Genesis roadmap
| Phase | Content | Status |
|---|---|---|
| 0 | Monorepo (Cargo + pnpm), Tauri 2 + React, CI | ✅ |
| 1 | Launcher, AI accounts (keychain), video editor (cut, join, trim, undo), automatic subtitles (local Whisper / OpenAI), smart export, MCP server, CLI | ✅ |
| 2 | AI Studio: image, video, voice and music with several providers (Gemini/Imagen/Veo, fal, Replicate, Runway, Luma, Stability, ElevenLabs), asset gallery, send to timeline; built-in LLM assistant (Claude, GPT, Gemini, Ollama) that uses Genesis's tools; MCP client for external servers; MCP over HTTP | ✅ |
| 3 | Image editor: WebGL layers, crop, adjustments, text; AI to remove backgrounds, inpainting, upscale and variations | ✅ |
| 4 | Audio: multitrack with waveform, fades, noise reduction, TTS voice-over, voice cloning, AI music; remove silences in audio and video | ✅ |
| 5 | Office: documents (DOCX/PDF/Markdown), spreadsheets with formulas (XLSX/CSV) and presentations (PPTX/PDF) with AI, generated natively and exposed via MCP | ✅ |
| 6 | Advanced video: multitrack, transitions, keyframes, vertical auto-reframe, smart-cut (re-encode only the edge GOPs), titles, speed and remove silences in the edit | ✅ |
| 7 | Distribution: plugin SDK (JS/TS and Python), signed auto-update, signed installers with FFmpeg and whisper.cpp included | ✅ |
| 8 | Local AI without accounts: text (llama.cpp), images (stable-diffusion.cpp), voices (Piper) and transcription (whisper.cpp) on your own computer; model catalog with resumable downloads and recommendations based on RAM; automatic use when no account is connected, also via MCP | ✅ |
| 9 | Automation: recipes that chain tools (visual editor, conditions, parallel loops) and are used as one more tool, also via MCP; batch processing; AI best moments for clips; document, presentation, spreadsheet and video templates (built-in and your own) | ✅ |
| 10 | Pro video: grading with 9 looks, .cube LUTs (3D and 1D), temperature and tint; green screen with eyedropper and spill removal; vignette, sharpen, blur and noise reduction; stabilization (two-pass vidstab); GPU preview with the same LUT as the export; English interface |
✅ |
| 11 | Smart Library: watched folders and incremental local indexing (SQLite FTS5); what's said in videos and audio with its moments, document text, colors and orientation; natural-language search (Spanish and English) with filters; search by meaning with a local embedding model; available via MCP | ✅ |
| 12 | Local vision: text models with vision (Qwen2.5-VL) in llama.cpp; ai.describe_image for photos and videos (description, keywords and visible text); the local Assistant understands images; the Library finds photos and videos by what they show |
✅ |
| 13 | Light, tailored AI: effort slider with 4 levels recommended for the computer (model, threads, context, vision detail); light models tuned per task (Qwen3 0.6B, SmolVLM 500M, EmbeddingGemma 300M) for background tasks; grammar-constrained JSON; import your own (fine-tuned) GGUF models for text, vision or search | ✅ |
| 14 | GPU acceleration: Vulkan builds of llama.cpp and stable-diffusion.cpp in the Windows and Linux installers (NVIDIA, AMD, Intel); GPU detection, automatic use, CPU fallback if the driver fails and a switch in Settings; local engines no longer keep running after closing Genesis | ✅ |
| 15 | Collaboration: shared spaces with end-to-end encryption (XChaCha20-Poly1305, per-space keys in the invite link); projects with their media, deduplicated chunks, versions, timed comments and live notices (long-poll); one-command self-hosted server (genesis share serve, admin key) or on the same network from the app |
✅ |
| 16 | Publish without certificates: "at your own risk" installers (ad-hoc signing on macOS so it opens on Apple Silicon, SmartScreen documented on Windows), generated release notes with the per-system notice, SHA256SUMS.txt, the install guide and a notice in About; updates are still signed with the free key |
✅ |
| 17 | Polish: genesis:// links open Genesis in Shared spaces with the invite filled in (also with the app already open: a single instance, which also opens files); space keys in the system keychain with automatic migration and a notice if there's no keychain; descriptions from the light vision model (English only) are translated with the local text model and the Library stores keywords in both languages |
✅ |
| 18 | Review from your phone: installable web page (PWA) served by the space server itself, with no app stores or developer accounts; decrypts in the browser (XChaCha20-Poly1305 with @noble, compatible with Rust via a shared vector) and the key never leaves the link's #; light encrypted preview when sharing (MP4 ≤720p, AAC or JPEG), comments at a second, approve versions (also from the app and share.approve), live notices and download of the original |
✅ |
| 19 | Real-time editing: video edits, documents and spreadsheets in a space are edited together (Yjs: the JSON↔CRDT binding merges per clip, cell and field; TipTap Collaboration in documents); WebSocket rooms on the server that only see encrypted changes (history that catches up on join and compaction with snapshots); presence (named cursors, each person's cell and clip), own undo, offline with resend on reconnect and media that uploads and downloads on its own | ✅ |
| 20 | Video and music with local AI: Wan 2.1 1.3B and Wan 2.2 5B (text or image → video) in stable-diffusion.cpp, converted to MP4 with FFmpeg, with the shared text encoder; MusicGen small on ONNX Runtime (included in the installers), with classifier-free guidance, the 4-codebook delay pattern and EnCodec; both in the AI Studio and via MCP without an account, with GPU-based recommendations | ✅ |
#How to add a tool
- Define the input with
#[derive(Deserialize, JsonSchema)]. - Declare the tool with
define_tool!in the right crate. - Register it in the crate's
registerfunction.
It will appear in the launcher (with an automatic form), in genesis tools and in MCP's tools/list.
To add tools without touching Genesis, write a plugin: see PLUGINS.md.