How it works

One objective. One AI team. Real work.

Every piece of work belongs to a project. You say what you want accomplished; Navorynth plans the team, shows the cost, and the result stays with the project.

  1. Project
  2. Objective
  3. AI team
  4. Work
  5. Deliverable
  6. Project Brain
  7. Next action

The AI team

The right specialists, assembled for the job.

Navorynth does not ask several models the same question and paste the answers together. Roles hand work to each other: one produces, another challenges, the producer answers, an independent role verifies, and a synthesis resolves the result.

You see what the team is doing, its findings and the result. You never need to read the models' private reasoning.

  • Strategistframes the decision and the options.
  • Researcherfinds the evidence.
  • Skepticchallenges weak assumptions.
  • Buildercreates the work.
  • Reviewerchecks it against what you asked for.
  • Synthesizerturns the team's work into one result.

Also assembled when a job needs them: Analyst · Evidence Checker · Product Manager · Planner · Verifier · Product Designer · Visual Reviewer · Financial Analyst

  1. Strategist proposes
  2. Analyst evaluates
  3. Skeptic challenges
  4. Evidence is checked
  5. Recommendation is revised
  6. Synthesizer writes the result
A completed Review in Navorynth: “What your AI team is doing” shows two independent reviewers whose findings were compared, then the review report. Total provider cost $0.35 of a $0.82 hard maximum.
A real Review: two independent reviewers, their findings compared before the report was written.

Six kinds of work

Different ways the same team works.

Choose one, or let Navorynth suggest it from what you typed.

Ask

Contextual help, from one quick answer to a full team discussion.

Quick is one AI with your project's context. Team and Deep bring in several specialists who build on, and push back on, each other's work.

Research

Evidence first: a Researcher gathers it, an Evidence Checker tests it, a Synthesizer writes it up.

The result separates evidence from inference and names the uncertainty, the limitations, any contrary evidence and the open questions.

Decide

A Strategist frames the options, an Analyst weighs them, a Skeptic challenges the favourite.

You get a recommendation with its reasoning and risks. The AI recommends; you decide. Nothing becomes a decision until you approve it.

Review

Independent reviewers check work against its requirements, evidence and constraints.

Findings are prioritised, actionable and tied to evidence. A review changes nothing on its own.

Improve

You approve specific findings. Navorynth makes those changes, verifies them and reviews them again.

It never widens the scope: only the findings you approved are changed.

Build

A product team plans, designs, builds and checks small working software on your PC.

Functional, product, visual and commercial-readiness reviews come before anything is called complete.

Reasoning depth

Quick
One AI, fast, with the project's context.
Team
Several specialists collaborate and challenge each other.
Deep
A richer team with more perspectives, challenge and verification where it helps.

Execution

Standard
Your providers' normal service.
Economy
The same team and the same reasoning, on a provider's own lower-cost service where one is available (it can take longer).

Economy is never lower-quality reasoning: it is the provider's own discounted service for the same request.

Project Brain

Your project remembers what matters.

Every project keeps its own Project Brain: what was decided, the evidence, the constraints, the findings and the open questions. The next job starts from there, so you do not explain it all again.

  • A recommendation stays a recommendation until you approve it. Only your approval makes it a decision.
  • Later work respects the decisions you approved.
  • Each project's knowledge stays in that project. It is never shared with your other projects.
Project Brain This project only
  • Decisionsonly the ones you approved
  • Evidencewith its sources
  • Constraintsbudgets, rules, must-haves
  • Requirementswhat the work must do
  • Findingsfrom research and reviews
  • Recommendationsuntil you decide
  • Assumptionslabelled as assumptions
  • Open questionswhat is still unknown
  • Deliverablesevery finished result
Recommendationfrom your AI team
Your approvalonly you can give it
Decisionlater work respects it
Illustration: what a Project Brain keeps, and how a recommendation becomes a decision.

Build

More than code generation.

Build assembles a product team for small working software that runs on your PC, and goes beyond writing code with product, visual and commercial-quality review.

  1. Plan
  2. Product design
  3. Build
  4. Functional review
  5. Visual review
  6. Polish
  7. Verification

Functional

It works, and its main journey is verified in a real browser.

Product / UX

It does what the plan promised, the way a user expects.

Visual

Its look is reviewed from real screenshots of the product.

Commercial readiness

It is reviewed as something you could put in front of people.

The Packwise project in Navorynth, marked Complete: Functional, Product / UX, Visual and Commercial readiness all passed, the core workflow verified in a real browser, Reviewer approved, $6.82 total of a $7.47 maximum.

Internal validation project

Packwise

A trip packing planner that runs in the browser: add items by category, tick them off, filter what's left. Planned, built, reviewed and polished by Navorynth from a plain-English description.

4 / 4
quality gates passed
13 / 13
planned tasks built
105
automated tests passed
11 / 11
main-journey steps verified in a real browser

Metered provider usage for the original build and two bounded polish passes: $6.82, inside a $7.47 maximum.

One internal example, not a price or a promise. Cost and results vary with the scope of the work, and Navorynth recommends your own review before you rely on anything it builds.

