Aishe

A personal AI assistant that works from the terminal, the desktop, and Telegram — one agent, one memory, one conversation.

Not a chatbot bolted onto three frontends: all three surfaces talk to the same DeepAgent backend over the same conversation thread, so a message sent from your phone continues the exact conversation you were having in the terminal.

Surfaces

Surface	How you reach it
Terminal	aishe
Desktop	aishe desktop — Electron app: chat, voice, history, memory, keys
Telegram	aishe telegram start — long-polling bridge

This repository is the CLI. The desktop app is aishe-desktop.

Install

curl -fsSL https://raw.githubusercontent.com/astromanish/aishe-cli/main/setup.sh | bash

Usage

aishe                        # interactive streaming chat
aishe "What is 15 * 4?"      # one-shot streaming chat
aishe desktop                # open the desktop app
aishe status                 # service health
aishe doctor                 # full diagnostics
aishe start-all              # bring up every service

aishe memory add "fact"      # save a fact
aishe memory search "query"  # search memory
aishe memory status          # embedder / plugin / vector DB

aishe config                 # view config
aishe config set <key> <v>   # set a value

aishe telegram start|status|stop
aishe gateway start|restart|stop|status
aishe version

In the interactive loop, /exit or Ctrl+C quits.

One shared foundation

Everything resolves through aishe_pkg/aishe_core.py — where the backend lives, where the desktop app lives, where the sidecars live, and which services exist on which ports.

That module exists because the alternative already happened once: the CLI, start-all.sh and the desktop each hardcoded their own paths, and they drifted. Two divergent copies of DeepAgent ended up competing for :8765, with the winner decided by whichever tool launched first.

from aishe_pkg import aishe_core

aishe_core.deepagent_dir()     # where server.py is
aishe_core.desktop_dir()       # the Electron app
aishe_core.telegram_bot()      # the Telegram sidecar
aishe_core.SERVICES            # deepagent :8765, ollama :11434, qdrant :6333
aishe_core.describe()          # all of the above, with existence flags

Resolution order: AISHE_HOME override → the repo checkout (so an editable clone is authoritative) → ~/.local/share/aishe-cli. Set AISHE_HOME to relocate the whole tree at once.

Services

Service	Port	Health	Purpose
DeepAgent	8765	/health	the agent: tools, memory, model routing
Ollama	11434	/api/tags	local models, embeddings
Qdrant	6333	/readyz	vector DB for semantic memory

Voice is not a service. STT and TTS go straight to Sarvam's cloud API (Saaras v3 and Bulbul v3) — no local ports, no models to download.

aishe doctor and the desktop's service panel read the same registry, so they can't disagree about what's healthy.

Memory

Two layers:

• Vector — mem0 + Qdrant (aishe_memories), embeddings by qwen3-embedding:0.6b on Ollama. This is semantic search.
• Short-term — SQLite with FTS5 at ~/aishe/aishe.db, which is what aishe search queries.

Check both with aishe memory status.

Requirements

• Python 3.9+, requests, pyyaml, ffmpeg
• Ollama (for embeddings), and Qdrant for vector memory
• A Sarvam API key for voice
• A provider key for whichever chat model you use

Data

All data lives in ~/aishe:

• ~/aishe/aishe.db — conversations, FTS5-indexed
• ~/aishe/memory/ — short-term memory files

Config: ~/.config/aishe/config.yaml

Development

python tests/run.py              # all suites (config, threads, memory, core)
python tests/run.py config       # one suite
python tests/test_threads.py     # a suite directly

70 assertions across 4 suites. Tests run against a throwaway AISHE_HOME and never touch the real ~/aishe or the live mem0/Qdrant store.

The desktop app has its own suite: cd aishe-desktop && npm test.

License

MIT