You write a sentence. Behind the scenes, a distributed agent pipeline turns that sentence into a packaged project — designed, coded, tested, and delivered.
Every request becomes a job on a Redis-backed queue. A permanent worker claims it, runs it through five phases, and publishes a downloadable package. You can watch the whole path on a live status page, and get a Slack alert when it’s done.
Your prompt and tier are accepted. The job is priced, stored, and placed on the pending queue. Nothing is built yet — the factory has simply taken the order.
The Architect agent creates a workspace, sketches structure, and prepares the directories the rest of the pipeline will use. This is the blueprint stage.
The Coder agent generates real source code with frontier models. Tier routing chooses the model: lighter Spark jobs can use one model path; Core and above lean on higher-capacity coding models. Output is written into the job workspace as actual files — not chat text.
The QA agent reviews what was produced, checks basic integrity, and prepares the tree for packaging. The goal is a coherent project, not a loose pile of snippets.
The Packager zips the workspace into a delivery archive. The job status flips to delivered, the download endpoint unlocks, and Slack is notified. You leave with a file you can open and run.
Creates the job workspace under an isolated path, lays down structure, and hands a clean work directory to the rest of the pipeline.
Calls the LLM layer with your prompt and tier, writes generated code into source files, and records which model path was used.
Runs a verification pass over the workspace so packaging doesn’t ship an empty or broken tree.
Collects source, README, and related files into a delivery zip bound to your job ID.
CodeForge is not a single chatbot process. It is a small production system:
The design is multi-node ready: when traffic grows, additional worker machines can attach to the same queue without changing how customers submit jobs.
Spark, Core, Build, Scale, and Forge map to different project ambition and pricing. Routing can also prefer different model paths by tier so small utilities stay fast while heavier products get deeper generation. You choose the tier on the homepage; the factory enforces the rest.
Describe the product. The agents handle design, code, checks, and packaging.
Start a project