# πŸͺ§ OpenPitch **The open, real-time intelligence layer for AI startups β€” that any agent can build on.** *A free, open-source alternative to PitchBook & CB Insights, focused on the AI companies VCs actually care about.* `MCP-native` Β· `zero-cost` Β· `fully-sourced` Β· `updated daily` [![CI](https://github.com/Avierovich/openpitch/actions/workflows/ci.yml/badge.svg)](https://github.com/Avierovich/openpitch/actions/workflows/ci.yml) [![PyPI](https://img.shields.io/pypi/v/openpitch.svg)](https://pypi.org/project/openpitch/) [![Python](https://img.shields.io/pypi/pyversions/openpitch.svg)](https://pypi.org/project/openpitch/) [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](LICENSE) > **Status: v0.1.3 β€” functional.** The pipeline, reconciliation engine, MCP server, and > dashboard all work end-to-end. Coverage and source breadth keep growing via the daily run. ![Ask your agent β€” get a sourced, confidence-scored answer](docs/demo/openpitch-demo.gif) [**Browse the live dashboard β†’**](https://avierovich.github.io/openpitch/)
--- ## Why OpenPitch exists PitchBook and CB Insights cost **$20k+/year** β€” and for fast-moving AI startups, their data is often **months stale**, because human verification is slow. For a company growing 3Γ— 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 β€” 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 β€” 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 β€” AI agents for customer service (sierra.ai) Valuation $15.4B [consensus Β· confidence 0.96] Β· as of 2026-05 ↳ 10 public sources Β· Reuters Β· CNBC Β· The Information Β· qz.com ↳ $950M round closed May 2026 β€” led by Tiger Global and GV ``` *(A real answer from the committed data β€” check it against the [live dashboard](https://avierovich.github.io/openpitch/).)* Every number carries **its source, a confidence score, and a tracked history** of how it changed. ## Features - πŸŽ™οΈ **Mines podcasts** β€” founders leak metrics on podcasts before any database catches them. We transcribe and extract them. - 🧾 **Always sourced** β€” every figure links to its origin (podcast timestamp, filing, article). No black-box numbers. - πŸ“Š **Confidence-scored** β€” built from source reliability, speaker authority, corroboration, and freshness (confidence *decays* as data ages). - πŸ”€ **Reconciles conflicts** β€” when sources disagree, you get a consensus range + a contradiction flag, not a silent guess. - 🧠 **Learns which sources to trust** β€” sources that prove right over time earn more weight. - πŸ•’ **Version-tracked** β€” the git history *is* the audit log. See exactly how a company's reported ARR evolved. - πŸ“‘ **Composable** β€” emits typed events other agents subscribe to (newsletters, press alerts, investor outbound). - 🀝 **A2A-discoverable** β€” ships an A2A agent card so agent ecosystems can find and describe it. - 🧯 **Grounding** β€” give your AI a sourced, confidence-scored fact base so it stops making up AI-company numbers. - ⚑ **60-second install** β€” no key, no signup; works in your agent in under a minute. - πŸ’Έ **Genuinely free** β€” runs entirely on free tiers. No cost to run, no cost to use. ## Quickstart β€” 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 β€” zero install** (reads the committed data from the public repo, no clone): ```bash uvx openpitch-mcp ``` **Or install the package:** ```bash 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): ```bash 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: ```jsonc // MCP config (Claude Code / Codex) β€” 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?"