Fabric-RLM: Open-source Recursive Language Model harness brings MIT's RLM to Microsoft Fabric notebooks
PawarBI · x · 2026-09-26
Sandeep Pawar, a Principal PM on Microsoft's Fabric CAT team, released Fabric-RLM, an open-source Python library built over 5 months that applies Recursive Language Models (RLM) to data-intensive tasks in Microsoft Fabric notebooks.
The problem it targets: asking an LLM about a large CSV, wide Excel file, or a pile of PDFs either overflows the context window or produces unverifiable numbers.
How it works:
- Instead of stuffing data into context, the model writes Python code to inspect the data and recursively calls itself on the pieces it needs, based on the MIT paper by Alex Zhang, Tim Kraska and Omar Khattab, built on DSPy's RLM module
- Users supply a plain-English task, inputs (Lakehouse files, Delta tables, semantic models, CSV), and an output contract with fields and types
- Code runs in a subprocess inside the notebook session; the model reviews results, fixes itself, and iterates
More from coding & agent
- Microsoft launches Autopilot, a managed service built on open-source OpenClaw — steipete · 2026-09-26
- Claude made a launch video by reading the repo — introducing Foleo, a local workspace for agent artifacts — joshgonsalves_ · 2026-09-26
- A weekend vehicular combat game coded entirely by Claude with 25+ prompts — DamageHelpful9364 · 2026-09-26
- Claude Opus 5.5 renders a whole Mumbai heritage sunrise scene purely in code, no 3D models — CurieuxExplorer · 2026-09-26
- ChatGPT Pro user regrets switching from Claude: 'output never what I asked' — Honest_Abroad_5846 · 2026-09-26
- Experiment: feeding thoughts to a NousResearch agent all day, no laptop needed — CurieuxExplorer · 2026-09-26