Download Tiny Agent
Free for personal use. This is an early test build, so expect rough edges.
Setup
Install Ollama
Download it from ollama.com. It runs the model on your computer.
Install and open Tiny Agent
Open Model & speed, choose a recommended model and click Install. Download progress appears in the app; you do not need a terminal.
Choose the model
Qwen 3.5 4B is the balanced small option. On a 16 GB laptop, Qwen 3.5 9B is the strongest practical choice. Tiny Agent shows the 35B-A3B model and its 24 GB download, but guards it behind a conservative 32 GB RAM tier.
About the Windows warning
Test builds aren't code-signed yet, so Windows SmartScreen may say it "protected your PC". Choose More info → Run anyway only if you downloaded the installer from this site.
Your data
Your local model, notes, tasks and saved history stay on your computer. Optional Realtime voice sends session audio and canvas to OpenAI with separate API billing. Spark with local speech sends text and canvas, not audio, through your Codex ChatGPT sign-in and uses your Spark allowance. Local Whisper setup needs Python 3.12 and downloads a speech model plus runtime libraries. Live sketch uses local speech and your local model only, and its sketches and scope record stay on your computer. Web search contacts its configured provider. Sharing with us is off by default. If you'd like to help improve Tiny Agent, see Contribute.