All work
AI & automationOpen source

SpecDesk

From app idea to a buildable plan in about two minutes: AI agents interview the founder, then write the features, screens, stack, timeline and price, with downloads and a booking link. Next.js and FastAPI, with seven LangGraph agents.

Product
AI project planner for agencies
Built with
Next.js, FastAPI, LangGraph
SpecDesk: describe your app idea and set a budget, with the agent flow below
7

planning agents, some running in parallel

~2 min

from idea to a buildable plan

0

prices set by the AI: a rule engine decides

Overview

The problem we set out to solve.

A founder types an app idea and sets a budget. An interviewer agent asks a few rounds of questions, each with a recommended answer, then seven planning agents write the plan: features, screens with wireframes, tech stack, timeline, risks and price. The founder can tweak the scope, download everything, and book a call with the agency.

  • Turn a vague, one-line idea into a plan a team could actually build from.
  • Ask the right questions without making the founder fill in a long form.
  • Give a price people can trust: the AI should never just make one up.
Plan workspace: price range, timeline, effort and features you can switch on or off
The plan arrives in the chat, with its files ready to download
“Why this price”: every step from hours to the final range
Screens tab with a wireframe for each screen
Admin panel: leads, plans, models and API keys
Researcher agent: real competitors found on the web, and how to stand out
The front desk: say hi and it asks about your app instead of starting a plan
The plan workspace on a phone
What we built

Every part, built to work in the real world.

Front desk and interview

A front-desk agent answers small talk and waits for a real idea; then an interviewer asks rounds of questions, each with a recommended answer and clickable options.

Planning agents in parallel

Analyst, product, UX, architect, estimator, backlog and critic on LangGraph, plus a researcher that finds real competitors on the web.

Rule-based pricing

The AI scores complexity; price and weeks come from a rule engine, and “Why this price” shows every step.

Plan workspace

Switch features on or off and the price updates. Every change is saved as a new plan version.

Downloads

spec.md, backlog CSV and JSON, a flow diagram and a build prompt, or everything as one ZIP.

Admin with full model control

Leads and plans, encrypted provider keys, a model per agent with fallbacks, and the cost of every model call tracked.

How it's built

A Next.js front end talks to a FastAPI backend, where one LangGraph graph runs each planning session. The interview pauses for the founder's answers and resumes exactly where it stopped, with state kept in a Postgres checkpointer, so a refresh or a redeploy loses nothing. Agents on the same level run in parallel and stream their progress to the browser. Money and weeks are always computed in code from a YAML rule file, never by the model.

  • Next.js 16 & React 19Web app with Tailwind CSS v4 and shadcn/ui
  • FastAPIPython 3.12 API, managed with uv
  • LangGraph & LangChainInterviewer and planning agents, Postgres checkpointer
  • GroqDefault model provider (Anthropic optional)
  • Neon PostgresData with SQLAlchemy 2 and Alembic migrations
  • Better AuthSign-in with a one-time email code
  • Brevo & CalendlyEmail delivery and call booking
The outcome

An open-source planner that turns an idea into a priced, buildable plan in about two minutes, with every price explained and a direct path to booking a call.

Keep exploringAll 10 projects
Prefolk landing page: “Your first users. Your blind spots. Before launch.”Up next · Prefolk
Next case study

Prefolk

AI user-testing SaaS

Have an idea? Let's build it.

Tell us what you want to build. You'll get a clear plan, a timeline and a fixed price within a week.