CrossMeet v1.0.0 · First stable release

CrossMeet 1.0 is here

After six months of internal development, CrossMeet 1.0 stable ships today. This is our first answer to the question "what should real-time translation actually feel like?"

Core capabilities

• Bidirectional interpretation — microphone and system audio run separate ASR pipelines, with source and translated text presented independently
• Local + API ASR — Whisper-faster, Qwen3-ASR, and Nemotron-ASR, plus OpenAI-compatible, Aliyun, Volcengine, and Deepgram routes
• Language selection — automatic detection or manual selection; coverage depends on the configured ASR, translation, and TTS engines
• Glossary management — brand names, person names, technical terms across three categories with per-meeting group activation
• Knowledge base RAG — upload PDFs / DOCX, vector retrieval injects context into prompts so the LLM speaks your business
• Virtual microphone (vMic) — routes TTS output through a virtual microphone for software that accepts a selectable microphone input
• Compose Bar (simul-interpret) — type in your language, watch live translation preview, manually correct, then send (text + TTS together)
• Quick phrases — preset common phrases with one-click send + TTS playback
• History and export — save, search, and export session content; available formats follow the current client UI
• LAN API sharing — expose local engines (ASR / TTS / LLM) as OpenAI-compatible APIs to LAN devices
• Cloud platform presets — OpenAI, Anthropic, Aliyun, DeepSeek, SiliconFlow, Ollama, Gemini

Engineering principles

• Two iron rules — ASR source text always arrives before translation; LLM calls never block ASR output
• Hardware gating + opt-in — local features auto-disable when no NVIDIA GPU is detected; local models install on demand
• Configurable data paths — local engines process on device; API engines send the relevant stage's data to the configured service
• License v2 multi-device binding — Stripe / Alipay / WeChat Pay; 2–5 device slots, hot-swappable

Known limitations

• Windows SmartScreen may show a "Windows protected your PC" warning on first install — click "More info → Run anyway"
• Local Whisper-faster recommends 6 GB+ NVIDIA VRAM; users without a GPU can rely on cloud APIs
• Windows 10 and 11 are the currently supported public platforms; other platforms have no public release date

See Getting Started for setup walkthrough · All releases for version history.