Your AI assistant.
Your machine.
A capable terminal assistant and a small async agent SDK. Work with local models, keep your data private, and build at your own pace.
ollama pull gpt-oss:20b && scripts/cortana
›
├─ tests/ — test suite
└─ pyproject.toml — dependencies
Everything you need to get moving.
From your first run to building your own agent, find your way around Cortana.
Introduction
Install Cortana, set up your workspace, and make your first agent.
Using the CLI
Learn the terminal UI, keyboard shortcuts, permission modes, and commands.
Core Concepts
Understand the agent loop, context, the knowledge graph, and learning.
Intelligence
Explore memory, subagents, planning, MCP, and background work.
Capabilities
Browse the web, use voice, generate media, and automate tasks.
Configuration
Tune your model, environment, and Cortana settings.
Everything runs on your machine.
A full assistant out of the box, built from parts you can use on their own.
Three local engines
Ollama, Docker Model Runner, or llama.cpp with automatic GGUF downloads.
Memory that lasts
Thread history, working memory, and semantic recall backed by Qdrant.
Subagents & planning
Delegate work to parallel subagents and verify results before answering.
MCP servers
Connect any MCP server; its tools load on demand in the background.
Voice in and out
Talk with Kokoro or Qwen3-TTS speech and Faster Whisper transcription.
Images & video
Generate and edit media locally, with a vision model checking the edits.
Real browser
Playwright browsing with accessibility snapshots small models can follow.
Git & your code
Safe git tools and a knowledge graph of your repository.
You stay in control
auto, plan and ask permission modes with diff previews before edits.
Schedules & heartbeat
Recurring tasks and a heartbeat that lets Cortana act on its own.
Skills
Drop in SKILL.md folders; Cortana finds and loads them when relevant.
An SDK underneath
Every piece is a small async Python API you can build your own agents on.