A personal assistant you reach from the terminal, the desktop, and Telegram — one agent, one memory, one conversation that follows you across all three.
Python 3.9+ · Linux · One-command setup
Installs the CLI, DeepAgent, Ollama for embeddings, and Qdrant for memory. Add your Sarvam and provider keys afterwards.
$ curl -fsSL https://raw.githubusercontent.com/astromanish/aishe-cli/main/setup.sh | bashSix core capabilities. No filler.
Terminal, desktop app, and Telegram share one thread. Start a thought on your phone, finish it at your desk.
Speak to it and hear the answer. Sarvam Saaras v3 in, Bulbul v3 out — 36 voices, 21 Indian languages.
mem0 + Qdrant, embeddings by Ollama. Searches before answering, and remembers across sessions.
Every message lands in one SQLite store with full-text search. Nothing fragments, nothing is lost.
The agent runs read-only commands on its own. Anything destructive is refused outright; anything consequential asks first.
One command checks every service, with STT, TTS and agent roundtrip tests included.
Nine tools, loaded by default. Add your own in tools.py.
AST-parsed arithmetic. No eval, no injection. Just math.
Memory and embeddings stay on your machine. Speech and inference go to the providers you choose — bring your own keys.
:8765
One agent behind all three surfaces — same memory, same conversation.
:8765
FastAPI backend. LangGraph agent with tool calling, streaming, and conversation memory.
:11434
Local embeddings (qwen3-embedding) and local models when you want to run fully offline.
:6333
Vector store for semantic memory. Collection aishe_memories.
cloud
Speech-to-text and text-to-speech. Saaras v3 and Bulbul v3, tuned for Indian languages.
DeepAgent · :8765
Ollama · :11434
Qdrant · :6333
Sarvam · cloud
Eleven subcommands. Type aishe to start a conversation.
aisheaishe "What is 15 * 4?"aishe desktopaishe telegram startaishe memory add "User prefers Hindi"aishe memory search "Hindi"aishe memory statusaishe search "aishe"aishe config set provider openrouteraishe gateway statusaishe doctoraishe statusOpen source under MIT. Clone it, run it, break it, improve it.
Get started