Ollama: Tightening Up the OpenAI Compatibility Layer
Today's activity centers on hardening the Responses API and Gemma 4 tool-call parsing, plus a wave of documentation cleanup. Together they reduce edge-case failures for agentic coding workflows and improve doc accuracy across install guides.
Duration: PT2M30S
Episode overview
This episode is a short developer briefing from Ollama.
It explains recent repository work in plain language.
- Show: Ollama
- Published: 2026-09-08T13:01:18Z
- Audio duration: PT2M30S
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 September 8th, and here's your Ollama development briefing.
The clear signal today: Ollama is closing gaps in how it talks to coding agents like Codex, and how it parses tool calls from Gemma 4.
Start with the Responses API. Three separate PRs fix real breakage reported by users switching between Codex and OpenAI models. PR eighteen-two-nine-six corrects tool search call IDs to use the prefix Codex actually expects. PR eighteen-two-nine-eight teaches the API to accept agent message input items, which Codex…
Second theme: Gemma 4 parsing is getting more robust. Contributor mann1x filed two related fixes — PR eighteen-three-oh-seven and eighteen-two-eight-eight — both about the thinking channel and tool-call tags getting confused, which could cause tool calls to be silently swallowed as reasoning text, or control tokens…
A few other notable fixes: PR eighteen-two-nine-one closes an integer overflow in tensor size calculations that a crafted file could exploit. PR eighteen-three-oh-two avoids sending an unsupported flag to llama-server in tensor split mode. And PR eighteen-three-oh-one loosens the model name part-length limit, fixing…
Rounding it out, five…
Nearby episodes from Ollama
- Closing the Gaps Between What You Configure and What Actually Runs
- Weekly Recap - MLX Hardens as a First-Class Engine
- Context Handling Gets Serious
- Codex Integration Push and Schema Hardening
- Structured Output Meets Speculative Decoding
- Hardening the MLX Engine and Fixing What Feeds the Scheduler
- MLX Grows Up, Model Intent Gets Respected
- Runner Reliability and a Documentation Cleanup Wave