Rogation

Turn 20 interviews into Friday's decision. Self-serve synthesis for Product Managers.

Paste transcripts, support tickets, or survey responses. Rogation clusters them into pain points, ranks opportunities against your weights, streams a spec in 30 seconds, lets you refine it over chat, then pushes it to Linear as a real issue. The whole loop — evidence in, ticket out — happens without leaving the browser.

• See CHANGELOG.md for what shipped in v0.1.0.0 (first release).
• See docs/designs/rogation-v1.md for the v1 product design.
• See DESIGN.md for the design system (tokens, components, a11y baseline).
• See TESTING.md for the test layers, local Postgres setup, and coverage gates.
• See CLAUDE.md for the full architecture reference and project conventions.

What's in v0.1.0.0

• Evidence ingestion. Paste text or upload .txt / .md / .log / .csv / .json / .yaml. Content-hash dedup, per-plan caps, 20 files / 10 MB per batch.
• Clustering. Two paths: full (cold start, Sonnet 4.6) and incremental (KNN-triaged, Haiku 4.5). The orchestrator picks per design §7. Async dispatch via an Inngest worker — the UI polls at 1.5s while a re-cluster runs and resumes across page reloads via latestRun. Every cluster carries the prompt_hash of the prompt that produced it, so eval regressions pinpoint the offending version.
• Opportunities. Five LLM-scored primitives feed a pure weighted score that re-ranks optimistically on slider drag; the server uses the exact same formula.
• Spec editor. Streaming SSE generation, A/B/C/D readiness grade, IR-aware refinement chat, every turn produces a new persisted spec version.
• Linear push. One click creates a real Linear issue with acceptance criteria + full markdown. OAuth, AES-256-GCM token storage, HMAC-signed state.
• Billing. Stripe Checkout + Customer Portal. Webhook-mirrored subscription state. Plan limits (ctx.assertLimit) enforced server-side.
• Observability. Sentry (errors, 3 runtimes, noise filter), PostHog (activation funnel), Langfuse (LLM traces with user attribution).
• Tenant isolation. Three layers: ESLint allowlist on @/db/client, transaction-scoped app.current_account_id, Postgres RLS with FOR ALL + WITH CHECK.

Getting started

Requires bun 1.3+ and Postgres 14+ with the vector extension (Supabase, Neon, or a local pgvector/pgvector container all work).

bun install
cp .env.example .env.local
# edit .env.local — see "Environment" below
bun run db:migrate
bun run dev

Then open http://localhost:3000.

Environment

Required at boot (the app fails fast via env.ts when any of these are missing):

• DATABASE_URL — Postgres connection string with pgvector + pgcrypto available.
• NEXT_PUBLIC_CLERK_PUBLISHABLE_KEY + CLERK_SECRET_KEY — Clerk auth.
• CLERK_WEBHOOK_SIGNING_SECRET — signature verification on /api/webhooks/clerk.
• ANTHROPIC_API_KEY — Sonnet 4.6 (synthesis) + Haiku 4.5 (generation / refinement / scoring).
• OPENAI_API_KEY — text-embedding-3-small for evidence embeddings.
• STRIPE_SECRET_KEY + STRIPE_WEBHOOK_SIGNING_SECRET + STRIPE_PRICE_ID_SOLO + STRIPE_PRICE_ID_PRO.
• NEXT_PUBLIC_APP_URL — used for OAuth callbacks and Stripe redirects.

Optional (fail open / no-op when unset — dev and CI run without them):

• SENTRY_DSN + NEXT_PUBLIC_SENTRY_DSN — error tracking.
• NEXT_PUBLIC_POSTHOG_KEY + POSTHOG_API_KEY + NEXT_PUBLIC_POSTHOG_HOST — product analytics.
• LANGFUSE_SECRET_KEY + LANGFUSE_PUBLIC_KEY + LANGFUSE_HOST — LLM trace capture.
• UPSTASH_REDIS_REST_URL + UPSTASH_REDIS_REST_TOKEN — rate limiting (fails open in dev).
• LINEAR_CLIENT_ID + LINEAR_CLIENT_SECRET + INTEGRATION_ENCRYPTION_KEY — Linear OAuth + token storage.

Local Stripe webhook testing: stripe listen --forward-to localhost:3000/api/webhooks/stripe and paste the printed signing secret into STRIPE_WEBHOOK_SIGNING_SECRET.

See .env.example for the full annotated list.

Scripts

• bun run dev — Next.js dev server.
• bun run build — production build.
• bun run check — typecheck + lint + build + test (the local gate; same steps CI runs).
• bun run typecheck / bun run lint / bun run test — run each gate in isolation.
• bun run test:watch / bun run test:ui — Vitest in dev modes.
• bun run storybook — Storybook on port 6006 for the shared UI primitives (components/ui/).
• bun run db:generate — generate a new migration from schema changes.
• bun run db:migrate — apply pending migrations to $DATABASE_URL.
• bun run db:push — push schema directly (dev only).
• bun run db:studio — Drizzle Studio (web UI).

Structure

app/              Next.js App Router (pages, layouts, route handlers, webhooks)
components/ui/    Shared UI primitives (DESIGN.md §6) + colocated stories
db/               Drizzle schema, migrations, RLS policies, scoped client
lib/              Domain logic — evidence, LLM router + prompts, spec IR, billing
server/           tRPC context + routers (auth'd procedures, RLS binding)
test/             Vitest unit + integration suites (DB-gated via TEST_DATABASE_URL)
env.ts            Typed env — import `env` from here, never process.env
docs/designs/     Solution designs (v1 plan lives here)
DESIGN.md         Design system (tokens, components, a11y)
CLAUDE.md         Project conventions, architecture, tenant-isolation rules
TESTING.md        Test layers and local setup
CHANGELOG.md      What shipped, per-version
TODOS.md          Deferred work ledger