{
  "markdown": "<div align=\"center\">\n\n# 🪧 OpenPitch\n\n**The open, real-time intelligence layer for AI startups — that any agent can build on.**\n\n*A free, open-source alternative to PitchBook & CB Insights, focused on the AI companies VCs actually care about.*\n\n`MCP-native` · `zero-cost` · `fully-sourced` · `updated daily`\n\n[![CI](https://github.com/Avierovich/openpitch/actions/workflows/ci.yml/badge.svg)](https://github.com/Avierovich/openpitch/actions/workflows/ci.yml)\n[![PyPI](https://img.shields.io/pypi/v/openpitch.svg)](https://pypi.org/project/openpitch/)\n[![Python](https://img.shields.io/pypi/pyversions/openpitch.svg)](https://pypi.org/project/openpitch/)\n[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](LICENSE)\n\n> **Status: v0.1.3 — functional.** The pipeline, reconciliation engine, MCP server, and\n> dashboard all work end-to-end. Coverage and source breadth keep growing via the daily run.\n\n![Ask your agent — get a sourced, confidence-scored answer](docs/demo/openpitch-demo.gif)\n\n[**Browse the live dashboard →**](https://avierovich.github.io/openpitch/)\n\n</div>\n\n---\n\n## Why OpenPitch exists\n\nPitchBook 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.\n\nMeanwhile, 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.\n\n**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.\n\n## What you get\n\nAsk your coding agent, get an answer with receipts:\n\n```\n> what's Sierra's valuation, with sources?\n\n  Sierra — AI agents for customer service (sierra.ai)\n  Valuation  $15.4B  [consensus · confidence 0.96] · as of 2026-05\n  ↳ 10 public sources · Reuters · CNBC · The Information · qz.com\n  ↳ $950M round closed May 2026 — led by Tiger Global and GV\n```\n\n*(A real answer from the committed data — check it against the [live dashboard](https://avierovich.github.io/openpitch/).)*\n\nEvery number carries **its source, a confidence score, and a tracked history** of how it changed.\n\n## Features\n\n- 🎙️ **Mines podcasts** — founders leak metrics on podcasts before any database catches them. We transcribe and extract them.\n- 🧾 **Always sourced** — every figure links to its origin (podcast timestamp, filing, article). No black-box numbers.\n- 📊 **Confidence-scored** — built from source reliability, speaker authority, corroboration, and freshness (confidence *decays* as data ages).\n- 🔀 **Reconciles conflicts** — when sources disagree, you get a consensus range + a contradiction flag, not a silent guess.\n- 🧠 **Learns which sources to trust** — sources that prove right over time earn more weight.\n- 🕒 **Version-tracked** — the git history *is* the audit log. See exactly how a company's reported ARR evolved.\n- 📡 **Composable** — emits typed events other agents subscribe to (newsletters, press alerts, investor outbound).\n- 🤝 **A2A-discoverable** — ships an A2A agent card so agent ecosystems can find and describe it.\n- 🧯 **Grounding** — give your AI a sourced, confidence-scored fact base so it stops making up AI-company numbers.\n- ⚡ **60-second install** — no key, no signup; works in your agent in under a minute.\n- 💸 **Genuinely free** — runs entirely on free tiers. No cost to run, no cost to use.\n\n## Quickstart — use it in Claude Code / Codex\n\n**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.\n\n**Fastest — zero install** (reads the committed data from the public repo, no clone):\n\n```bash\nuvx openpitch-mcp\n```\n\n**Or install the package:**\n\n```bash\npip install openpitch          # the MCP server (mcp is a core dependency)\nopenpitch-mcp                  # start the read-only server\n```\n\n**Or run from a clone** (for the pipeline / to rebuild data):\n\n```bash\ngit clone https://github.com/Avierovich/openpitch && cd openpitch\npython -m venv .venv && source .venv/bin/activate\npip install -e \".[pipeline]\"   # core + pipeline LLM deps\nopenpitch seed                 # build the data/ database from the committed seed (offline, no key)\n```\n\nThen point your agent at the local server:\n\n```jsonc\n// MCP config (Claude Code / Codex) — zero-install via uvx:\n{\n  \"mcpServers\": {\n    \"openpitch\": { \"command\": \"uvx\", \"args\": [\"openpitch-mcp\"] }\n  }\n}\n// (or \"command\": \"openpitch-mcp\" if you pip-installed the package)\n```\n\nAsk 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).\n\n### Or just browse the data\n- 🌐 **Live dashboard** — [avierovich.github.io/openpitch](https://avierovich.github.io/openpitch/) (sourced company cards, refreshed daily) — or build locally: `openpitch build-dashboard`\n- 📁 **Raw data** — [`data/companies/`](data/companies/) — plain JSON, diffable, yours to use\n- 🤝 **A2A Agent Card** — generated at `dashboard/dist/.well-known/agent.json`\n\n> **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](https://avierovich.github.io/openpitch/quality.html). See the [methodology](docs/METHODOLOGY.md) and the [correction workflow](docs/CORRECTIONS.md).