Dyad: A Local AI App Builder That Keeps Your Keys and Your Code

What Dyad Actually Replaces

Dyad positions itself as a local alternative to hosted app builders like Lovable, v0, Replit, and Bolt. Those services generate front-end code from natural language prompts, but they run in the cloud and usually require you to send your code and context to their servers. Dyad's pitch is that you run it on your own machine, so generation is faster because there is no round trip to a remote service, and private because your prompts and generated files stay local. The intended user is someone who already works with AI models and wants more control than a hosted builder offers. The README explicitly targets power users, not beginners. That distinction matters: you are expected to bring your own API keys, which means you handle billing, rate limits, and key management yourself. If you are used to a hosted tool that manages all of that, Dyad shifts that burden onto you.

The Local-First Mechanism

The core mechanism is straightforward: Dyad runs on your machine and calls AI models through your own API keys. The description lists support for providers like OpenAI, Anthropic, Google Gemini, DeepSeek, Qwen, and Ollama. Ollama is notable because it means you can run models entirely offline, which eliminates the privacy concern completely. The repository is written in TypeScript and uses Next.js and React, which suggests the app itself is a web-based interface served locally. The data flow is likely: you type a prompt, Dyad sends it to the model provider you configured, receives the response, and generates or edits files in your project. Because it is local, there is no vendor lock-in at the storage level; your projects live on your disk. The README emphasizes 'no lock-in' twice, once for the platform and once for API keys. That is a clear design goal, and it is a real difference from hosted builders that keep your projects on their servers.

Getting Dyad Running

The README is sparse on installation details. It says 'No sign-up required. Just download and go' and points to a download page at https://www.dyad.sh/#download. There is no command-line install command, no npm package, and no Docker instruction in the material. The download page is the only entry point. Once downloaded, you presumably run the app and then configure your AI API keys. The README does not document the exact configuration keys or environment variables. If you are the kind of user who wants to see the setup steps before committing, this is a gap. The repository does have a CONTRIBUTING.md file, which suggests that building from source is possible, but the README does not give the commands. For a tool aimed at power users, that lack of documented setup detail is a weakness. You will need to rely on the website or the repository's other files, which are not in the material provided.

The Split License Is the Real Catch

The license is where Dyad gets complicated. The README states that all code outside src/pro is Apache 2.0, which is a permissive open-source license. But everything inside src/pro is 'fair-source' under the Functional Source License 1.1, which is not an OSI-approved open-source license. That means the core open-source claim is only partial. If you are evaluating this for a company that requires all dependencies to be fully open source, the src/pro folder will be a problem. The FSL 1.1 typically restricts production use or imposes a delayed open-source release, but the exact terms are in the license file, which is not reproduced here. You must read that file before adopting. The split also creates uncertainty about which features are in the open part and which are in the pro part. The README does not specify that. If a feature you need lives in src/pro, you may be using fair-source code under different terms than the rest of the project.

What the Repository Does Not Tell You

The material is thin on operational details. There is no documentation of the user interface, no example prompts, no description of the generated output format, and no mention of how Dyad handles project state, version control, or deployment. The README says it is cross-platform for Mac and Windows, but Linux is not mentioned. That is a concrete limitation if you are on Linux. The release history shows a v1.14.0 from September 2026, with beta releases shortly before, which indicates active development, but the README does not include a changelog or feature list per version. You cannot tell from the material whether Dyad supports multi-file projects, database integration, or authentication flows, which are common in hosted builders. If you need those, you will have to test the app yourself or look at the source code in the repository. The absence of this information is a risk for a power user who expects to evaluate a tool quickly.

Alternatives and the Difference in Approach

The obvious alternatives are the hosted builders Dyad names: v0, Lovable, Replit, and Bolt. Those are cloud services where you do not manage keys or infrastructure; you pay for a subscription or usage. The difference in approach is fundamental. Hosted builders handle the model calls, the storage, and often the deployment of the generated app. Dyad makes you handle all of that locally. In exchange, you get privacy and control. Another alternative is to use a local AI coding assistant like Continue or Cursor, but those are not covered in the material. The key comparison point is that Dyad is a builder, not an assistant: it generates whole apps from prompts, not just code completions inside an editor. If you already use a hosted builder and are satisfied with its privacy trade-offs, Dyad's local model may not be worth the setup effort. If you are uncomfortable sending proprietary code to a third party, Dyad's approach is different in the right way.

Maintenance and Upgrade Cost

The repository shows an active release cadence: v1.14.0, with two beta versions in the days before. That suggests the project is maintained, but it also means you should expect frequent updates. For a local app, that is a moderate cost: you will need to download new versions regularly to get fixes and features. The README does not mention an auto-update mechanism, so you may have to check the download page manually. The split license adds a maintenance consideration: if the project changes the license terms in future releases, you may need to re-evaluate. The README says the code outside src/pro is Apache 2.0, but that is only true for the current state. There is no governance document or contribution policy in the material beyond a CONTRIBUTING.md link. For a power user, the upgrade cost is manageable but not zero. You should plan to track releases if you adopt Dyad.

Editorial conclusion

Adopt Dyad if you are a power user who wants a local AI app builder with your own API keys, values privacy, and is comfortable with a split license where the pro folder is fair-source. Skip it if you require every line of code to be Apache 2.0 or if you need a fully managed cloud service. Before adopting, verify which features live in src/pro, read both LICENSE files, and confirm that your preferred model providers (OpenAI, Anthropic, Ollama, etc.) are supported in the current release.

Official sources

1. dyad-sh/dyad on GitHub
2. Issues
3. Project website
4. README
5. Releases