Packwise, built by Navorynth: the Weekend in Lisbon trip, packed 2 of 4.
Packwise itself, as Navorynth built it.
  • Build plans first: you approve the plan and its cost before anything is built.
  • Building and testing happen on your PC, in a secure build environment. Build does not deploy or publish anything.
  • Every finished build comes with a Proof of Build: what was built, tested and reviewed, and what to check yourself.
  • Build needs OpenAI and Anthropic keys, plus Claude Code, Git and the secure build environment. The app lists exactly what is missing.

Providers

Leading AI providers, working as one team.

Your team can use models from several providers behind the scenes. You work with roles, not model names.

  • OpenAI
  • Anthropic
  • Google Gemini
  • xAI

Bring your own keys

  • You connect the providers you want. One is enough to start; Build needs OpenAI and Anthropic.
  • Each provider bills you directly for what your team uses, at its own prices. Navorynth does not resell or mark up provider usage.
  • Your keys stay on your PC, in Windows Credential Manager, and are sent only to the provider they belong to.
  • Navorynth's account service never receives your keys, your projects or your files.

Cost control

Know the ceiling before the team starts.

An estimate before work starts

Every plan shows the team, the expected provider cost and a hard maximum. Nothing is spent until you press Start.

A hard maximum

Navorynth stops before starting work that would go over the maximum you set.

Usage you can read

Afterwards, Usage shows what each provider billed, by project and by kind of work.

Funding checked before Start

If a provider the team needs has reported a billing or usage-limit problem, you are told before the job starts, with a link to that provider's billing page.

Safe stops

If a provider runs out of credit during a job, Navorynth stops before its next request and keeps everything already done.

Resume

Add credit on the provider's page, choose Test again, then Resume. The job continues from where it stopped, without repeating or paying again for finished steps.

Navorynth's Usage page: what the AI providers billed this month, split by provider (Anthropic, OpenAI) with buttons to each provider's own billing page, and by kind of work.
Usage: what each provider billed, with a button to each provider's own billing page.

Most providers do not let an API key see the remaining balance, so Navorynth cannot always know it in advance. We recommend a monthly limit at each provider as well.

Needs you

It keeps moving until it actually needs you.

Everything that waits for you, from every project, is in one place. Each item says what happened, what was kept, and the one thing to do next.

  • A provider needs billing attention
  • A plan or a change waits for your approval
  • The team needs a clarification
  • A job reached its maximum
  • A build needs your decision
  • Your subscription or access needs attention

Questions

Pricing questions are on the Founding Beta page; security and data on Security.

What is Navorynth?

A Windows desktop app that runs your work as an AI team. You give it an objective inside a project; it assembles the specialists the job needs, coordinates them, and produces a deliverable. Each project keeps a Project Brain, so the next job starts from what you already know.

How is this different from ChatGPT or Claude?

Chat assistants answer one prompt at a time, and you carry the context. Navorynth plans each piece of work as a team: one role produces, another challenges, an independent one verifies, and a synthesis resolves the result. The project remembers decisions, evidence and open questions between jobs.

Navorynth uses AI providers' models through their APIs, with your own keys. It is not affiliated with or endorsed by any provider.

Do I need accounts with AI providers?

Yes: an API account with at least one supported provider. Chat subscriptions such as ChatGPT Plus or Claude Pro are a different product and cannot be used. The app walks you through creating a key on the provider's own page.

Which providers are supported?

OpenAI, Anthropic, Google Gemini and xAI.

Do I need all four providers?

No. One provider is enough to start Ask, Research, Decide, Review and Improve. With two or more, the team can draw on independent perspectives. Build needs both OpenAI and Anthropic, plus Claude Code, Git and the secure build environment on your PC.

Where are my API keys stored?

In Windows Credential Manager on your PC. Each key is sent only to the provider it belongs to. Navorynth's account service never receives your keys, your projects or your files.

Does Navorynth turn AI recommendations into decisions?

No. A recommendation stays a recommendation until you approve it. Only your approval creates a decision in the Project Brain, and later work respects the decisions you approved.

What happens if an AI provider runs out of credit?

If Navorynth already knows a provider the team needs has a billing or usage-limit problem, it tells you before the job starts. If credit runs out during a job, it stops before the next request, keeps the finished work, and shows the job under Needs you. Add credit on the provider's page, choose Test again, then Resume.

Most providers do not let an API key see the remaining balance, so Navorynth cannot always know in advance. Setting a monthly limit at each provider is still wise.

Can I set a spending limit?

Yes. Every job has an estimate and a hard maximum before it starts, and Navorynth stops before starting work that would go over it. Usage shows what each provider actually billed.

What is Project Brain?

Each project's memory: decisions you approved, evidence, constraints, requirements, findings, recommendations, assumptions, open questions and deliverables. It stays with that project and is never shared with your other projects.

What can Build create?

Small working software tools that run on your PC, such as a browser-based planner or a command-line calculator. Build plans, builds, tests and reviews them on your computer; it does not deploy or publish anything. Every build ends with a Proof of Build, and we recommend your own review before you rely on it.

Does Navorynth work on Mac?

Not yet. Navorynth V1 is a Windows desktop app for Windows 10 and 11.

Is provider usage included in my Navorynth subscription?

No. Your subscription pays for Navorynth. Each AI provider bills you directly for what your team uses, at its own prices. Navorynth does not resell or mark up provider usage.

Can I download Navorynth today?

Not publicly yet. Navorynth is in a Founding Beta: we are onboarding a small group of early users personally. Request access and we'll be in touch.

Bring your AI team its first objective.