Understand the code AI writes.
Verify it before it ships.
AI tools write code fast. OpenThunder shows you how your repository fits together, walks you through how each feature works, and checks every change: it runs the tests the change touches, flags risky edits, and points to the file and line behind every answer.
npx @skillstech/openthunder can-i-ship
Learning a repo and checking a change run on your machine, free, with no account and no AI key.
See a real agent run, start to finish: what the reviewer caught, what ran, and the verdict →
Study a codebase until you can explain it.
Open any repository: one you inherited, or one an AI wrote for you. OpenThunder traces it from the code itself, not the folder names, and turns it into something you can learn.
See how it fits together
Components, the routes that call them, and the tables they write, traced across frontend, backend, and database.
Understand a feature
Pick one thing a user does and follow it step by step to the file and line that runs it. Links OpenThunder could not trace stay labeled.
Test your understanding
Questions built from the traced code, so the right answer is always the code itself. Repo Mastery adds walkthroughs and flashcards.
When the code moves, flows you already studied are marked Changed since you reviewed it. Guide: understand a repo you didn’t write →
Keep the AI tools you like. Add the part they leave to you.
They write and explain code. OpenThunder shows what that code does to your system and whether the change is ready to merge.
Claude
Plan a change and write it with Claude Code. OpenThunder gives Claude a map of your repo over MCP, then checks what it changed and hands back a fix list it can act on.
Codex
Let Codex implement a task. OpenThunder writes the AGENTS.md it reads, runs the tests the change touches, and names what still needs a person, with file and line.
ChatGPT
Explore an idea or a bug in chat. Then ask OpenThunder, which answers from your actual files and cites them, and check the idea against your real tests.
Visual Studio Code
Where the code lives. The OpenThunder extension shows the verdict for your working tree in the status bar and puts findings next to the lines they are about.
Connect your agent: openthunder mcp install --client claude. Then have it check its own work every time it finishes: openthunder agent-check --install claude --fix. How it works →
Prefer one window? Build inside OpenThunder tooA terminal for your agents, AI tasks with an independent reviewer, overnight runs, and a built-in browser.
Your agents, your terminal, checked as they work.
Run Claude Code, Codex, or any agent in OpenThunder’s terminal, or describe a task and let OpenThunder run it. Either way, the work goes through the same Check before it ships.
AI tasks with a reviewer
Describe a change in plain language. One model builds it, a second one challenges it, your real tests run, and you get the evidence with the result.
Overnight runs
Queue tasks to run while you sleep. Wake up to changes that already went through a Check, with the evidence attached.
Built-in browser Preview
Open your running app, OpenThunder Cloud, and Repo Mastery next to your code and your agents.
Prove it works before anyone else has to.
Every change gets a Check: what changed, what it affects, what was proven, what is still unknown, and what to do next. One answer, the same in the app, the terminal, and the pull request.
Needs review: 1 thing needs a person
✓ Measured 14 impacted tests ran and passed
✓ Measured No secrets or unauthenticated routes added
? Not checked No test covers src/billing/refund.ts:48
! Needs a person This change touches payments: ask the billing owner
Next: send the fix list to your agent, or approve with proofRuns the tests the change touches
Only the test files tied to the changed code, so the answer comes in seconds. A test that could not run is Not checked, never a pass.
Finds what no test notices
OpenThunder can undo each changed block and re-run its tests. If nothing fails, it names the change no test protects.
Blocks only real problems
Failed tests, leaked secrets, routes that lost their login check, and broken behavior contracts block a merge. Style and refactoring ideas never do.
Sends a fix list back to your agent
Each problem comes with the file, the line, a command to reproduce it, and what should happen. Your agent fixes it, and the Check runs again.
Tells you what to test by hand
The routes and tables the change affects, ranked by risk, plus a sign-off that records who checked them.
Leaves proof on the pull request
Every claim says where it came from: measured, found by a scanner, stated by a person, or AI’s opinion, which never counts as proof.
Guide: verify an AI change → · Try it on a diff in your browser →
Explain it when someone pushes back.
Being able to explain a change is the last check. Take what you studied in OpenThunder into practice, and share only what you approve.
Explain Your Code
Practice explaining one feature or one change in Skills Tech Talk®, with an interviewer who asks follow-up questions. Only the context you approve is shared.
Show your evidence
Bring a Check’s proof to a design review or an interview: what you changed, what you tested, and what you still had a person look at.
Learn how AI tools work
Free lessons on agents, MCP, context, and reviewing AI-written code in Build with AI at Skills Tech Talk®.
How it comparesWhere OpenThunder fits next to Claude, Codex, ChatGPT, and Visual Studio Code, and what it is not.
AI made building faster. Ownership got harder.
The hard part is no longer generating code. It is understanding what changed, knowing what could break, and being able to explain the system after AI touched it. OpenThunder gets you from a guess to proof.
Use your AI tools for building. Use OpenThunder to own the result.
