---
title: "We got local models to triage the OpenClaw repo for FREE!*"
description: "OpenClaw receives hundreds of issues and pull requests daily, creating a triage problem for maintainers who need near-real-time filtering without consuming a paid closed-model quota. The system uses Gemma and Qwen local models inside a Pi-based agent harness: each item is normalized, enriched with GitHub context, classified into finite topic labels, and required to return structured output. For pull requests, the agent can inspect the repository through reposhell, a read-only command-restricted shell, while localpager, SQLite, a worker queue, and deterministic notification policy handle orchestration and Discord delivery. Tests described for gemma-4-26b-a4b and qwen3.6-35b-a3b report hundreds of tokens per second and good zero-shot classification accuracy without fine-tuning, while an LLM audit loop monitors errors; the authors present the pattern as agentic classification for high-throughput triage."
---

# We got local models to triage the OpenClaw repo for FREE!\*

[Hugging Face](https://yomu.fyi/company/hugging-face) · Onur Solmaz, ben burtenshaw, shaun smith · Jun 22, 2026

**Type:** Problem & solution

## Summary

OpenClaw receives hundreds of issues and pull requests daily, creating a triage problem for maintainers who need near-real-time filtering without consuming a paid closed-model quota. The system uses Gemma and Qwen local models inside a Pi-based agent harness: each item is normalized, enriched with GitHub context, classified into finite topic labels, and required to return structured output. For pull requests, the agent can inspect the repository through reposhell, a read-only command-restricted shell, while localpager, SQLite, a worker queue, and deterministic notification policy handle orchestration and Discord delivery. Tests described for gemma-4-26b-a4b and qwen3.6-35b-a3b report hundreds of tokens per second and good zero-shot classification accuracy without fine-tuning, while an LLM audit loop monitors errors; the authors present the pattern as agentic classification for high-throughput triage.

## Context

OpenClaw receives hundreds of issues and pull requests each day, and maintainers need to triage, prioritize, and route them quickly. Running a paid closed model for every item could consume the available quota or require batching, which would delay notifications. The work focuses on filtering and notifying a maintainer only about issues in their responsible topics using local open-weight models on existing hardware.

## Approach / What changed

Issues and pull requests are mirrored and normalized into localpager's SQLite database, where new items become classification jobs. A worker supplies GitHub metadata and selected diff context to a Pi-based agent harness running Gemma or Qwen through a local endpoint. The agent may gather additional repository context through reposhell, a read-only restricted shell, and must submit labels through a final\_json schema. Results are stored and passed to Discord according to deterministic notification rules.

## Takeaways

- Reposhell limits the model to read-only repository commands such as ls, find, rg, grep, sed, cat, and git inspection; unsupported commands such as curl are rejected.
- The architecture keeps classification agentic but makes notification deterministic: the model labels each item, while localpager applies the user's topic policy before sending Discord alerts.
- An audit loop used roughly 40,000 GPT-5.5 tokens per two-hour check, costing about 2–3 cents per run or approximately $9 per month at 12 runs per day.

**Tags:** [Architecture](https://yomu.fyi/topic/architecture), [LLMs](https://yomu.fyi/topic/llm), [Open Source](https://yomu.fyi/topic/open-source), [Performance](https://yomu.fyi/topic/performance)

- Source: [Hugging Face](https://huggingface.co/blog/local-models-pr-triage)
- Source URL: https://huggingface.co/blog/local-models-pr-triage
- Ingested by Yomu: 2026-08-29T01:50:53.102Z

[Read original post](https://huggingface.co/blog/local-models-pr-triage)