Introduction

Sparse is an AI assistant and coding agent that picks the right model for every task.

It sits between you and every AI model you have access to. Ask it a question and it answers. Point it at a project and it plans the work, turns each change into a reviewable diff, checks it, and learns what works on your repo. Behind the scenes it routes each step to the model that fits it: fast, cheap models for routine work, frontier models for the hard parts. You stop choosing models. You just build.

At a glance

• One assistant, every model. One OpenRouter key gives Sparse the whole catalog: Anthropic, OpenAI, Google, DeepSeek, Moonshot, Z.ai and more. Its own agent runs the work, so there’s no other tool to install. Prefer local? Point it at Ollama.
• Chat or code. Questions, emails, summaries and research get one well-routed answer. Coding tasks go through a full pipeline: plan, edit, validate, review.
• Routing you can inspect. Every model pick is journaled with its reason, and routing improves from the outcomes of your own runs.
• Spend where it matters. Renames, boilerplate and “where is X?” go to fast, cheap models. Deep reasoning, refactors and architecture get the frontier ones.

Three ways in

	What it is
Desktop app	The full UI: chat, plans, loops, sessions, cost tracking, images, a browser and a terminal. See The desktop app.
CLI	sparse for the terminal: an interactive REPL plus one-shot commands. See the CLI reference.
VS Code / Cursor	A side chat for the folder you have open. See VS Code and Cursor.

Sparse is also an MCP server, so any MCP host can use it as a set of tools.

Where to start

1. Install the desktop app or the CLI.
2. Follow the quick start.
3. Read how routing works and where edits land.