Why OpenPitch exists

PitchBook and CB Insights cost $20k+/year — and for fast-moving AI startups, their data is often months stale, because human verification is slow. For a company growing 3× a year, a figure verified six months ago can be off by multiples.

Meanwhile, the real numbers are already public: founders state ARR on podcasts weeks before any database, funding hits SEC filings, hiring velocity reveals growth. They're just scattered, unstructured, and contradictory — exactly the problem an AI agent is built to solve.

OpenPitch's bet is latency, not coverage. For the AI companies that matter, a fresh, fully-sourced, confidence-scored number beats a verified-but-stale one. We don't claim certainty — we show you the receipts.

What you get

Ask your coding agent, get an answer with receipts:

> what's Sierra's valuation, with sources?

  Sierra — AI agents for customer service (sierra.ai)
  Valuation  $15.4B  [consensus · confidence 0.96] · as of 2026-05
  ↳ 10 public sources · Reuters · CNBC · The Information · qz.com
  ↳ $950M round closed May 2026 — led by Tiger Global and GV

(A real answer from the committed data — check it against the live dashboard.)

Every number carries its source, a confidence score, and a tracked history of how it changed.

Features

• 🎙️ Mines podcasts — founders leak metrics on podcasts before any database catches them. We transcribe and extract them.
• 🧾 Always sourced — every figure links to its origin (podcast timestamp, filing, article). No black-box numbers.
• 📊 Confidence-scored — built from source reliability, speaker authority, corroboration, and freshness (confidence decays as data ages).
• 🔀 Reconciles conflicts — when sources disagree, you get a consensus range + a contradiction flag, not a silent guess.
• 🧠 Learns which sources to trust — sources that prove right over time earn more weight.
• 🕒 Version-tracked — the git history is the audit log. See exactly how a company's reported ARR evolved.
• 📡 Composable — emits typed events other agents subscribe to (newsletters, press alerts, investor outbound).
• 🤝 A2A-discoverable — ships an A2A agent card so agent ecosystems can find and describe it.
• 🧯 Grounding — give your AI a sourced, confidence-scored fact base so it stops making up AI-company numbers.
• ⚡ 60-second install — no key, no signup; works in your agent in under a minute.
• 💸 Genuinely free — runs entirely on free tiers. No cost to run, no cost to use.

Quickstart — use it in Claude Code / Codex

No API key. No signup. No cost. The data is already built and committed; the MCP server just reads it, and your agent does the reasoning.

Fastest — zero install (reads the committed data from the public repo, no clone):

uvx openpitch-mcp

Or install the package:

pip install openpitch          # the MCP server (mcp is a core dependency)
openpitch-mcp                  # start the read-only server

Or run from a clone (for the pipeline / to rebuild data):

git clone https://github.com/Avierovich/openpitch && cd openpitch
python -m venv .venv && source .venv/bin/activate
pip install -e ".[pipeline]"   # core + pipeline LLM deps
openpitch seed                 # build the data/ database from the committed seed (offline, no key)

Then point your agent at the local server:

// MCP config (Claude Code / Codex) — zero-install via uvx:
{
  "mcpServers": {
    "openpitch": { "command": "uvx", "args": ["openpitch-mcp"] }
  }
}
// (or "command": "openpitch-mcp" if you pip-installed the package)

Ask your agent: "What's Cognition's ARR, with sources and confidence?" — it calls get_metric/get_provenance and answers from committed data (and will flag the public-source discrepancy).

Or just browse the data

• 🌐 Live dashboard — avierovich.github.io/openpitch (sourced company cards, refreshed daily) — or build locally: openpitch build-dashboard
• 📁 Raw data — data/companies/ — plain JSON, diffable, yours to use
• 🤝 A2A Agent Card — generated at dashboard/dist/.well-known/agent.json

Data status: live, refreshed daily by CI. Figures are probabilistic, public-source intelligence — every number carries its source, confidence score, and date, and open quality items are tracked in public. See the methodology and the correction workflow.

Docs

• Trust model — methodology · data policy · corrections
• Interfaces — MCP spec · events spec
• Architecture — full design doc · more product docs in docs/

How it works

  Sources              Daily pipeline (free GitHub Actions)         Interfaces
  ──────────           ───────────────────────────────────         ──────────
  Podcasts ─┐          1. select top-50 (VC-attention score)        ┌─ MCP server (local, BYO agent)
  News ─────┤    ───▶  2. collect · 3. transcribe · 4. extract ───▶ ├─ static dashboard
  SEC EDGAR ┤          5. reconcile · 6. score sources              ├─ event feed (JSONL)
  Web ──────┘          7. publish → git commit (the database)       └─ "what moved today" digest

The git repo is the database. There's no server to run. See the FRD for the full design.

Build on it (composability)

OpenPitch emits typed, confidence-scored events when something material changes — so other agents can react:

Events ship on MCP and a raw events/feed.jsonl. Schemas are versioned. See the events spec.

How we compare

OpenPitch is complementary to the incumbents, not a rip-and-replace. We win a narrow wedge; we lose on breadth and verification — and we're honest about both.

The honest pitch: the free, fresh, AI-native first look — every number sourced — before you pull the expensive verified report. For an investment decision, you still need the incumbents. Full mapping, feature matrix & pricing: docs/COMPETITIVE-ANALYSIS.md · spreadsheet.

Coverage

Global AI startups — 140+ profiled across 12 sectors (including Chinese AI labs and European names Western trackers miss), with a top 50 dynamically ranked by VC attention (valuation + funding activity — not ARR, to avoid circularity). The list moves as attention shifts; companies entering/leaving the top 50 is itself a tracked signal, and auto-discovery grows the universe daily.

MENA AI/tech segment — a dedicated regional set (an open, AI-native alternative to MAGNiTT/Wamda). Honest caveat: MENA disclosure is lighter than the US, so this segment launches with lower confidence/coverage, clearly labeled.

Seed universe: config/watchlist.yaml.

Honest disclaimer

OpenPitch is transparently probabilistic. Many figures are estimates derived from public, self-reported, sometimes-contradictory sources. We surface confidence and provenance precisely so you can judge for yourself. This is not investment advice, and figures are not guaranteed accurate. Always verify before acting.

Roadmap

• Seed universe (global AI + MENA segment) + auto-discovery (news, funding digests, 21-sector backfill, China feed)
• Core data model + reconciliation engine (confidence, consensus, contradiction) — tested
• Source adapters: podcast, news, EDGAR, company-site — tested
• Extraction stage: batched LLM claim extraction + model rotation — tested; data QA still required
• MCP server — local read-only data tools
• Daily GitHub Actions pipeline — wired for LLM, Groq transcription, and SEC user-agent secrets
• Static dashboard + company pages — generated from committed data
• Event feed — JSONL feed and digest generated from publishes
• A2A agent discovery card — generated with dashboard
• MENA adapters (regional news, free-zone registries)
• Rich-source expansion (GitHub, hiring, app-ranks) — post-PMF scaling
• v2: implied-ARR model, intra-day funding fast-lane

Contributing

Contributions welcome — especially new source adapters (one file each) and watchlist curation. See the FRD for architecture.

Who built this

OpenPitch is built and run by Mohamed Abdulhadi, a product manager — working with AI agents (Claude Code) that wrote much of the code and now operate the daily pipeline and its public data corrections. That's not a footnote; it's the product demonstrating itself: an agent-native database, built and maintained agent-natively, with every commit and correction in the open. Questions, feedback, or collaboration — connect on LinkedIn or open an issue.

License

MIT