Ante Files

v0.2.12 - 2026-10-15

Added

• Remote MCP servers over Streamable HTTP: an mcpServers entry takes url in place of command, with an optional transport of stdio or streamable-http (http is a synonym). An entry that mixes command and url, or names the legacy sse transport, fails that server at warm-up with a message saying what to change. /mcp shows the URL, and rage redacts a URL's query string
• ante offline use <MODEL> takes --context <TOKENS>, --thinking on|off, --temperature <T>|default and --args "<llama.cpp options>" to save per-model load preferences; each flag replaces only its own value, they are validated at use, and every invocation ends with a Loads with: line
• Gemini 4 Argon in the native Gemini, Vertex and OpenRouter catalogs, with explicit low/medium/high thinking levels; Muse Wonder in the OpenRouter catalog

Changed

• The local provider is derived from the offline runtime state at the moment a model becomes ready, so a StartSession sent on the ready reply always finds it registered
• Headless runs and subagents are told that no one can answer them and the session ends with the turn
• llama.cpp bumped to b11540, picking up the router control-pipe fix and CUDA, Vulkan and Metal correctness fixes

Fixed

• A session that names a provider without a model now gets that provider's default model instead of the global default, which the named provider could not serve
• Read recognizes PNG and other images saved without an image extension (or as .bin) and shows them as images
• Sharded GGUF models selected by path or saved default are inspected the same way discovery does, reporting the full set; incomplete sets are rejected at resolution
• ante offline stop help and messages say what it does: it stops the home's router whether or not a session is using it

Wire

• McpServerInfo.url added: the endpoint of a remote MCP server; command is empty for such a server

Crates

• Public crates published as 0.2.12: ante-protocol-shape, ante-llm, ante-exec, ante-sdk