Tags: Lythaeon/cmdcode
Tags
v0.6.0 - Responses API tool-call streaming release Fixes broken tool-call streams on the openai-responses frontend so clients like OpenCode receive proper function call lifecycle events. Highlights: - Full Responses API tool-call lifecycle events (output_item.added/done, function_call_arguments.done) - Tool call deduplication in the Responses stream renderer - Incomplete-stream detection returning a proper 502 - Migrates the workspace to Rust edition 2024
v0.5.0 - OpenCode compatibility release Fixes duplicate tool call ID errors and context overflow detection so OpenCode's auto-compaction works seamlessly through the proxy. Highlights: - Request-side tool call deduplication - Response-side streaming deduplication - Context overflow error forwarding for auto-compaction - Duplicate finish event suppression
v0.4.0: provider-agnostic LLM gateway Five protocol frontends (OpenAI, Anthropic, Gemini, Responses API, Ollama-native) over multi-provider upstream routing (command-code, openai-compat, anthropic-native, gemini-native) with hot-reloadable providers.json, runtime enable/disable via connect CLI/TUI, taste injection on every path, Responses API session chaining, and a ~600ms per-request performance fix.