\n\n## Docs\n\n- **Trust model** — [methodology](docs/METHODOLOGY.md) · [data policy](docs/DATA-POLICY.md) · [corrections](docs/CORRECTIONS.md)\n- **Interfaces** — [MCP spec](docs/MCP-SPEC.md) · [events spec](docs/EVENTS-SPEC.md)\n- **Architecture** — [full design doc](docs/FRD.md) · more product docs in [`docs/`](docs/)\n\n## How it works\n\n```\n  Sources              Daily pipeline (free GitHub Actions)         Interfaces\n  ──────────           ───────────────────────────────────         ──────────\n  Podcasts ─┐          1. select top-50 (VC-attention score)        ┌─ MCP server (local, BYO agent)\n  News ─────┤    ───▶  2. collect · 3. transcribe · 4. extract ───▶ ├─ static dashboard\n  SEC EDGAR ┤          5. reconcile · 6. score sources              ├─ event feed (JSONL)\n  Web ──────┘          7. publish → git commit (the database)       └─ \"what moved today\" digest\n```\n\nThe git repo **is** the database. There's no server to run. See the [FRD](docs/FRD.md) for the full design.\n\n## Build on it (composability)\n\nOpenPitch emits typed, confidence-scored **events** when something material changes — so other agents can react:\n\n| You're building… | Subscribe to | OpenPitch becomes… |\n|---|---|---|\n| A newsletter agent | all material events | your content pipeline's data source |\n| A press/PR workflow | funding/valuation events, confidence ≥ 0.8 | your \"time to call the company\" trigger |\n| Investor outbound | universe entries, growth thresholds | your targeting signal |\n\nEvents ship on MCP and a raw `events/feed.jsonl`. Schemas are versioned. See the [events spec](docs/EVENTS-SPEC.md).\n\n## How we compare\n\nOpenPitch 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.\n\n| | PitchBook / CB Insights | Crunchbase | Harmonic | MAGNiTT / Wamda | **OpenPitch** |\n|---|:--:|:--:|:--:|:--:|:--:|\n| Price | $20k–100k/yr | Freemium | Custom | $/regional | **Free & open** |\n| Freshness | Weeks–months | Variable | Days | Weeks | **Daily** |\n| In your AI agent (MCP) | ✗ | ✗ | ◐ | ✗ | **✓** |\n| Every figure sourced + confidence-scored | ◐ | ◐ | ◐ | ◐ | **✓** |\n| Contradiction detection | ✗ | ✗ | ✗ | ✗ | **✓** |\n| Coverage breadth | **✓✓✓** | **✓✓✓** | **✓✓** | ✓ (MENA) | narrow (by design) |\n| Verified, diligence-grade | **✓** | ◐ | ◐ | ◐ | ✗ (probabilistic) |\n\n**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](docs/COMPETITIVE-ANALYSIS.md) · [spreadsheet](docs/competitive-matrix.xlsx).\n\n## Coverage\n\n**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.\n\n**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.\n\nSeed universe: [`config/watchlist.yaml`](config/watchlist.yaml).\n\n## Honest disclaimer\n\nOpenPitch 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.\n\n## Roadmap\n\n- [x] Seed universe (global AI + MENA segment) + auto-discovery (news, funding digests, 21-sector backfill, China feed)\n- [x] Core data model + reconciliation engine (confidence, consensus, contradiction) — *tested*\n- [x] Source adapters: podcast, news, EDGAR, company-site — *tested*\n- [x] Extraction stage: batched LLM claim extraction + model rotation — *tested; data QA still required*\n- [x] MCP server — local read-only data tools\n- [x] Daily GitHub Actions pipeline — wired for LLM, Groq transcription, and SEC user-agent secrets\n- [x] Static dashboard + company pages — generated from committed data\n- [x] Event feed — JSONL feed and digest generated from publishes\n- [x] A2A agent discovery card — generated with dashboard\n- [ ] MENA adapters (regional news, free-zone registries)\n- [ ] Rich-source expansion (GitHub, hiring, app-ranks) — *post-PMF scaling*\n- [ ] *v2:* implied-ARR model, intra-day funding fast-lane\n\n## Contributing\n\nContributions welcome — especially **new source adapters** (one file each) and **watchlist curation**. See the [FRD](docs/FRD.md) for architecture.\n\n## Who built this\n\nOpenPitch is built and run by [Mohamed Abdulhadi](https://www.linkedin.com/in/mohamedibnomer),\na product manager — working with AI agents (Claude Code) that wrote much of the code and now\noperate the daily pipeline and its public data corrections. That's not a footnote; it's the\nproduct demonstrating itself: an agent-native database, built and maintained agent-natively,\nwith every commit and correction in the open. Questions, feedback, or collaboration —\n[connect on LinkedIn](https://www.linkedin.com/in/mohamedibnomer) or open an issue.\n\n## License\n\n[MIT](LICENSE)\n\n---\n\n<div align=\"center\">\n<sub>Built in the open. Free forever. If a number looks wrong, open an issue — provenance means you can check our work.</sub>\n\n<!-- mcp-name: io.github.Avierovich/openpitch -->\n</div>\n",
  "bytes": 11436,
  "sha": "5c76493d971799611ca0c07efb67cef13d6a3f209582d020fbc546cd19c93db3",
  "repo_slug": "avierovich/openpitch",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/mcp_io_github_avierovich_openpitch_f1e5ef57/readme"
}