June is a personal AI assistant that runs entirely on your local machine — no subscriptions, no cloud, no payments.
"Hello, what are we working on today?"
Four pieces, each its own open-source component, assembled into one voice agent:
- The brain: Ollama running
qwen2.5:7b(orllama3.2) locally. Fast, private, free. - The mouth: backtalk — hold a key, speak, and June answers through your speakers a second later. Hearing and voice run on free local models (Whisper + Kokoro).
- The face: ai-visualizer — full-screen visualizer that idles, listens, thinks, and speaks in sync with the conversation.
- The memory: ai-memory-vault — persistent memory built on plain Markdown files in Obsidian.
- Windows 10/11 (64-bit)
- Ollama installed
- Python 3.11 (installed via
uvautomatically) - A microphone and speakers
Launch Ollama from the Start Menu. Wait for the llama icon to appear in the system tray.
ollama pull qwen2.5:7bThis downloads ~4.4 GB. Do it once. (Or use llama3.2 which is already installed).
$env:Path = "C:\Users\$env:USERNAME\.local\bin;$env:Path"
cd backtalk
uv sync --inexactDouble-click start.bat or run:
.\start.batHold the Home key (or say "go hands free") and speak. June answers through your speakers.
Say "goodbye June" to hang up.
These exact phrases, spoken alone, control the session:
Edit backtalk/backtalk.json:
{
"agent_dir": "d:/Personal/Projects/AI Trials/Friday",
"name": "June",
"model": "llama3.2:latest",
"deep_model": "qwen2.5:7b",
"ollama_url": "http://localhost:11434/v1",
"ptt_key": "home",
"voice": "bm_lewis",
"stt_model": "small.en",
"stt_device": "cpu"
}June's personality lives in AGENTS.md.
See TROUBLESHOOTING.md for common issues. If something breaks, open a chat and describe the problem — June is built to diagnose and fix her own stack.
Code components are licensed under the GNU Affero General Public License v3 or later (AGPL-3.0-or-later). See LICENSE.
Built by Akhil, powered by open-source tools.
