Models and Providers
🏪 A curated model store. Memo detects your GPU and RAM and tags every download with a hardware-fit badge ("Fits your device", "Runs (CPU)", "Too large"). Plain-language quality labels replace quant codes, with capability filters and a one-click Start that configures GPU offloading. ⚡ Local models, faster. Memo asks each bundled engine which GPUs it sees and picks accordingly. With a Vulkan driver, prompt processing was about 2.2× faster in our test, and code-heavy replies up to 3× faster with n-gram speculative decoding. Measured on one model and one machine; Linux for now (details). |
|
- 18 provider types, one interface: OpenAI, Claude, Gemini, Grok, Groq, OpenRouter, Ollama, bundled
llama.cpp, OpenAI- and Anthropic-compatible custom endpoints, OpenCode Zen and Go, Kilo Code, Cline, plus Subscriptions: sign in once with an Antigravity, Claude or Codex account and every model it offers appears in the selector,/modeland the local/v1gateway, with no API key. Auto-fallback, auto-disable after 3 errors, live switching mid-chat, keys encrypted with AES-256-GCM. - 🖥 Coding agents as providers (Beta): point a chat at your installed Claude Code or Codex CLI. ChatGPT and claude.ai (Beta) can be driven through a real browser window instead of spending API quota.
- 🎵 Orchestra: a Chief model splits a complex task across specialist roles (planner, frontend, backend, bug fixer, reviewer, security, DevOps, generalist) and mixes providers freely.
- 🎤 Live Mode v2: native audio-to-audio conversation through Google Live or OpenAI Realtime, with barge-in and memory recall; falls back to on-device whisper.cpp plus local Piper TTS.