TokenRhythm launches NeoHorse-1, turning agent run experience into model training
Prajwal Tomar published a long thread introducing TokenRhythm, targeting a common pain point: an AI agent can find a bug, solve it through trial and error, yet make the same mistake again the next day—the conversation may be remembered, but the underlying model itself doesn't get any smarter, wasting vast amounts of valuable trial-and-error work.
Confirmed
- TokenRhythm consists of three parts: OpenSquilla runs tasks, the TokenRhythm API connects tasks to different models, and NeoHorse-1 learns from experience.
- The first step is Data-RSI: the system observes the entire task execution process—which model did the work, which tools were used, where it failed, how it recovered, and whether the final answer was valid—useful run records then become training samples for the next generation of models.
- The second step is a test-driven data flywheel: models are tested to find weaknesses (e.g., struggles with coding, tool use, or instruction following), the next training batch increases the share of such tasks, and the updated model goes through another round of testing, repeating the cycle.
Why it matters
- Most agent run experience today stays at the conversation level and never flows back into model weights; TokenRhythm offers a path to systematically turn real agent execution traces into training data, forming a self-improving loop of "execute—record—train—re-test."
2026-09-16 ~ 2026-09-16 · 9 related posts
Primary sources
- NeoHorse-1 open-sourced: 4B/9B Qwen3.5 post-trained models targeting recursive self-improvement — PrajwalTomar_ ·
- NeoHorse-1 4B beats its Qwen3.5 base on all ten tests, average score up 58.94 to 64.87 — PrajwalTomar_ ·
- NeoHorse-1 debuts: Apache 2.0 4B/9B models trained from agent runs, 262K context — PrajwalTomar_ ·
- AI agents fix bugs then forget — TokenRhythm turns agent runs into model training data — PrajwalTomar_ · 2026-09-16
- TokenRhythm's bet: train models on lessons from your agent's own runs — PrajwalTomar_ · 2026-09-16
- Inside TokenRhythm's three-part stack: OpenSquilla, a model API, and NeoHorse-1 — PrajwalTomar_ · 2026-09-16
- Data-RSI: turning useful agent runs into training examples for the next model — PrajwalTomar_ · 2026-09-16
- TokenRhythm's loop: test for weak spots, weight next training batch toward them — PrajwalTomar_ · 2026-09-16
- [source] NeoHorse-1 debuts: Apache 2.0 4B/9B models trained from agent runs, 262K context — PrajwalTomar_ · 2026-09-16
- [source] NeoHorse-1 4B beats its Qwen3.5 base on all ten tests, average score up 58.94 to 64.87 — PrajwalTomar_ · 2026-09-16
- [source] NeoHorse-1 open-sourced: 4B/9B Qwen3.5 post-trained models targeting recursive self-improvement — PrajwalTomar_ · 2026-09-16
1 near-duplicate retellings: PrajwalTomar_