From vague idea to deployable spec — in one sitting.
AI-powered depth interviews that extract requirements non-engineers miss, then generate production-ready PRDs and implementation specs.
Live Demo · Product Hunt · GitHub · 日本語
AI can build a PoC in minutes. But shipping requires more:
- Edge cases nobody thought of
- Security & availability requirements
- Acceptance criteria engineers can actually implement
- Validation that users truly have this problem
Non-engineers can't see what's missing. DeepForm bridges that gap.
Your idea → AI Depth Interview → Facts → Hypotheses → PRD → spec.json
↓
Claude Code / Cursor / Any Agent
| Step | What happens | Output |
|---|---|---|
| 1. AI Depth Interview | AI interviewer probes for concrete examples, frequency, severity, workarounds | Structured conversation |
| 2. Fact Extraction | Extract facts, pains, frequencies with source evidence | Evidence-linked facts |
| 3. Hypothesis Generation | Generate hypotheses with supporting facts, counter-evidence, unknowns | Falsifiable hypotheses |
| 4. PRD Generation | MVP-scoped PRD with testable acceptance criteria (ISO 25010) | PRD.md |
| 5. Spec Export | API specs, DB schema, test cases | spec.json |
Share an interview link with your users to collect feedback at scale:
- Create a campaign from any interview
- Share the link — respondents go through the same AI interview
- View aggregated pain points, common facts, and keyword analysis in real-time
- Use AI cross-analysis to find patterns across all responses
- Evidence-linked specs — Every requirement traces back to actual user statements
- Counter-evidence on hypotheses — Prevents confirmation bias with falsification patterns
- Purpose-built interview AI — Not generic chat; specialized depth interview techniques
- Campaign interviews — Collect and aggregate feedback from multiple users
- Agent-ready export —
spec.jsondrops directly into coding agents - Production readiness check — ISO/IEC 25010 quality checklist before you ship
Visit deepform.exe.xyz and sign in with GitHub.
git clone https://github.com/susumutomita/DeepForm.git
cd DeepForm
make startRequires Bun runtime. Set ANTHROPIC_API_KEY for Claude API access.
Once DeepForm generates your spec, hand it to a coding agent:
| Agent | How |
|---|---|
| Claude Code | Place PRD.md in repo root → claude |
| Cursor | Paste spec.json into Composer |
| Any agent | spec.json is standard JSON |
frontend/ Vite + TypeScript (vanilla)
src/
main.ts Entry point & routing
api.ts Typed API client
interview.ts Core interview & analysis flow
campaign-analytics.ts Campaign results dashboard
shared.ts Campaign respondent interview
sessions.ts Session list management
auth.ts GitHub OAuth login
i18n.ts JA / EN / ES
ui.ts UI utilities
src/ Hono + TypeScript server
app.ts Hono app with middleware
llm.ts Claude API client (API key / OAuth / gateway)
db/
index.ts Kysely database setup + migrations
migrations/ Schema migrations (001–003)
routes/
auth.ts GitHub OAuth + exe.dev auth
analytics.ts Page view tracking
billing.ts Stripe webhook handler
feedback.ts User feedback
sessions/
crud.ts Session CRUD
interview.ts AI interview (streaming)
analysis.ts Facts / hypotheses / PRD / spec / readiness
campaigns.ts Campaign management
pipeline.ts Full analysis pipeline (SSE)
middleware/
auth.ts Auth middleware (GitHub OAuth + proxy headers)
analytics.ts Request analytics
- Runtime: Bun
- Server: Hono + TypeScript
- Frontend: Vite + TypeScript (vanilla, no framework)
- Database: SQLite via Kysely (WAL mode)
- AI: Claude API (Anthropic)
- Auth: GitHub OAuth + exe.dev proxy headers
- Payments: Stripe (optional, disabled by default)
- i18n: Japanese / English / Spanish
make install # Install dependencies
make dev # Start dev server (hot reload)
make before-commit # Run lint + typecheck + test (317 tests)See CLAUDE.md for development guidelines.
See LICENSE.
Built with Claude AI
Open source. Self-hostable. Free.