agents Releases

A major release (vLLM-Omni) featuring Day-0 support for Nvidia Cosmos 3, expanded quantization support, and production-ready audio/TTS serving. It includes significant architectural updates like OmniCoordinator integration.

Jun 6, 2026

A patch release focused on UX improvements and bug fixes for the Crush CLI, including a permission dialog key press delay and a fix to show all models in the `crush models` command. It also includes continued work on the experimental client-server architecture.

Jun 5, 2026

Focuses on workflow improvements, new model literals for LLM/STT/TTS, and worker protocol updates, along with fixes for avatar animations and inference request headers.

Jun 3, 2026

A maintenance release fixing bugs related to OAuth and startup crashes. It includes fixes for Hyper OAuth dialogs appearing mid-session and a rare startup crash if default models are missing.

Jun 2, 2026

A feature update for RustDesk including new Android deployment flows, privacy mode improvements, and various bug fixes across platforms.

Jun 2, 2026

A minor patch release for the React grab library adding a confirmation prompt during the skill agent-selection view installation process.

Jun 1, 2026

A minor patch release for the grab library, updating dependencies to match react-grab@0.1.44.

Jun 1, 2026

A patch release for the React grab CLI adding a confirmation prompt during the skill agent-selection view installation.

Jun 1, 2026

A patch release setting up automated npm releases via GitHub Actions using trusted publishing (OIDC).

May 31, 2026

Adds new STT providers like Cartesia ink-2 and SmallestAI Lightning v3.1 Pro, fixes OpenAI tool ordering and serialization, and updates various dependencies.

May 29, 2026

A Friday patch release addressing specific issues with the `bash` tool and OpenCode integration. It also includes continued fixes for the experimental client-server architecture.

May 29, 2026

A release adding MCP section to config web UI and various bug fixes related to MQTT, tool feedback, and documentation. Includes updates to dependencies like Go and React libraries.

May 29, 2026

This release focuses on enhancing the client-server architecture, improving notification capabilities across SSH, and fixing issues with various AI models and AWS Bedrock services.

May 28, 2026

Introduces new STT plugins (GnaniAI, Soniox), fixes voice race conditions, and adds avatar examples like LemonSlice with switchable personas.

May 27, 2026

Deprecates mcp_servers parameter, adds Perplexity Agent API (Responses) LLM support, and enhances OpenAI realtime capabilities including gpt-realtime-whisper STT.

May 21, 2026

Fixes Deepgram TTS websocket errors, adds Speechmatics to inference STT, and introduces a CLI command for asset fetching along with Rime coda model support.

May 18, 2026

Introduces Answering Machine Detection (AMD) and adds Perplexity LLM plugin, along with various fixes for ElevenLabs chunk scheduling and Google STT memory leaks.

May 13, 2026

This is a major production release rebased to vLLM v0.20.0, featuring large-scale serving improvements for Qwen3-Omni, expanded quantization coverage, TTS speedups, and Wan2.2 production readiness on NPU.

May 7, 2026

Focuses on AMD improvements, fixes for warm transfer fallback behavior and AWS stream handling, and adds Soniox TTS plugin support.

May 5, 2026

A maintenance release focusing on bug fixes for STT and TTS plugins, addition of new AI model support (GPT-5.4-mini, XAI), and improvements to metrics and session handling.

Apr 30, 2026