Skip to content

AI Free 0.4.25 — Qwen/Hermes tool-call reliability

Latest
Latest

Choose a tag to compare

View all tags
@Staks-sor Staks-sor released this 20 Aug 18:56
v0.4.25
2ebd800

AI Free 0.4.25

AI Free 0.4.25 improves Qwen compatibility with Hermes and other OpenAI-compatible agents while continuing the shared-core cleanup.

Changes

  • Valid unfenced Qwen JSON tool calls are converted to OpenAI tool_calls instead of being returned as ordinary assistant prose.
  • The streaming bridge holds detected tool JSON until it can emit one unambiguous tool turn with the terminal finish_reason: "tool_calls" chunk.
  • Bare calls are accepted only when their names match tools supplied by the client, reducing false positives and invented-tool failures.
  • Slow Qwen generations can wait up to four minutes for first content and ten minutes overall while the idle timeout still detects stalled streams.
  • Long technical answers that merely discuss 401 Unauthorized no longer invalidate a healthy Qwen session.
  • Desktop, standalone API and VS Code load .env before the provider module graph.
  • Shared model catalog, localization, parser and memory utilities now live in packages/core with compatibility re-exports for Desktop and VS Code.
  • Desktop, VS Code and JetBrains/PyCharm versions are synchronized at 0.4.25.

Thanks to @firegoaway for the detailed Hermes diagnostics and the reference implementation shared in issue #21. Thanks to @dimazzq92 for the shared-core migration in PR #20.

Compatibility

  • Node.js 18 or newer.
  • macOS, Windows and Linux desktop/API runtime.
  • VS Code 1.75 or newer.
  • PyCharm 2025.2 or newer with JCEF and Node.js 18+.
Assets 4
Loading