Each tool is great at its job. This is the part that stays with you.
| Tool | Great for | What you still have to do |
|---|---|---|
| Claude Code | Planning and writing changes in your repo | Know what the change affects, and prove it works |
| Codex | Running a coding task from start to finish | Review every line it changed before you merge |
| ChatGPT | Exploring ideas and explaining concepts | Check the answer against your real code and tests |
| Visual Studio Code | Editing, running, and debugging code | See the impact of an edit beyond the open file |
| OpenThunder | Learning the repo and checking each change | Works with all of them; it does not replace them |
Claude, Codex, ChatGPT, and Visual Studio Code are products of their respective companies. OpenThunder is not affiliated with them.
What OpenThunder is, and is not
- A way to learn a repository from its real code, and to check that you understood it
- A Check for every change: the tests it touches, the risks it adds, and what still needs a person
- The repo context your AI tools were missing (CLAUDE.md, AGENTS.md, Cursor rules, MCP)
- A record of proof you can hand to a reviewer, tied to the exact version it checked
- Not a replacement for your editor or coding agent: it works beside them
- Not a generic chatbot: answers point to files and lines, and an AI’s opinion is labeled as one
- Not an auto-merge bot: OpenThunder never merges for you, and a person approves every change
- Not a promise that software is safe: it says what was proven and what is not known
Download
One file. Double-click and run it like any desktop app.
Verify this release yourself
OpenThunder is built with AI and verified with evidence, and that includes its own releases. Each desktop release publishes its Evidence Pack alongside the installers: SHA-256 checksums for every artifact, a full CycloneDX SBOM (every resolved dependency with its integrity digest), and SLSA-style provenance recording the source commit and the CI workflow that built them. Don’t take the download on trust; check it:
cd ~/Downloads
curl -fsSLO https://openthunder.dev/releases/OpenThunder-mac-universal.dmg
curl -fsSLO https://openthunder.dev/releases/evidence/SHA256SUMS
shasum -a 256 -c SHA256SUMS --ignore-missing
Re-download the installer so you verify the current build (an older copy in Downloads will not match).
OK means the file on your disk is byte-for-byte the build this evidence describes.
With OpenThunder installed you can verify the whole manifest instead:
openthunder verify-artifacts checksums.json --dir .
Want more proof? See the engine run in public against well-known open-source repos in the Evidence Gallery, or read how Evidence Packs work.
How it works
No setup required to start. Add test commands, team rules, or cloud sync only when you want them.
-
1. Open your project
Point OpenThunder at any folder. It maps the architecture from the code and highlights the risky parts.
-
2. Learn how it works
Follow a feature step by step, ask "how does login work here?" and get answers that cite your files, then test yourself.
-
3. Build with your AI tool
Use Claude Code, Codex, Cursor, or OpenThunder’s own AI tasks. Connected agents can ask OpenThunder what a change affects.
-
4. Check the change
Get one answer, Ready to merge, Needs review, or Blocked, always with the reason and the proof behind it.
Local-first by default. Cloud only by choice.
Source code is not uploaded by default. API keys stay on your machine. OpenThunder does not force-push or auto-merge. Cloud sync and repo training are optional and opt-in. See exactly what happens to your code →
Runs locally
Repo analysis happens on your computer. Your source code is not uploaded by default.
No account required
Open a repo and start offline. Sign in only if you want cloud sync or repo training.
Your keys stay yours
API keys live on your machine. They never touch OpenThunder servers, logs, or your repo.
Your git stays yours
OpenThunder never force-pushes or auto-merges to your main branch, and warns before working on a dirty tree.
Repo Mastery, includedWalkthroughs, flashcards, and a mastery score for any repo, built into OpenThunder.
Master any repo. Repo Mastery, included.
Repo Mastery is built into OpenThunder on Desktop, Cloud, and the web. It reuses the same analysis, so there is no second scan and no separate account.
- A plain-English repo summary
- Architecture walkthroughs
- Tech stack explanations
- Critical flow breakdowns
- Flashcards
- Multiple-choice questions
- Interview questions
- A repo mastery score
Recognizing the right answer is not the same as mastering the code. Practice explaining it to find out.
Guides and how-tos
Short, practical walk-throughs.
Open your first repo
Install OpenThunder, point it at a folder, and read the map of your code.
Understand a repo you didn’t write
Map the architecture, get a grounded brief, and catch up on what changed.
Connect Claude Code, Codex, or Cursor
Give your agent the repo map over MCP, so it knows what a change affects.
Verify an AI change
Run a Check, read the result, and send the fix list back to your agent.
See the risks before you ship
Leaked secrets, unauthenticated routes, and risky dependencies, found on your machine.
Use OpenThunder in your editor
The verdict in the status bar and findings next to the code they are about.
Releases and updates
OpenThunder keeps itself up to date. Here is what is in each build.
- Loading release notes… See them as a file.
OpenThunder Desktop is free. Teams are the upgrade.
Desktop and Repo Mastery are free for individuals: learn, check, and own any repo on your own machine. OpenThunder Cloud adds shared repos, team rules, and proof across your organization. Skills Tech Talk® is the connected practice product for interviews and certification.
Stop shipping code you cannot explain.
Learn how your repository works, build with the AI tools you like, and check every change before it ships.
Understand it. Verify it. Own it.