Menu

1 / 2

fix(agent): honor reasoning and auxiliary output-budget overrides by gcp · Pull Request #156198 · NousResearch/hermes-agent

GitHub Public Timeline Feed·gcp·about 10 hours ago
#hAmtqvn9
Reading 0:00
15s threshold

What does this PR do?

Explicit reasoning settings should be usable when an operator needs to correct an unsupported default. Title generation currently forces thinking off, title/approval output budgets cannot be overridden, and thinking-only continuations replace an explicit session effort with none .

Related Issue

Fixes #156196

Type of Change

Bug fix

Changes Made

Let configured title-task reasoning override the title caller's default disable, following the existing approval caller's approach. Generic explicit-caller precedence is preserved.

Add positive-integer auxiliary.title_generation.max_tokens and auxiliary.approval.max_tokens overrides. Defaults remain 512 and 16; explicit budgets reach supported routes, including max_output_tokens on the public OpenAI Responses API.…

Continue reading — create a free account

Join HashtagPLUS to read full articles, follow hashtags, vote, and join the conversation.