DigestAI news desk

Cut through the AI noise.

Agents & Tools6 min read

Oh My Pi updates Custom model configurations for version 18.2.7 and later

Recent updates to the Oh My Pi (omp) assistant tool require users to modify their custom model configurations in models.yml for version 18.2.7 and later. Users must rename custom providers from "local" to a unique name to avoid conflicts with omp's new built-in local provider for on-device models. Additionally, users running Qwen 3.8 models must add qwenTemplateReasoningEffort: true to the…

1 source

Key points

  • Users must rename custom local providers to avoid conflicts with the new built-in local provider in omp 18.2.7.
  • Qwen 3.8 configurations require the qwenTemplateReasoningEffort parameter to prevent thinking effort from defaulting to maximum.
  • An allowlist in config.yml can restrict omp to local models and prevent accidental fallbacks to paid hosted APIs.

The update also affects how omp handles inference servers like vLLM, llama.cpp, LM Studio, Ollama, and SGLang. To prevent omp from falling back to hosted models when local servers fail, users can implement an allowlist using the enabledModels setting in config.yml. A Python logging script can be used to debug and inspect the exact request payloads omp sends to these custom backends.

Model page: Claude Opus 5.5 →

Full story from doug.sh · by Doug Calobrisi · via Reddit AI communitiesOpen source ↗

Custom Models in Oh My Pi: vLLM, llama.cpp, SGLang and More

doug.sh · 23 September 2026

Loading the full article…

This text was published by doug.sh and written by Doug Calobrisi. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗

Coverage and discussion

1source
Topics · follow one to build your own front page

The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.

Comments

via GitHub Discussions

More in Agents & Tools

All →

Related stories