Build exactly what your business signed off.
nTra is where business, product and engineering take a requirement from conversation to merged code. The business signs off the spec; AI agents build only approved work and can’t drop, reword or fake a requirement; your engineers merge it. You get what was agreed, without the rework.
- Naaply
- Scriblo
- Verlo
Naaply
Scriblo
VerloVibe coding gets you a demo. nTra gets you to production.
Requirements live in meetings and documents. AI tools work from prompts. Things get lost in between, so code drifts from what was agreed: fine for a demo, not for something you ship.
- Manager approves
- Email on decision
- Max 5 days carry-over
- HR sets allowance
- Manager approves
- Email on decision
- Max 5 days carry-over
- HR sets allowance
- Manager approves
- Email on decision
- Max 5 days carry-over
- HR sets allowance
- Manager approves
- Email on decision
- Max 5 days carry-over
- HR sets allowance
- Unlimited carry-over
nTra keeps every detail through every hop, and every task points back to it.
Four steps. Four sign-offs: brief, design, plan, merge.
Describe it. nTra asks what people forget.
Talk it through or import your documents, tickets and code. nTra asks the open questions and builds the brief as you go.
nTra works through 300 questions, flags security and personal data before you mention them, and never overwrites an answer you gave.
Sign-off: the briefEvery requirement becomes approved work.
nTra turns the approved brief into the requirements document, the design and a plan of epics and tasks. The design is approved at a saved version; change it and it needs approval again. Before any task is written, nTra checks every requirement is covered.
Every line of the requirements document cites a source, and nTra may not add a requirement nobody asked for. Design approval is pinned to the version you saw.
Sign-off: the design and each epicEach task says what done means.
Acceptance criteria in plain Given, When, Then, what is out of scope, and the decisions already made. Tied back to the requirement it came from.
Vague criteria are rejected before anyone sees them. The task doctor finds requirements with no task yet, and a person releases work to engineers.
You get: tasks an agent can finishAI builds it. Your engineers merge it.
The agent your team chooses runs every step and logs every action it takes. A second AI reviews the change, and nothing merges until one of your engineers approves. Every pull request names the task and the requirement it came from.
nTra re-runs the agent’s checks itself. Secrets in a diff block the commit. The finish is locked: open PR, your engineer approves, CI, merge. Never a push to main.
Sign-off: review and merge
A person approves the brief
Your engineers stay in control.
- The re-run decidesnTra doesn’t trust the agent’s report. It re-runs every check the agent says it ran, and the re-run decides the gate.
- Claims are judgedEach acceptance criterion needs proof. A second model is asked which “proven” claims the code doesn’t support, and UI work needs a screenshot.
- A locked finishCheck, commit, push, open PR, read review comments, wait for your engineer, CI, merge. No flow can remove a step. Never a force-push, never main.
- Done means a person said soA task closes only when an engineer confirms it: PR merged, branch deleted, each criterion ticked. Your agents, Claude Code, Codex, Gemini, Copilot, Kiro or Antigravity, on your machines or in a throwaway cloud machine.
$ ntra watch watching 2 linked repos for runs started from the board: github.com/acme/claims-portal ~/code/claims-portal github.com/acme/claims-api ~/code/claims-api one run at a time · Ctrl+C to stop → picked up TASK-42 · Claim detail is read-only agent: claude-code · plan · build · test ✓ acceptance evidence attached · PR opened waiting for review
Your business runs on systems that already exist. nTra starts there.
Atlas is what every agent reads before it touches code: your repositories, documents and tickets indexed into one map. The run log says when an agent didn’t ask.
- One map of your projectCode, modules, dependencies, documents, decisions and past tasks, kept current by every run.
- Docs vs code, decided by a personWhere your documents and your code disagree, nTra shows both sides and someone decides. The answer becomes the truth every future task reads.
- Nothing unconfirmed spreadsWhat an agent learns is recorded with its source, and reaches another agent only after a person confirms it.
- Ask it, and get the trailWho changed this file, why, and in which pull request: the task and the PR, not a guess.
Business teams move faster. Engineering keeps the keys.
Your code stays yours
Changes arrive as pull requests in your repositories, and you own all output.
Not used for training
Your content is not used to train AI models, and our AI providers are contractually bound to the same.
You choose where it runs
On your own machines through the nTra CLI, or in a throwaway cloud machine per run that holds no standing secrets.
Approvals on record
Who signed off the brief, the documents, the design and the merge, and when. Every read of your knowledge is audited before it is answered.
Enterprise controls
SSO, SCIM provisioning and enforced two-factor on the Enterprise plan. SOC 2 report as an add-on.
Questions leaders ask first.
Anything else, ask us on a call. We’ll answer it on one of your own features.
Do I need engineers to use nTra?
Yes. Business and product people describe the work and sign off the spec. Engineers review and merge the code that AI agents produce. nTra makes that hand-off clear, it does not remove it.
Who owns the code?
You do. Everything nTra produces for you is yours, and it lands as pull requests in your own repositories.
What happens if the AI gets something wrong?
It gets caught before it ships. Every task has acceptance criteria the agent can’t change; nTra re-runs the checks itself; a second model is asked which claims the code doesn’t prove; secrets block the commit; and nothing merges until one of your engineers approves.
Where does our code run, and who can see it?
On your own machines through the nTra CLI, or in a separate cloud machine for each run. Your content is not used to train AI models.
Which AI tools does it work with?
Claude Code, Codex, Gemini, Copilot, Kiro and Antigravity. You can also use your own Anthropic key.
How is it priced?
Launch is $59 a month with 500 credits; Build is $79 a month with 1,000 credits. Yearly plans cost at least 10% less. Enterprise terms are agreed with us.
See it on one of your own features.
Don’t vibe code it. Bring a real requirement and we’ll take it from conversation to a reviewed pull request, and you’ll see who signed off what along the way.