Skip to content

Make your AI
better at its job.

Change the instructions your AI follows. Try them on real examples. See which answers work better.

Manage prompts outside your code. Keep your test criteria, examples, and reviews together. Turn every comparison into a decision report your team can inspect and repeat.

Apply for beta Explore a decision report

In development. Now accepting beta applications.

For example: a customer support reply

Same request. Better reply.

The customer asks
“Please cancel my subscription. I’m not using it.”

Original prompt

“Try to keep customers from canceling.”

The AI replies
“Before you go, how about 20% off your next month?”

Still trying to sell.

Revised prompt

“Help customers cancel. Don’t pitch an offer.”

The AI replies
“You can cancel in Settings → Billing → Cancel subscription.”

Answers the question.

Illustrative example.

Find every prompt in one library.

See the current version, model, owner, and tests. Open a prompt to inspect its history or draft a change without editing application code.

NIBUNO/PRODUCT PREVIEWPrompt Library
Prompt library organized by project, showing prompt roles, owners, current versions, model settings, and test results.

Scroll the preview to explore →

Product preview — Nibuno is currently in development. Examples use illustrative data.

  1. 01
    Organized by project

    Find the prompts that write, check, and score for each application.

  2. 02
    A named owner

    See who is responsible for each prompt.

  3. 03
    Current version and tests

    Open the prompt to compare a draft with the version in use.

“I just need to change the wording. Why does that need a developer?”
“Which prompt is actually live? And who owns it?”
“This model costs twice as much. Is it actually better for us?”

THE OUTPUT THAT BRINGS IT TOGETHER

Run the test.
Get the decision report.

Your prompt. Your test criteria. Your Golden Dataset. Nibuno is being designed to turn them into a repeatable report: what improved, what broke, what it costs, and what to do next.

No rebuilding a presentation after every experiment. Keep the answers, comparisons, human reviews, and reasons together—then use the same report structure for the next test.

Explore an example report ↗See what every feature produces →
ILLUSTRATIVE REPORT · NOT LIVE RESULTS

SUPPORT / REPLY WRITER / P-013

Should we use the revised prompt?

Promising. More review needed.
9 → 11acceptable answers / 12
0newly broken cases
$0.30candidate run cost

Two failures fixed. One policy ambiguity remains. Independent evaluation and approval are still required.

Parameters → Examples → Results → Review → Decision
Open the evidence behind the recommendation ↗

Common problems Nibuno solves

Product owner

The interview needs a better question.

The person responsible for the customer experience shouldn’t need an engineer to test a wording change.

Read the use case
Business team

A wording change is waiting on a developer.

Let the business owner draft the wording and test it against saved examples.

Read the use case
Engineering lead

One point better. Twice the price.

Compare the actual improvement against the additional cost.

Read the use case
Operations leader

The AI rated the opportunity. Now show me why.

Inspect the outputs and reviewer decisions behind each rating.

Read the use case
Developer

The new prompt fixed one problem and created another.

See what changed, why it changed, and which version to roll back to.

Read the use case
Software team

Your product has more prompts than you think.

Give the writer, checker, grader, and judge their own owners, versions, and tests.

Read the use case
All 12 use cases

See the answers behind the score.

Read the original input and both answers. Save the reviewer’s decision and explanation with the test case so they can be used when evaluating the next version.

NIBUNO/PRODUCT PREVIEWHuman Review
Human review of a customer reply, with the original request, baseline and candidate answers, evaluator results, and a reviewer decision form.

Scroll the preview to explore →

Product preview — Nibuno is currently in development. Examples use illustrative data.

  1. 01
    Original input

    Check what the model was actually asked to do.

  2. 02
    Compared outputs

    Read both answers alongside the evaluator’s result.

  3. 03
    Reviewer decision

    Record which answer is acceptable and explain the choice.

Your team edits the prompt.
Your engineers protect the app.

The people who know your customers can improve the wording and test the answers. Engineers keep control of the data, tools, and actions the AI can use.

For example: a customer support assistant

Your team

Edits the instructions

“Help customers cancel. Don’t pitch an offer.”

Engineering

Sets the permissions

  • Can read account details
  • Cannot issue refunds
Your customer support assistant

New wording. Same permissions.

Know what changed and why.

Keep each version’s tests, reviewer feedback, and release decision together. See which version is live and what you can return to if a change causes problems.

NIBUNO/PRODUCT PREVIEWVersion History
Version history showing a local draft, a candidate with test results and pending review, and the current production version.

Scroll the preview to explore →

Product preview — Nibuno is currently in development. Examples use illustrative data.

  1. 01
    Version history

    Inspect the change from the current version to the candidate.

  2. 02
    Tests and review

    Check the comparison results and unresolved reviewer feedback.

  3. 03
    Release and rollback

    Keep the approved version in use while a candidate is reviewed.

Explore all product previews

From a prompt change to a release decision.

  1. 01

    Bring a prompt

    Start with one job: write a reply, conduct an interview, extract information, or score a result.

  2. 02

    Set the standard

    Collect examples and agree on which answers are acceptable and which should fail.

  3. 03

    Try the change

    Run the same examples through prompt and model combinations. Compare accuracy, failures, latency, and cost.

  4. 04

    Make the call

    Approve the new version or keep working. Either way, the test results stay with it.

Find the right model for your work.

Try the same prompt and examples with different models. Compare their answers, response times, and costs before you switch.

See model comparisons

See what the extra cost buys.

Read the answers that improve and the ones that get worse. Decide whether the difference matters to your customers.

Choose based on the results.

Upgrade, try a cheaper model, or keep the one you have. Save the comparison and your team’s decision together.

Journal

Notes on prompt management, evaluation, model selection, and the problems we’re encountering while building Nibuno.

All articles
The principle4 min read

Why you should separate prompts from code

Manage prompt versions, owners, tests, and releases separately from application logic.

Read the article
For business teams3 min read

A prompt change should not start with a ticket

Business owners should be able to test a wording change while engineers protect the application’s technical requirements.

Read the article
For developers3 min read

Is the better model worth twice the cost?

Compare the outputs your business actually needs before you upgrade.

Read the article

BETA APPLICATIONS

Have a prompt that’s causing problems?

Apply to work with us on one prompt your team wants to improve. Selected teams will help define what a good answer looks like and review examples with us.

Next: we review your workflow and contact selected teams to discuss fit, scope, and any pricing. Nibuno is in development; applying does not promise immediate access. Please leave out customer data and confidential prompts.

Apply to be a beta tester

Beta FAQ