* β€” 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** β€” [avierovich.github.io/openpitch](https://avierovich.github.io/openpitch/) (sourced company cards, refreshed daily) β€” or build locally: `openpitch build-dashboard` - πŸ“ **Raw data** β€” [`data/companies/`](data/companies/) β€” plain JSON, diffable, yours to use - 🀝 **A2A Agent Card** β€” generated at `dashboard/dist/.well-known/agent.json` > **Data status:** live, refreshed daily by CI. Figures are **probabilistic, public-source intelligence** β€” 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). ## Docs - **Trust model** β€” [methodology](docs/METHODOLOGY.md) Β· [data policy](docs/DATA-POLICY.md) Β· [corrections](docs/CORRECTIONS.md) - **Interfaces** β€” [MCP spec](docs/MCP-SPEC.md) Β· [events spec](docs/EVENTS-SPEC.md) - **Architecture** β€” [full design doc](docs/FRD.md) Β· more product docs in [`docs/`](docs/) ## How it works ``` Sources Daily pipeline (free GitHub Actions) Interfaces ────────── ─────────────────────────────────── ────────── Podcasts ─┐ 1. select top-50 (VC-attention score) β”Œβ”€ MCP server (local, BYO agent) News ────── ───▢ 2. collect Β· 3. transcribe Β· 4. extract ───▢ β”œβ”€ static dashboard SEC EDGAR ─ 5. reconcile Β· 6. score sources β”œβ”€ event feed (JSONL) Web β”€β”€β”€β”€β”€β”€β”˜ 7. publish β†’ git commit (the database) └─ "what moved today" digest ``` The git repo **is** the database. There's no server to run. See the [FRD](docs/FRD.md) for the full design. ## Build on it (composability) OpenPitch emits typed, confidence-scored **events** when something material changes β€” so other agents can react: | You're building… | Subscribe to | OpenPitch becomes… | |---|---|---| | A newsletter agent | all material events | your content pipeline's data source | | A press/PR workflow | funding/valuation events, confidence β‰₯ 0.8 | your "time to call the company" trigger | | Investor outbound | universe entries, growth thresholds | your targeting signal | Events ship on MCP and a raw `events/feed.jsonl`. Schemas are versioned. See the [events spec](docs/EVENTS-SPEC.md). ## 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 β€” and we're honest about both. | | PitchBook / CB Insights | Crunchbase | Harmonic | MAGNiTT / Wamda | **OpenPitch** | |---|:--:|:--:|:--:|:--:|:--:| | Price | $20k–100k/yr | Freemium | Custom | $/regional | **Free & open** | | Freshness | Weeks–months | Variable | Days | Weeks | **Daily** | | In your AI agent (MCP) | βœ— | βœ— | ◐ | βœ— | **βœ“** | | Every figure sourced + confidence-scored | ◐ | ◐ | ◐ | ◐ | **βœ“** | | Contradiction detection | βœ— | βœ— | βœ— | βœ— | **βœ“** | | Coverage breadth | **βœ“βœ“βœ“** | **βœ“βœ“βœ“** | **βœ“βœ“** | βœ“ (MENA) | narrow (by design) | | Verified, diligence-grade | **βœ“** | ◐ | ◐ | ◐ | βœ— (probabilistic) | **The honest pitch:** *the free, fresh, AI-native first look β€” every number sourced β€” 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) Β· [spreadsheet](docs/competitive-matrix.xlsx). ## Coverage **Global AI startups** β€” **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 β€” *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** β€” 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`](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 - [x] Seed universe (global AI + MENA segment) + auto-discovery (news, funding digests, 21-sector backfill, China feed) - [x] Core data model + reconciliation engine (confidence, consensus, contradiction) β€” *tested* - [x] Source adapters: podcast, news, EDGAR, company-site β€” *tested* - [x] Extraction stage: batched LLM claim extraction + model rotation β€” *tested; data QA still required* - [x] MCP server β€” local read-only data tools - [x] Daily GitHub Actions pipeline β€” wired for LLM, Groq transcription, and SEC user-agent secrets - [x] Static dashboard + company pages β€” generated from committed data - [x] Event feed β€” JSONL feed and digest generated from publishes - [x] A2A agent discovery card β€” generated with dashboard - [ ] MENA adapters (regional news, free-zone registries) - [ ] Rich-source expansion (GitHub, hiring, app-ranks) β€” *post-PMF scaling* - [ ] *v2:* implied-ARR model, intra-day funding fast-lane ## Contributing Contributions welcome β€” especially **new source adapters** (one file each) and **watchlist curation**. See the [FRD](docs/FRD.md) for architecture. ## Who built this OpenPitch is built and run by [Mohamed Abdulhadi](https://www.linkedin.com/in/mohamedibnomer), a product manager β€” 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 β€” [connect on LinkedIn](https://www.linkedin.com/in/mohamedibnomer) or open an issue. ## License [MIT](LICENSE) ---
Built in the open. Free forever. If a number looks wrong, open an issue β€” provenance means you can check our work.