Nuanced
Nuanced is a macOS desktop app for spec-driven development with coding agents. It turns a rough idea into an explicit, living specification, generates tasks from it, then lets you review changes and commit without leaving the app.
What is Nuanced?
Nuanced is a macOS desktop application for building software with coding agents, published by Nuanced, Inc. and launched publicly on 24 June 2026. Its starting observation is that agents have made code generation fast while making something else harder: holding the system in your head, knowing which details actually matter, and tracing where an implementation drifted away from what you meant.
Its answer is to treat the spec as a first-class object in the development loop rather than a disposable plan buried in a chat transcript. The spec is a living document sitting alongside the conversation. It is the source of truth the agent implements against, and the thing you review the code against as it evolves.
The workflow runs in four stages. You start from raw intent, a short and deliberately unfinished description of what you want to build or change. Instead of firing that straight into code generation, the app opens what it calls progressive disambiguation: an iterative pass that helps you settle goals, constraints, invariants and what is explicitly out of scope, one layer at a time, without demanding that you define everything upfront. Once the spec is precise enough, tasks are derived from it and executed, with progress shown as they run. Finally you review the generated files and commit from inside the app, without switching to a terminal.
Three spec features arrived on 14 July 2026: selecting a section and requesting a change scoped to it, a turn-by-turn diff showing which sections moved as the conversation progressed, and Spec Tour, a guided interactive walkthrough for rebuilding context on a long spec without reading it end to end.
Practically, Nuanced ships as a packaged .dmg distributed through GitHub releases, observed at v0.0.155. It requires a GitHub sign-in and a connected ChatGPT account, and uses Codex as its execution agent. A workspace is simply the local folder or repository it operates on. Product documentation amounts to a single Getting Started page, with two blog posts and a Discord community alongside it: a genuinely working but very young product.
What it does
- Turn a rough idea into an explicit specification before any code is generated
- Clarify goals, constraints, invariants and out-of-scope items one layer at a time
- Generate tasks from the specification and follow them as they run
- Select a single section of a spec and ask for a change scoped to it
- See which sections of the spec changed, conversation turn by turn
- Rebuild context on a dense spec through a guided Spec Tour
- Review the generated files and commit without leaving the app
When to use Nuanced / When not to
A quick filter to help you decide if Nuanced is the right fit.
When to use Nuanced
- Developers on macOS who already drive coding agents against a local repository
- Engineers tackling medium to large changes where the original intent gets lost in a long chat thread
- Teams that need to trace why a technical decision was made, not only what changed
- Practitioners of written design docs and RFCs who want that discipline back at agent speed
- Solo builders and startup founders already equipped with GitHub, ChatGPT and Codex
When not to use Nuanced
- Anyone not on macOS: no Windows, Linux, web or mobile build exists
- Teams standardised on an execution agent other than Codex, the only one supported so far
- Organisations that require a GDPR framework, a Data Processing Agreement or a stated hosting jurisdiction
- Developers working on confidential or NDA-bound source code, given how broad the content licence is
- Buyers who need a predictable budget, since no pricing whatsoever is published
How to use Nuanced
A typical end-to-end flow, from setup to results.
- Download the .dmg from the Download for macOS link on the homepage
- Open the .dmg and drag Nuanced into your Applications folder
- Launch Nuanced from Applications
- Sign in with GitHub: the browser opens, you authenticate there, and control returns to the app automatically
- Connect your ChatGPT account
- Add a workspace with cmd + O, meaning the local folder or repository you want to work in
- Start a spec by typing in the main prompt or pressing cmd + N
- Describe what you want to build or change; a rough idea is enough at this stage
- Work through the disambiguation questions until the spec is precise enough to execute
- Run the tasks derived from the spec, then review the generated files and commit from the app
Pros & Cons
Pros
- The spec outlives the conversation instead of being buried in the backscroll
- Intent stays traceable: you can revisit why a decision was made, not just what changed
- Turn-by-turn spec diffing is rare among agentic coding interfaces
- Spec Tour tackles a real problem, regaining context on a long specification
- The whole loop lives in one app: specify, execute, review and commit, no terminal
- Short, documented installation with automatic background updates
- The team builds Nuanced with Nuanced, and the founder's background is stated openly
Cons
- macOS only, with an arm64 build: no Windows, Linux, web or mobile version
- Codex is the only execution agent supported so far
- Two mandatory external dependencies: a GitHub account and a connected ChatGPT account
- No privacy policy is published anywhere on the site
- Terms section 5 grants a perpetual, irrevocable, sublicensable licence over your content, model training included, with no documented opt-out
- No pricing is published, even though the Terms already contemplate payment
- Very young product: v0.0.155, a single documentation page and one published contact address
Pricing & Plans
Nuanced publishes no pricing information of any kind. There is no pricing page, the sitemap listing four URLs in total, and no amount, currency or tier appears anywhere on the site. The application can be downloaded and installed with no payment announced, but the vendor never qualifies this as a permanent free plan or as a free trial, so neither can be recorded as such. Section 9 of the Terms of Service nonetheless contemplates payment, capping total liability at the amount paid over the preceding twelve months, or USD 100 where nothing has been paid. Prospective users should therefore treat the current cost as unstated rather than confirmed as free.
Data, GDPR & hosting
A consolidated view of how Nuanced handles your data.
GDPR overview
There is no GDPR framework on this site at all. Nuanced publishes no privacy policy: the sitemap lists only four URLs, and a search across every archived page returns no mention of the GDPR, of data protection or of the CCPA. No Article 27 EU representative is named, no Data Processing Agreement is offered or refused, no subprocessor list is published, and no hosting country or region is declared. The only legal document available is the Terms of Service, last updated April 2026 and governed by the law of the State of California. This record therefore marks GDPR compliance as N, and the distinction matters: the site does not deny compliance, it says nothing on the subject. Anyone routing personal or client data through the app should raise the question with the vendor directly.
