Free trial
The first 3 submissions are scored for free on kodwai's own key, with no payment details mentioned.
Learn morekodwai
Practice directing your own AI coding agent on real challenges, then get scored
Intermediate
What is kodwai?
kodwai is a coding tool. kodwai lets developers solve real coding challenges locally with Claude Code, Cursor or Codex, then scores how well they directed the agent across Direction, Outcome and Lift, with weekly leagues and public profiles.
kodwai is a practice platform for agentic coding, the habit of building software by directing an AI agent instead of typing every line. Developers pick a challenge, solve it on their own machine with Claude Code, Cursor or Codex, and receive a score out of 100 that reflects how well they steered the agent rather than whether the tests happen to pass. The workflow runs through a CLI. One command downloads the problem, starter files and tests and starts a timer. The developer then works as usual in their own editor: planning, prompting, running tests and pushing back when the agent goes wrong. A second command submits the code, commits, test runs and agent transcript. The score splits into Direction (how the work was steered, checked and broken down), Outcome (tests, code quality or the challenge's own rubric) and Lift (hidden traps caught and improvement over a solo AI). Direction carries the most weight. Each scored run also lists up to six key moments, labelled in a chess style from brilliant to miss, with evidence from the session. The catalog holds 15 challenges across backend, frontend, data, AI, distributed systems and other areas, graded easy, medium and hard, with time limits from 60 to 150 minutes. Each hides traps, meaning requirements that are easy to miss when a spec is pasted straight into an agent. Weekly leagues group players, promote and demote them by points weighted by difficulty, and a Direction Rating, an Elo-style number, sets a tier from Bronze to Grandmaster. Public profiles and shareable score cards sit on top. Positioned against algorithm-practice sites, kodwai targets a different skill: ticket-sized project work with an AI agent, useful for candidates preparing for AI-enabled interview rounds. A separate hiring offering lets teams run private interviews on their own problem with a rubric and full transcripts. Developers get the first three scored submissions on the platform's key, then supply their own Anthropic API key for further runs.
How do you use kodwai?
1Choose a challenge
Browse the catalog by area and difficulty. The easy Bookshelf REST API is the suggested first pick.
2Start from the terminal
Run the npx @kodwai/cli challenge command with the challenge slug. The CLI fetches the problem, starter files and tests, and starts the timer.
3Solve with your agent
Work in your own editor with Claude Code, Cursor or Codex. Write a spec, add tests and push back when the agent misses requirements.4Submit your run
Run the CLI submit command to send code, commits, test runs and the agent transcript for scoring.5Review your score and league
Open the web app to see Direction, Outcome and Lift, the key moments, and your weekly league standing and profile.
Pros and cons
Pros
- Scores the quality of agent direction, not just passing tests, which matches how AI-assisted work really happensAI
- Runs on your own machine with the agent you already use, so there is no browser sandbox to adapt toAI
- Hidden traps in each challenge reward careful specs and verification over one-shot promptingAI
- Moment-by-moment feedback with evidence shows exactly what raised or lowered a scoreAI
- Free for developers, with the first three scored submissions covered by the platformAI
Cons
- After three free submissions, unlimited runs require supplying your own Anthropic API keyAI
- Only three agents are supported (Claude Code, Cursor, Codex), so other assistants are left outAI
- The catalog is small at 15 challenges, which limits variety for frequent practiceAI
- Scoring depends on submitting your code and full agent transcript, which some users may not want to shareAI
- Challenges take 60 to 150 minutes, a real time commitment for casual practiceAI
How much does kodwai cost?
Pricing
Free for developers. The first 3 submissions are scored on the platform's key; after that, users add their own Anthropic API key in Settings for unlimited runs. Hiring team pricing is not stated on the homepage.
Learn moreSupport
Support is by email at hakan@kodwai.com, and the homepage includes an FAQ covering scoring, agents, pricing and leagues.
Integrations
Works with Claude Code, Cursor and Codex, and runs through the @kodwai/cli npm package. Scoring uses an Anthropic API key after the first 3 submissions.
Features
Local challenge runs via CLI, scoring on Direction, Outcome and Lift, hidden traps, key-moment feedback, 15 challenges, weekly leagues, Direction Rating tiers, public profiles, badges, share cards, and private interviews for hiring teams.
Learn moreFrequently asked questions about kodwai
How much does kodwai cost?
Free for developers. The first 3 submissions are scored on the platform's key; after that, users add their own Anthropic API key in Settings for unlimited runs. Hiring team pricing is not stated on the homepage.Does kodwai offer a free trial?
The first 3 submissions are scored for free on kodwai's own key, with no payment details mentioned.How do you use kodwai?
The walkthrough on this page covers 5 steps: 1. Choose a challenge 2. Start from the terminal 3. Solve with your agent 4. Submit your run 5. Review your score and league.What platforms does kodwai support?
kodwai is available on Web App, MacOS, Windows and Linux.What does kodwai integrate with?
Works with Claude Code, Cursor and Codex, and runs through the @kodwai/cli npm package. Scoring uses an Anthropic API key after the first 3 submissions.What are the limitations of kodwai?
After three free submissions, unlimited runs require supplying your own Anthropic API key. Only three agents are supported (Claude Code, Cursor, Codex), so other assistants are left out. The catalog is small at 15 challenges, which limits variety for frequent practice.
