mnfst/llm-gateway manifest@6.28.1

manifest v6.28.1

on GitHub

🐛 Patch Changes

• 8f84cf6: Keep Claude prompt caching growing with the conversation on Amazon Bedrock and custom Anthropic providers, instead of stopping at the system prompt and tools.
• 13ddffd: Count Gemini thinking tokens as output tokens. Google reports them in thoughtsTokenCount, apart from candidatesTokenCount, so thinking models on Google routes were recorded and priced without them.
• 82e7c6a: Forward tool_choice to Gemini as toolConfig on Google routes. It was dropped, so a forced (required or named function) or disabled (none) tool call was left to the model's own choice.
• 5fe5150: Model parameters dialog now saves any value you set, even when it equals the provider default, and shows unset params as "Not set" so the client's value is used. Fixes a max_tokens set in the dashboard being ignored (#3022).