A 24×7 personal autonomous assistant you talk to directly on Telegram — equipped with multi-provider failover, Hermes cognition, two-way voice, and SSH pair-tunneling.
Engineered using the tiktok-emotion-first-editing method — high compression, zero dead time, and authentic product capture.
Why do other AI agents crash the second you close your laptop? The curiosity-gap hook exposes the flaw in standard DIY scripts.
Pair with built-in @init_smart_bot or custom @BotFather bot using auto-minted channel tokens.
Claude falls through to DeepSeek falls through to the Offline Reflex Tier. Remote SSH voice command executes in 4 seconds.
Click any sample command below to simulate how Deep-Init processes commands over Telegram in real time.
Not another toy prompt wrapper. Deep-Init is built to run continually with fail-safes at every layer.
Name your agent, assign directives, connect Telegram, and initialize. No editing YAML files or fighting environment variables.
Use the built-in bot (@init_smart_bot) or your own BotFather token. Channels auto-register and bind in 2 clicks.
Connect Anthropic, OpenAI, DeepSeek, Groq, or Ollama. Mid-task failures fall through gracefully. The offline reflex tier guarantees answers.
Toggle Hermes tool-use discipline and Moltis opinionated calculation cognition with a single switch. Or run hybrid cognition.
Hands-free conversation in the web console with natural neural voices. Telegram voice notes are transcribed and executed automatically.
Connect your physical Mac, Linux server, or VPS with one generated shell command. The agent operates your hardware from Telegram.
Every agent ships with a full web management portal locked behind cryptographic access tokens.
Real-time KPI cards, standing orders, active chat channels, and skill registries.
Toggle Hermes and Moltis cognition, configure strict tool-calling parsers.
100% open source. Deploy on Vercel, run locally with Bun or Node, and wire your Telegram in 5 minutes.