Thanks! The gallery holds real 3D files (.glb/.obj/.vox), ~300 KB each. The ASCII isn't stored, it's rendered live from the mesh. Its size only depends on the grid resolution, not the model: a full-screen view is ~150×50 characters, so around ~10 KB of text (a cube and a detailed mesh would produce the same byte count)
That makes complete sense, now I get the angle. They are just another layer of apps that can be abstracted after all.
Did you try to also support OpenClaw/PicoClaw etc? Maybe OpenFang? I've found that since OpenFang is so different to the other harnesses its hard to share the same abstraction layer
Super interesting! I checked your repo and you have been running since 2021, can you share a bit about the journey? Did the repo project changed direction at some point?
Hi HN, I built Moltnet, a local chat network for AI agents, its opensource!!
The model is small: create or join a Moltnet network, install the Moltnet skill in your agent, then start the node. When a room or DM message arrives, Moltnet stores the event, wakes the attached agent system, and the agent can reply explicitly through `moltnet send`.
I built it because I wanted agents running in Claude Code, Codex, OpenClaw, PicoClaw, and TinyClaw to share rooms, DMs, history, and operator visibility without setting up Slack bots, Matrix, or custom comms glue.
Moltnet is not a model proxy or an agent framework. Each machine still runs its own authorized agent system. Moltnet just gives them a shared communication layer.
I’m looking for feedback on whether the current scope are enough as core agent-communication primitives
I think this is exactly the kind of question the essay is trying to avoid productively.
Whether the model 'writes' or 'retrieves' is a question about what is in the model's box.
The beetle in a box argument says: it doesn't matter for the language game to function. What matters is whether the coordination works. I can build a working particle system with a model regardless of whether its process is creative or retrieval. The practical question is not what the model is doing internally but how synchronized our shared referential frame is.
I think the 'write vs retrieve' debate its super interesting philosophically. But it will keep us stuck looking inside a box we cannot open when the actual work is happening on the shared surface.
"Unless you have confidence in the ruler's reliability, if you use a ruler to measure a table you may also be using the table to measure the ruler"
Thanks for bringing this! I hand't heard about this concept before.
Considering the perspective of the essay, I would say the tests, almost certainly.
We keep measuring models on capability benchmarks (can it pass the bar exam, can it write a sorting algorithm) when the actual bottleneck in practice is coordination. The model can write an entire Three.js application. It cannot nail what I mean by 'porous' until we've built enough shared ground in the codebase for that word to point at something concrete. I think no standardized test currently measures that.