Private AI character chat — running entirely on your machine.
A free, open-source desktop app for immersive AI conversations. Run models locally for full privacy, or connect to OpenRouter and other remote APIs. Persistent memory, voice chat, a Realism Engine that makes every character feel alive — and now The Stoop, a community character hub built right in.
Two channels — pick your speed. Both are free, open-source, and install side-by-side without touching each other's data.
Latest release · v0.9.9.1.3 · tagged on main
Fresh builds every night from the Rawhide branch
Grab the newest Nightly rawhide.* pre-release · may have rough edges
A community character hub built right into the app. Browse featured & mod-picked cards, follow creators, and download straight into your library — no browser, no separate website, no copy-paste imports. Whole group casts travel too: members, lorebooks, and their pre-seeded Realism & Needs state all arrive intact, ready to play. Only the hub touches the network; everything else stays local.
Run AI models locally with KoboldCpp — your conversations never leave your machine. No account needed and no telemetry for local use; remote APIs and The Stoop are strictly opt-in.
Characters develop bonds, trust, and emotions across conversations. A dynamic relationship system that evolves with every interaction.
A community character hub built right in — browse, share, and download cards, including whole group casts with their Realism/Needs state. Open-source, local-first, opt-in, and 18+.
Full V2/V2.5 character card support. Import from SillyTavern, Chub.ai, or create from scratch with the built-in editor.
Full-duplex voice conversations with Whisper STT and Kokoro TTS. Push-to-talk or hands-free call mode.
RAG-based memory system remembers past conversations. Characters reference details from sessions days ago.
Multi-agent novel writing pipeline. Generate full stories with character arcs, world-building, and structured narratives.
Native desktop app for Windows, macOS, and Linux. Apple Silicon optimized. GPU acceleration via CUDA, Metal, Vulkan, or ROCm.
A full web app brings the desktop experience to any browser — or your phone over Tailscale — kept in sync with the desktop app.
Available on