Ollama: Cleaning Up the Agent Loop and Widening Model Support
Ollama's team shipped new Laguna model support and an agent skills system, while trimming redundant work out of the TUI event loop and tightening streaming correctness for both OpenAI and Anthropic-compatible endpoints.
Duration: PT2M56S
Episode overview
This episode is a short developer briefing from Ollama.
It explains recent repository work in plain language.
- Show: Ollama
- Published: 2026-07-18T13:00:46Z
- Audio duration: PT2M56S
Transcript excerpt
This excerpt keeps the crawler page concise. Listen to the episode or use the RSS feed for the full update.
Good morning. It's July 18th, 2026, and here's what moved in Ollama's codebase.
The clearest pattern today is cleanup and correctness in the agent and API layers, running alongside a push into new model territory.
Start with the agent TUI. Parth Sareen opened two nearly identical pull requests, 17240 and 17241, both removing calls to a synchronous context-window refresh function that was blocking the interface's update loop. That function was firing on every tool start, tool finish, and run completion — even though the…
Second theme: API streaming correctness. Eva H's PR 17225 fixes an Anthropic streaming bug where a text block wasn't closed before a thinking block started, producing invalid stream output for models like Gemma 4 that emit text before reasoning. Separately, PR 17239 addresses the OpenAI-compatible responses…
Third, model and platform groundwork. Dan Hiltgen's pair of PRs, 17237 and 17238, add Laguna MLX support, including custom kernels shared across mixture-of-expert models, aimed at closing the prompt-performance gap with llama-server. He also bumped the Linux build toolchain to GCC 13, fixing crashes in AMX code on…
Smaller but practical: PR 17245 routes the bare…
Nearby episodes from Ollama
- Agent Package Gets a Major Consolidation
- Agent Hardening and a Path Traversal Fix
- Resource Leaks and GPU Placement Get a Clean-Up Pass
- Cleaning Up Memory Planning and the Launch Experience
- Fixing What Defaults Got Wrong
- Weekly Recap - Model Correctness and Runtime Hardening
- Tightening Up the Serving Layer
- Thinking Model Output Is Getting a Real Fix