Forge brings rigorous, adversarial architectural planning and reliable step-by-step execution to AI-assisted development. It works with Claude Code, Codex CLI, Cursor, or directly via API — the provider is pluggable.
How It Works
Planning — Two independent AI architects design competing solutions. Adversarial critics tear each design apart. Refiners address the critiques. A judge synthesizes the best elements into a final plan. The result: plans that have been stress-tested before a single line of code is written.
Execution — A deterministic pipeline runs your plan through configurable steps (code, build, lint, test, review, PR) with evidence-based validation at each stage. State lives on disk, not in LLM memory — every step gets a fresh agent with zero context rot.
Thanks. While we began with Express, we intend to add support for more ORMs, so that users can pick the one they want.
We did come across Prisma and we are exploring supporting it.
For now, we support MySQL, PostgreSQL and SQLite, but are also planning to add support for Mongo. Do you have Mongo/NoSQL databases on your roadmap?
It would be great to connect on a zoom - I'm the co-founder of imagine, my email is [email protected]
Yes we will generate configs and scripts that would allow a single command deployment to AWS Lambda and Google Cloud Functions.
We already generate a docker container that is ready to be deployed to Cloud Run or Fargate.
Thanks. Which feature is the one you found to be the most useful? Anything you would like us to add that could help during the initial phases of starting projects?
Hey :) it’s not multiple accounts - this post is by one of the core contributors to the project who is a functional programming expert and I thought it would be helpful if he shared his perspective on why Haskell was chosen for this!
(I couldn’t see my earlier post for some reason on Show HN and I thought I accidentally deleted it, so I thought rather than me reposting it, it would be nice to have one of our contributors share his perspective).
Imagine is a labor of love for my co-founder and me. I’ve been a software engineer for over a decade and we’ve overseen the build out of several (80+) software products over the years.
Each time we were acutely reminded of how much redundancy there exists while writing code, especially for commonly used components and features in software applications. Also, while we believe the current version of the no-code movement is awesome, we found it hard to scale these solutions for applications that required any reasonable level of custom logic.
So we built Imagine - a configurable, cross-framework platform for developers to easily generate high-quality code for common application components, which they can seamlessly integrate into their projects, without any lock-in - and allowing them to focus on writing custom business logic. Our goal has been to make the generated code clean, readable and following known software best practices (we love the 12—factor app philosophy!) - basically indistinguishable from what a skilled engineer would write.
To build the Imagine code generator, we initially considered building with regexes, but discarded that approach as it would have been highly error-prone. Given the high configurability we wanted to offer, we needed to be obsessed with correctness. To achieve this, we decided to use Haskell to build the code generator, as we were able to use Haskell’s built-in type system to represent rules to ensure correctness of the generated code. It also allowed us to enforce various constraints across an app’s specification using static checks.
In terms of features, our beta release allows you to define your starter settings, data models and relations via a UI and we instantly generate code for a dockerized project starter, ORMs, REST & GraphQL CRUD APIs and tests in both Django & Node.js.
We also have the following features on our roadmap:
- Backend frameworks:
- Node/Typescript
- Backend features:
- NoSQL databases (MongoDB, DynamoDB)
- Support for migrations
- Kubernetes deployment scripts
- Logging and monitoring
- Authentication (via multiple providers)
- Payments integration (via multiple providers)
- Frontend frameworks:
- React (incl Redux, Hooks)
- React native
- Frontend features:
- API calling
- Views & components
- Storybooks
We’d like to thank you in advance for trying our product out - we’re still early in development, and would love your feedback and suggestions as we push to make Imagine better everyday!
Well the reason FB/Foursquare/Google etc add these to 3rd party apps is so they can get data. Example if you visit a website which has a Facebook like button, your browser fetches the js files/which maybe even makes an API call to let FB know your IP (and hence location). All this data is fed to the giant system that feeds you ads.
Adding their SDK to other apps/sites (even if there is no user facing need) is a common strategy used by most big companies to get data. In return the app that puts in the SDK gets $ from the company.
Yeah these SDKs are siphoning off data without authorization like your location, songs you listen to and who knows what else (eg other processes or apps running on your phone). Eg the latest iOS exploit allows any app to get access to all SMS data.
Tech companies of today trample upon individual privacy openly. Amazing!
Also learn generic problem solving VS just pattern matching on how to do a specific thing. Frameworks, languages and patterns themselves will keep changing. Core problem solving skills and understanding will stay relevant always.
A skill of knowing how to use React in a particular way to create simple views is something that will be redundant in a few years. These people will unfortunately be out of jobs. The mundane work of converting sketch files to react views will be automated soon
How It Works
Planning — Two independent AI architects design competing solutions. Adversarial critics tear each design apart. Refiners address the critiques. A judge synthesizes the best elements into a final plan. The result: plans that have been stress-tested before a single line of code is written.
Execution — A deterministic pipeline runs your plan through configurable steps (code, build, lint, test, review, PR) with evidence-based validation at each stage. State lives on disk, not in LLM memory — every step gets a fresh agent with zero context rot.