Full Changelog: v0.3.5...v0.3.6

Full Changelog: v0.3.4...v0.3.5

Full Changelog: v0.3.3...v0.3.4

What's new

• Add explicit prompt cache breakpoints (llm.Part.CacheBreakpoint) across Anthropic, Bedrock, and OpenAI Responses/Codex adapters.
• Add built-in catalog metadata and preflight reasoning controls (ThinkingLevelsOnly) for Gemini 3.8 Flash on Google and Vertex AI.
• Report raw usage counter decreases properly in Gemini streaming responses.
• Fix OpenAI Responses wire serialization for explicit prompt cache breakpoints and retention options.

Full Changelog: v0.3.2...v0.3.3

Fix Gemini tool-call continuation by omitting unsigned empty response parts while preserving signed replay state.

What's new

• Add built-in OpenAI and OpenAI Codex catalog entries for GPT-6 Astra.
• Add built-in Anthropic catalog metadata for Claude Fable 5.1.
• Refresh catalog snapshot coverage for Claude Fable 5 and 5.1.

Full Changelog: v0.3.0...v0.3.1

What's new

• Add Azure OpenAI Responses, Mistral Conversations, Google Vertex AI, and Amazon Bedrock ConverseStream adapters.
• Expand protocol factories and the built-in catalog with Mistral, Vertex AI, and Amazon Bedrock model metadata.
• Add context-overflow detection, tolerant streaming JSON parsing, and coercing JSON Schema tool-argument validation.
• Add environment credential discovery for API keys, Vertex ADC hints, and Amazon Bedrock credential sources.

Full Changelog: v0.2.1...v0.3.0

Full Changelog: v0.2.0...v0.2.1

What's new

• Add deferred tool loading and provider-native tool discovery replay for OpenAI Responses, Codex, and Anthropic.
• Add durable OpenAI Responses background jobs with start, fetch, and cancel handles.
• Normalize tool-call IDs and repair missing, orphaned, or mismatched tool results during history transformation.
• Preserve background capabilities through retry and pricing wrappers.

Full Changelog: v0.1.0...v0.2.0

Full Changelog: https://github.com/XiaoConstantine/llm-go/commits/v0.1.0