- [new] Global defaults for
extra_headers,temperature,top_p,max_completion_tokens,max_retries,inference_idle_timeout_secs, andstream_tool_callscan now be set once under[models]so they apply to every model (built-in, discovered, or custom). Per-model values in[model.<id>]continue to take precedence. - [new] Global
extra_headersare merged into every inference request; a per-model header with the same name (case-insensitive) overrides the global value while other global headers are still inherited.
Under the hood
- Documentation for custom models and configuration was expanded with dedicated sections explaining the new global defaults and their interaction with per-model settings.
- Internal version metadata and bundled manifest updated to 0.2.78.
Upgrade notes
No manual steps required. Existing per-model configuration continues to work unchanged.