Helix-AGI: dual-RWKV agent ditches JSON tool calls for latent-space action loop
LowDistribution3995 · reddit · 2026-10-17
An indie dev shared an update to Helix-AGI, an open-source agent harness built on two coupled RWKV models: a conversational Speaking Head plus a Sensory-Actualizer head that operates directly over a non-language vocabulary of primitive action/perception tokens (sense, move, read, write, run), coupled in latent space via hidden-state gain of 0.15 — bypassing the usual natural-language JSON tool-calling loop.
A shared Habitus hourglass graph (conserved edge weights, ∑w=1.0) turns OS file inspections and script executions into reflexive motor pathways, letting the agent build habitual behaviors without eating token space. Focus trajectory mapping and rolling JIT context bidding let it drift into subtasks and snap back. A real-time UI shows the graph and the agent's location in its environment. The repo is still messy; the author is looking for testers and collaborators.
More from coding & agent
- Claude Code Pushes Urgent Security Update Across All Installations — umeshonai · 2026-10-23
- Claude cut a 2.6-hour video export to 20 minutes — but only after being told to rebuild the server — byhewar · 2026-10-23
- 3 flight-booking rules my AI agent checks that I'd never do by hand — Efistoffeles · 2026-10-23
- OpenJet claims Qwen3.8-27B coding runs 1.9x faster than llama.cpp on one RTX 3090 — JeremyCMorgan · 2026-10-23
- Building an autonomous SEO agent in Hermes with three skills: programmatic pages, Keywords Everywhere, DataForSEO — EXM7777 · 2026-10-23
- Building a sports betting AI agent with onchain execution and human approval by default: architecture and lessons — Dry-Comfortable-2678 · 2026-10-23