Who owns the data?
Under the Terms of Service you keep ownership of your content, but that ownership is largely nominal. Section 5 grants Nuanced, Inc. a worldwide, perpetual, irrevocable, transferable, sublicensable and royalty-free licence to use, reproduce, store, process, modify, adapt, analyse and create derivative works from your content and any related data, for any purpose. Section 7 confirms the company retains all rights to the software, systems, models and technology. Section 6 adds that feedback you send is not treated as confidential and can be exploited without compensation. Because the licence is transferable and sublicensable, its reach is not limited to Nuanced, Inc. itself.
Reuse rights
The Terms say far more about what the vendor may do with your data than about what you may do with it. You keep ownership of your content and remain free to use it in your own work, but the software is licensed rather than sold, and section 2 forbids copying, modifying, reverse engineering or redistributing it, or using it to build a competing product. On the vendor side the permitted purposes are listed explicitly in section 5: operating the service, improving and developing products, training and evaluating models, generating insights and analytics, and commercialising products or data derived from that data. Technical data, usage data, metadata, logs, interactions and system behaviour are collected as well, and may be disclosed in aggregated, de-identified or anonymised form, including combined with other users' data. No opt-out from model training is documented anywhere on the site.
Data retention & training
Hosting summary
Nuanced declares nothing about where data is hosted. There is no trust or security page, no stated hosting country or region, no certification such as SOC 2 or ISO 27001, and no mention of encryption anywhere on the site. The only jurisdictional statement published is section 12 of the Terms, which places the agreement under the law of the State of California. One observation is technical rather than contractual and should not be read as a hosting commitment: the domain resolves to 216.150.1.1, geolocated in the United States on AS16509 (Amazon.com, Inc.), with an anycast indication. That describes the website, not the storage of user data. Because the app is a desktop program working on a local folder, part of the processing is necessarily local. Execution nonetheless runs through Codex and requires a connected ChatGPT account, so user content leaves the machine along a path the vendor never documents. Anyone with data-residency requirements will need to ask directly.
Things to keep in mind
Risks and trade-offs to weigh before adopting Nuanced.
- No privacy policy exists on a tool that authenticates through GitHub and operates directly on local code repositories
- Terms section 5 grants a worldwide, perpetual, irrevocable, transferable and sublicensable licence over your content for any purpose, training and evaluating models included, with no documented opt-out
- Logs, interactions, metadata and system behaviour are collected, and may be shared in aggregated or anonymised form combined with other users' data
- Feedback you send is explicitly non-confidential and exploitable without compensation under section 6
- No DPA, no subprocessor list, no declared hosting country and no security certification; liability is capped at USD 100 where nothing has been paid
- Delegating the writing of specifications can erode the very skill the product celebrates, the habit of thinking a system through in writing before building it
- A well-formed spec can create false confidence: the Terms themselves warn that outputs may be incorrect and must be independently reviewed before you rely on them
Setup & Integrations
Technical difficulty
Low for anyone comfortable on macOS. Installation follows the standard route: download the .dmg, drag Nuanced into Applications, launch it. There is nothing to compile and no command line involved. The only friction is the two authentications, signing in with GitHub, which completes in the browser and hands control back automatically, then connecting a ChatGPT account, followed by choosing a local folder or repository as your workspace. Updates install themselves in the background. The real prerequisites are environmental rather than technical: an Apple Silicon Mac, a GitHub account, and access to ChatGPT and Codex.
Deployment
Integrations
Behind Nuanced
Fundraising
Social
Resources
All the official URLs gathered for verification and reference.
Frequently asked questions
Which systems does Nuanced run on?
Do I need an account to use it?
Which agent actually writes the code?
How much does Nuanced cost?
Is my code used to train models?
Does the site publish a privacy policy?
What exactly is a workspace?
How mature is the product?
Should you pick Nuanced?
Nuanced is a real, verifiable product rather than a landing page. The app downloads as a packaged .dmg from dated GitHub releases, the installation documentation works, and two detailed engineering posts describe the workflow with product screenshots. Its central idea is genuinely uncommon too: making the specification a durable object that can be edited, diffed turn by turn and toured, rather than a plan that dissolves into a chat transcript, addresses a bottleneck most agentic tools leave untouched.
The reservations are equally concrete. The product is young, at v0.0.155, launched publicly less than two months before this review, with a single page of documentation. Its perimeter is narrow: macOS only, Codex as the sole execution agent, and mandatory GitHub and ChatGPT accounts. Nothing is published about pricing, so the eventual cost is unknown even though the Terms already anticipate payment.
The most serious point is contractual rather than technical. Nuanced publishes no privacy policy at all, while section 5 of its Terms grants the company a worldwide, perpetual, irrevocable and sublicensable licence over user content for any purpose, explicitly including training and evaluating models and commercialising derived data. For a tool whose entire function is to operate on a source repository, that combination deserves attention before adoption, particularly for client work, proprietary code or anything under a confidentiality obligation.
For an individual developer on macOS experimenting with spec-driven agentic workflows on non-sensitive code, Nuanced is worth trying today. For a team carrying compliance obligations, the absence of any data-protection framework, with no privacy policy, no DPA, no stated hosting jurisdiction and no subprocessor list, is the question to settle with the vendor first.
- Choosing a selection results in a full page refresh.
- Opens in a new window.