Ollama: Legacy Model Cleanup and Agent Launcher Polish
Ollama is retiring legacy model compatibility shims in favor of on-load conversion, while tightening its launch experience with Claude Code and Codex. Inference fixes for reasoning tags, tool parsing, and raw output round out the activity.
Duration: PT2M28S
Episode overview
This episode is a short developer briefing from Ollama.
It explains recent repository work in plain language.
- Show: Ollama
- Published: 2026-10-09T13:00:32Z
- Audio duration: PT2M28S
Transcript excerpt
This excerpt keeps the crawler page concise. Listen to the episode or use the RSS feed for the full update.
Hello, it's Friday, October 9th, 2026, and this is your Ollama briefing.
The lead pattern is cleanup and compatibility: the project is removing old workarounds while making local models behave better inside external agents.
First, legacy model handling. PR 18882 removes the private llama-dot-cpp compatibility patches and migrates older model files on load, with atomic manifest replacement and garbage collection of unreferenced blobs during startup. That is the Phase Two store retirement. The merged companion, PR 18874, hides the…
Second, the agent launcher. Two small merged fixes landed, captured in commits 531946e and c2b7368 and b061384: one sets Claude Code's maximum context setting to the model's actual context length to avoid compacting too often on large-context models, and another sets a default service tier to silence inherited…
Third, inference correctness. PR 18877 stops Mistral bracketed think tags from leaking reasoning into answers, PR 18872 flushes the Qwen 3 point 5 tool parser on an empty final chunk so trailing tool text isn't dropped, and PR 18891 preserves retained reasoning through compaction replay. PR 18881 also fixes raw…
What's next: watch conversion…
Nearby episodes from Ollama
- Native Decision Models and Precision Fixes
- Weekly Recap - Decision Models Go Native, Embeddings Go Multimodal
- GPU Discovery Fallback and Docs Cleanup
- OpenAI Parity and Reasoning Fixes
- Memory Accuracy, Tool Fixes, and Simpler Onboarding
- Multimodal Embeddings and Cloud Convergence
- MLX Performance Push and Tool-Call Reliability
- Tokenizer Parity and RC-Aware Updates