Tess creator open-sources fine-tuning recipe: turn Claude Code sessions into a local model
udmrzn · x · 2026-11-08
Open-source model author Miguel Tissera (Tess/Synthia, 100+ models) has released his complete LLM fine-tuning recipe as a GitHub skill repo, tinker-ai-skill (MIT, 205 stars).
- Core idea: every coding-agent session is a worked example of how you want a model to behave — your repos, conventions, tools, debugging style.
- The skill hands your agent a production-tested pipeline: convert sessions into an SFT dataset, train a LoRA on Tinker, merge it, and ship a GGUF that runs locally in llama.cpp.
- Claims one afternoon and a few dollars per model; Tess-4 and Tess-4.5 were built with this pipeline.
- Works in Claude Code, Codex, and OpenCode, with Python scripts and full instructions included.
More from coding & agent
- DataTalksClub frees course materials for Production AI Agents for Generative Media, projects application-only — Al_Grigor · 2026-11-08
- GitHub's new Git storage architecture handles 7,200 pushes/sec with 5,000 concurrent agents per repo — nicolascraske · 2026-11-08
- Engineer's data: 11x more merged PRs per month at AI-native Tano vs pre-AI PostHog — bibryam · 2026-11-08
- No Chinese models allowed: what's left for local document OCR when contracts rule out PaddleOCR and Qwen? — aidenclarke_12 · 2026-11-08
- ADLC means two different things: Agentic Development Lifecycle vs Agent Development Lifecycle — Thirumalaivasan_GJ · 2026-11-08
- Anthropic's Official claude-quickstarts Repo Nears 20k Stars for Claude API Developers — anthropics · 2026-11-08