Ollama: Tool Call Ordering Fixes and Decision Models
Parallel tool results misattributed by position get ordering fixes in the OpenAI layer, while decision models gain structured criteria and Strands support. A Windows manifest path fix for registry hosts with ports and a Cogito streaming parser correction round out reliability work.
Duration: PT2M20S
Episode overview
This episode is a short developer briefing from Ollama.
It explains recent repository work in plain language.
- Show: Ollama
- Published: 2026-10-03T13:01:03Z
- Audio duration: PT2M20S
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 October 3rd, 2026.
The headline is reliability for agent workflows, with parallel tool results being attributed to the wrong call, plus steady expansion of decision models.
First, tool calling correctness. Two related proposals, PR 18763 and PR 18767, both address the same report where tool results arriving out of order were paired by position instead of by tool call I D. The fix normalizes complete groups of results to follow the assistant's call order, but only when every result…
That same theme shows up in streaming. PR 18759 fixes the Cogito parser where a chunk boundary split inside a tool tag caused the partial tag to leak out as plain text and the whole call to be lost. The parser now holds back a trailing partial tag until the next chunk confirms it. If you've seen intermittent missing…
Second, decision models are getting more expressive. PR 18768 accepts structured criteria — strings, objects, and arrays — for choice, no-op, and score questions while preserving null semantics, and updates the scoring schema and guide. That pairs with work on PR 18755 to separate decision preparation from the…
Finally, a long-standing Windows fix: PR 18771 encodes colons in…