Stopguessing.Solvetherightproblems.
Spar watches how the attempt actually goes — where you stall, what you reach for first, which case breaks you — and sets the one problem worth your next hour.
LeetCode. Codeforces. Problems written for you. One rating, one history.
Spar, playing a recorded session: a learner writes an answer to “Count the Root’s Children”, runs it, submits, and fails a hidden case. They ask for something simpler, and the agent drafts, validates, repairs and publishes a new challenge, “Identify a Tree’s Root”.
Challenges are set in
Python
JavaC++CGoRustSwiftRubyA roadmap is an order somebody fixed once, for somebody who isn’t you. Spar picks your next problem from how your last attempts actually went.
Every attempt is evidence. Which case broke, on which input, after how many tries — kept, not graded and forgotten.
A map of what you can do. Concepts, each with the attempts behind it, marked where Spar still needs evidence.
A rating that explains itself. Every move says what caused it, with rough Codeforces and LeetCode equivalents.
What happens next
Your next problem is a decision.
Found where it already exists. LeetCode and Codeforces first, when a real problem already trains the thing.
Written when it doesn’t. A challenge for exactly your gap, with its own tests, in the language you train in.
Checked before you see it. A reference must pass every test and a plausible wrong answer must fail one. If not, it’s repaired.
The agent
A coach that writes the next problem.
- Talk to it like a coach.
- Ask for something simpler, a hint, or the case you keep missing.
- It changes the plan, not just the reply.
- Every new challenge is drafted, validated and published before it lands.
Bring your own model
Runs on the model you already have.
Sign in with a subscription, paste a key, or run one on your own machine.
Supported providers: OpenAI Codex, Claude Code, GitHub Copilot, OpenAI, Anthropic, Google, xAI, DeepSeek, Ollama, LM Studio, OpenRouter, Moonshot, MiniMax, Z.ai, Vercel AI Gateway, Cloudflare AI Gateway, Cline.
Yours, end to end.
Graded by tests. Your code runs against visible and hidden cases. No model decides you passed.
Every run kept. Passed, replaced or still open — the whole history, and why.
No telemetry. Keys in the system keychain. No analytics in the app or on this site.
Get in the ring.
Free, and the whole product is on GitHub.
v0.7.1 · macOS, Windows and Linux