KetanAI

A personal AI assistant that runs entirely on your machine — no cloud, no API keys, no subscriptions.

Built with Ollama for local LLM inference and a custom memory layer using ChromaDB for persistent long-term memory across all sessions.

 ██╗  ██╗███████╗████████╗ █████╗ ███╗   ██╗ █████╗ ██╗
 ██║ ██╔╝██╔════╝╚══██╔══╝██╔══██╗████╗  ██║██╔══██╗██║
 █████╔╝ █████╗     ██║   ███████║██╔██╗ ██║███████║██║
 ██╔═██╗ ██╔══╝     ██║   ██╔══██║██║╚██╗██║██╔══██║██║
 ██║  ██╗███████╗   ██║   ██║  ██║██║ ╚████║██║  ██║██║
 ╚═╝  ╚═╝╚══════╝   ╚═╝   ╚═╝  ╚═╝╚═╝  ╚═══╝╚═╝  ╚═╝╚═╝
                                               ketan.ai

---

Features

• 100% local — runs on your machine via Ollama, nothing leaves your device
• Long-term memory — facts are extracted and stored in ChromaDB, persists across all sessions forever
• Session history — every chat is saved, auto-named from your first message
• Streaming responses — tokens appear as they're generated
• Fast — async memory search, small context window, lightweight models
• Transparent — shows exactly what's happening before each response

---

Install

Requires macOS with Homebrew.

git clone git@github.com:Ketansuhaas/ketanai-my-local-personal-assistant.git
cd ketanai-my-local-personal-assistant
bash install.sh

Open a new terminal:

ketanai

install.sh handles everything: Ollama, models, Python venv, and shell alias.

---

Commands

Command	Description
/model [name]	Show current model or switch to another
/models	List all installed Ollama models
/memory [query]	Browse long-term memories
/remember <fact>	Explicitly store a fact
/forget <query>	Delete memories matching a query
/sessions	List past sessions (sorted by recent)
/load <id>	Resume a past session
/exit	Save session and quit

---

Adding models

ollama pull mistral
ollama pull llama3.2
ollama pull phi4
ollama pull qwen2.5:3b

Switch inside ketanai with /model <name>.

---

Architecture

You type a message
    │
    ├─ search memory (ChromaDB, 2s timeout, background thread)
    ├─ build prompt (system + memory facts + last 10 messages)
    ├─ stream response (Ollama, num_ctx=4096)
    └─ store new facts (background thread, non-blocking)

Storage — all data lives in ~/.ketanai/:

~/.ketanai/
  ├─ config.json      # active model, user settings
  ├─ chroma_db/       # long-term memory vector store
  └─ sessions/        # chat history, one JSON file per session

Two separate memories:

	Session history	Long-term memory
Storage	JSON files	ChromaDB vectors
Scope	One session	All sessions forever
Used for	Conversation context	System prompt injection
Cleared by	Starting a new session	/forget <query>

---

Stack

• Ollama — local LLM inference
• ChromaDB — vector store for memory
• nomic-embed-text — local embeddings
• Rich — terminal UI
• llama3.2:1b — default model (fast, 1.3GB)