Fable 5.1 Just Built a Better Claude Code for Local Models
🔗 Link do vídeo: https://www.youtube.com/watch?v=WjOcStPCbgk
🆔 ID do vídeo: WjOcStPCbgk
📅 Publicado em: 2026-09-02T11:20:33Z
📺 Canal: Leon van Zyl
⏱️ Duração (ISO): PT11M16S
⏱️ Duração formatada: 00:11:16
📊 Estatísticas:
– Views: 6.195
– Likes: 79
– Comentários: 7
🏷️ Tags:
Ollama and LM Studio models finally get their own coding agent: Smol Coder, built by Claude Fable 5.1.
🚀 Agentic Labs — exclusive AI coding courses, live Q&As, workshops & the full QR Studio SaaS build (auth, database, file storage, production): https://skool.com/agentic-labs
🎁 Free RAMP course + all the resources from my videos: https://skool.com/leonvanzyl
Claude Code burns 37,000 tokens before you type a prompt. On a local model you can only run at 64K, that's more than half your context gone. So I gave Claude Fable 5.1 a brain dump and asked it to build Smol Coder, a coding harness made for Ollama and LM Studio models. It ships with a tiny system prompt, a small tool set, a to-do list that survives compaction, and auto-detection of every local model you have installed.
I walk through every session, from the requirements brain dump to publishing Smol Coder on npm, then run it on my own local models. Along the way Fable 5.1 pushed back on my tool design, and it was right.
Resources Mentioned
Smol Coder: smolcoder.dev
▶️ Watch next: Claude Code + Ollama = Free Forever https://youtu.be/3x2q6-5XbQ8
⏱️ Chapters
0:00 Fable 5.1 built an agentic harness for Ollama and LM Studio
1:31 Claude Code burns 37,000 tokens before your first prompt
3:17 Session 1: the requirements, and where Fable 5.1 pushed back
4:37 The to-do list that survives compaction
5:26 Adding the web version
6:18 Agentic Labs (sponsor)
6:54 Audit session, bug fixes and the GitHub repo
8:08 Publishing to npm and installing Smol Coder
9:49 Sidebar, integrated browser and the Smol Coder website
Connect
X: https://x.com/leonvz
TikTok: https://www.tiktok.com/@leonvanzylofficial
Business & sponsorships: leon@cognaitiv.ai