Seline is a local-first AI desktop application that brings together conversational AI, visual generation tools, vector search, and multi-channel connectivity in one place.
# Add to your Claude Code skills
git clone https://github.com/tercumantanumut/seline
Selene is a desktop app that runs AI agents on your machine. Connect them to your WhatsApp, Telegram, Slack, or Discord. Write code, generate images, build personal assistants. All from one place. Your data stays on your device.
Most AI apps work like this: you click a button, the AI responds. Selene flips that.
In Selene, your agent can do everything you can do, and the app follows along. When you ask your agent to "create a PR for this feature," it creates a workspace, copies the branch, writes the code, pushes, and opens the pull request. The UI automatically switches to show the git workspace as the agent works. You watch it happen, or step in whenever you want.
Every action the agent takes has a manual button too. You can create worktrees, stage files, push branches, and open PRs yourself. Same UI, same result. The difference is you don't have to. The app is built so the agent can operate it end-to-end, and you choose how much you want to steer.
This is how Selene develops itself. The app has been building its own codebase for weeks, running multi-hour sessions, managing parallel workspaces, creating its own PRs. 99% of the code you're reading was written by Selene agents.
AI agents are powerful but expensive. Most of that cost is context. Every turn, the model re-reads your files, conversation history, and tool definitions all over again.
Selene uses two agents instead of one. Your main agent handles the conversation. A smaller utility agent works in the background: it searches your files, finds what's relevant, and hands it over. The main agent never digs through thousands of files. It asks, gets the answer, and moves on.
No comments yet. Be the first to share your thoughts!
What this means for you:
Everything you need to write and ship code, built into one place.
Personal AI companions with personality.
Team agents for company workflows.
Connect your agent to your apps. Not through webhooks, as a native integration.
| Channel | Setup | What works | |---------|-------|------------| | WhatsApp | Scan a QR code | Messages, voice notes, attachments | | Telegram | Paste a bot token | Messages, voice bubbles, interactive buttons | | Slack | Socket Mode | Messages, files, native UI elements, threads | | Discord | Paste a bot token | Messages, threads, buttons, attachments |
Voice notes from WhatsApp and Telegram are automatically transcribed. When the agent replies with audio, Telegram users hear a native voice bubble. When the agent needs to ask a clarifying question, each app shows it in its own style: buttons, menus, or inline prompts.
Pair channels with the scheduler: set a task to run daily at 9am and have results delivered to your Telegram or Slack automatically.
Talk to your agent. Choose from cloud or local speech-to-text, including a local option that supports 32 languages with no API key.
Hear your agent. Three text-to-speech providers, including a free one that's always available. Clone a voice to make it unique. Long responses get summarized before speaking so you get the gist, not a monologue.
See your agent. Turn on the 3D avatar and your agent speaks with a moving face. Enable emotion detection (off by default) and it reacts: smiles, looks thoughtful, shows surprise.
Quick voice actions. After you speak, run one-click actions: fix grammar, rewrite professionally, summarize, or translate.
Images. Generate locally or through cloud providers. Pass in reference images for style matching, virtual try-on, or blending. Import custom image generation workflows and they become agent tools automatically.
Video. Turn session images into assembled videos with transitions, motion effects, and text overlays. AI plans the sequence, you get a rendered MP4.
Deep Research. Give the agent a question and it searches the web, reads sources, writes up findings with citations, and refines through multiple passes.
Your agents get better over time.
After conversations, Selene suggests memories like "user prefers TypeScript strict mode" or "always use pnpm." These show up on the memory page where you approve, edit, or remove them. Approved memories carry into every future conversation.
You can also just tell your agent: "remember that we deploy to Vercel," and it saves immediately.
Set tasks that run on their own: daily standups, weekly digests, code reviews, learning quizzes.
Skills. Reusable instructions your agent can follow. 37+ ship out of the box: deployment, dev tools, productivity, creative tasks. Create your own from the UI.
Plugins. Bundle multiple skills and tools together. Install from GitHub or a URL.
MCP. Connect externa