Local Model Builds Automation Graph, Runs on Pure Python with Zero Token Cost
gustaw221133 · reddit · 2026-08-27
A developer shared a workflow for building automation tools using local models: converting natural language file chores into a graph of Python steps. The core idea is to let the model work only during the build phase (selecting from a pre-built library or writing custom code). The generated graph runs independently of the LLM, executing pure Python for zero token cost and deterministic results. The author solved reliability issues with small quantized models by relying heavily on the library, enforcing a tight schema, using typed sockets, and adding a planning step.
More from coding & agent
- Yutori's Batra: most of the web will never get agent APIs — pixels in, clicks out — DhruvBatra_ · 2026-08-27
- Testing local Ornith 1.5 35B: solid performance but hardware is the bottleneck — DanWahlin · 2026-08-27
- LLMs generate diagrams faster via DSLs like Penrose than from scratch — keenanisalive · 2026-08-27
- Open-source AI video project hits 42k stars: one prompt to finished video, clones viral hits — Vjeux · 2026-08-27
- Grok Build generates Mars simulator game from simple prompts — elonmusk · 2026-08-27
- Instinct personal agent launched: uses phones and computers like a human — saranormous · 2026-08-27