Frontier LLM Intelligence Dashboard

Local-first web dashboard for daily frontier LLM monitoring. It ingests curated primary sources, deduplicates overlapping coverage, tags companies and topics, scores impact, generates a daily digest, and exposes source health plus pipeline run logs.

What v1 does

• Ingests verified primary-source feeds for OpenAI, Google DeepMind, Hugging Face, Microsoft Research, NVIDIA, and arXiv
• Includes an optional Anthropic official-page parser with visible failure reporting if the page shape changes
• Deduplicates overlapping items into canonical developments
• Tags topics, companies, models, and source types
• Scores trust, frontier relevance, novelty, release significance, and confidence
• Generates a daily digest for a rolling 24-hour window
• Renders a dashboard with today view, archive, item explorer, source health, and run logs
• Keeps X/Twitter and Reddit as honest disabled integrations until credentials and compliant adapters are added

Stack

• Next.js 16 App Router + TypeScript
• Tailwind 4 + shadcn-style UI primitives
• Prisma schema and client
• SQLite for local dev with a lightweight SQL migration runner
• Zod for environment validation
• Node/TypeScript ingestion scripts with optional OpenAI summarization
• Vitest for smoke coverage

Why the migration flow is custom

Prisma 7’s local SQLite runtime is adapter-based, and the Prisma schema engine path was not reliable in this environment. The repo still uses Prisma schema + generated client as the source of truth, but local migrations are applied through committed SQL files in prisma/migrations/ via:

npm run db:migrate

That keeps the schema readable, the migration history explicit, and local setup dependable.

Quick start

npm install
cp .env.example .env
npm run db:generate
npm run db:migrate
npm run db:seed
npm run sweep:daily
npm run dev

Open http://localhost:3000.

Commands

npm run dev
npm run db:migrate
npm run db:generate
npm run db:seed
npm run ingest
npm run digest
npm run sweep:daily
npm run demo:seed
npm run sources:list
npm run sources:enable <key>
npm run sources:disable <key>
npm run test

Source coverage

Enabled primary adapters

• openai-news-rss
• google-deepmind-rss
• huggingface-blog-rss
• microsoft-research-rss
• nvidia-developer-atom
• arxiv-llm-api
• anthropic-news-page as an experimental official HTML parser

Disabled optional signal layers

• x-selected-accounts
• reddit-llm-communities

See docs/social-setup.md for the setup contract and disabled-state expectations.

App structure

• src/app: Next.js routes and views
• src/lib/ingest: adapters, normalization, dedupe, tagging, scoring, and pipeline runner
• src/lib/digest: deterministic digest generation
• src/lib/queries: server-side data access for dashboard pages
• prisma/schema.prisma: data model source of truth
• prisma/migrations/0001_init/migration.sql: initial committed schema
• .agents/skills/frontier-llm-daily-sweep/: reusable operator skill for running the daily sweep

Verification

The intended verification flow is documented in docs/verification.md. The core checks are:

1. Apply migrations and seed the source registry
2. Ingest live primary-source data
3. Generate the daily digest
4. Start the dashboard and inspect today, archive, items, sources, and runs
5. Run npm test

Notes

• The dashboard never pretends disabled social adapters are active.
• Daily digests are citation-backed and fall back to a quiet-day brief if no material changes are found.
• Demo data is opt-in through npm run demo:seed and is visibly labeled as sample content.