Everything you need.
Nothing you don't.

Every generation is constrained to a maximum of 5 features and 1200 words. Not as a suggestion — as a hard structural rule enforced at the schema validation layer. No matter how big your idea, the output will be something you can actually build in a weekend or a sprint.

Most AI tools will generate whatever you ask for. Ask ChatGPT for a PRD and you'll get 20 features, 4,000 words, and a rabbit hole. Specd's system prompt is engineered to push back, cut features, and force prioritization. It marks extras as explicitly out of scope so you know what you're deferring, not just ignoring. It's opinionated by design, because scope creep kills more side projects than bad code.

Paste your idea and get a complete, structured PRD in under 15 seconds. Problem statement, target user, user flow, feature list with acceptance criteria, build order, file structure, and database schema — all formatted and ready to paste into your AI coding tool.

No back-and-forth prompting. No formatting. No waiting for clarifying questions. One input, one output, ready to build. The generation uses Claude claude-sonnet-4-20250514 with a prompt-cached system prompt, which keeps latency consistently under 15 seconds even for complex ideas. If the first generation isn't perfect, section refinement lets you adjust any field with a single instruction.

Tell Specd what you're building with — Next.js, Python, React Native, Supabase, Stripe, whatever — and every section of the PRD adapts. Acceptance criteria reference your actual technologies. The build order reflects your framework's conventions. The file structure matches your project layout.

Over 30 technologies supported across frontend, backend, database, auth, payments, hosting, and AI categories. If your tool isn't listed, use the "Other" field to specify it. The system prompt is instructed to never mention technologies outside your selected stack, so you won't get generic advice that doesn't apply to your environment. Available on all plans.

The Markdown output is structured to work as a system prompt, a project brief, or a task list inside any AI coding tool. The build order tells Cursor what to build first. The file structure tells Claude Code what files to create. The acceptance criteria tell any AI assistant when a feature is done.

Compatible with Cursor, Claude Code, ChatGPT, Windsurf, GitHub Copilot, and any tool that accepts Markdown or plain text context. The format follows a deliberate sequence: Problem → Target User → Features → Constraints → Build Order → File Structure → Success Criteria. This order maps directly to how AI coding tools consume context — broad picture first, specific implementation second.

Every PRD you generate is automatically saved to your private dashboard. Browse, search by title, and re-export any past generation. Switch between grid and list view. Great for iterating on an idea across multiple sessions or revisiting a project you shelved three months ago.

Your dashboard also shows aggregate stats: total documents, generations this month, and total words generated across all PRDs. The original idea input, tech stack selections, and context settings are all stored alongside the output — so you can see exactly what you asked for and regenerate with adjustments. Available on Pro and Architect plans.

Every PRD exports as a clean `.md` file. Copy to clipboard or download. No proprietary format, no lock-in, no account required to view it after export. Commit it to your repo, drop it in Notion, paste it into a Slack message — it works anywhere Markdown does.

The export includes every section: problem statement, target user, user flow, feature specs, build order, file structure, database schema, architecture diagram source, assumptions, out-of-scope items, and success metrics. It's designed to be self-contained — a developer who has never heard of your product should be able to read the export and start building without asking questions. Available on all plans including Free.

Free tier idea inputs are not stored after generation — they exist only long enough to produce your PRD. Paid tier PRDs are saved to your private account only, protected by row-level security. No one else can see your ideas, not even us during normal operations.

Specd is built on Supabase with RLS (Row-Level Security) policies that restrict all data access to the authenticated user who created it. We do not train on your data. We do not sell your data. We do not use your product ideas to improve our models. The only people who will ever see what you're building are the people you choose to show. See our Privacy Policy for the full technical details.

No team features, no agile ceremonies, no Jira integrations, no stakeholder approval flows. Specd is built for the solo developer or tiny team that needs to move from idea to code without any process overhead. The entire workflow — paste idea, get PRD, start building — takes under two minutes.

The free tier gives you 3 generations per month, which is enough to scope a new idea each week without thinking about it. The Pro plan at $9/month covers 30 generations — more than any normal human will use in a month. There are no seats, no per-user pricing, no minimum commitments. One person, one idea, one spec. Ship it.