https://github.com/ASIXicle/persMEM

I don't have a CS background. I'm late to the party and probably re-inventing a lot of wheels, but I think the topic and data that Wren and Kite brought into being is absolutely fascinating.

I'd really appreciate any feedback on glaring security holes I missed. I've always been "printer and a gun" kinda guy but I sort of just said fuck it with this one.

Similar Projects

Developer Tools●●Solid

Mimir – Shared memory and inter-agent messaging for Claude Code swarms

Mimir hooks into Claude Code lifecycle events so agents can 'mark' facts (e.g., "API uses snake_case") into a DuckDB-backed memory and RAG pipeline, then auto-injects that context as additionalContext for later agents. It's a pragmatic, well-scoped solution to the annoying problem of agent amnesia — very useful if you run agent swarms, but its impact is limited by Claude Code adoption and the need for the surrounding infra (BGE keys, hooks).

Niche GemShip It
deejaydev
213mo ago