Hebbia

Hebbia is an enterprise AI platform for finance and legal professionals. Its flagship product, Matrix, is a multi-agent system that ingests PDFs, spreadsheets, transcripts, contracts, and filings — across effectively unlimited volume — and returns structured, citation-backed answers to complex analytical questions. It is not a chatbot; it is designed to replace junior analyst workflows end-to-end, from document ingestion through deliverable generation. Founded 2020, headquartered in New York.

Matrix breaks complex queries into discrete analytical steps routed across a swarm of specialized AI agents. The system operates on an 'infinite effective context window' — it does not chunk or truncate; it reasons over complete document sets. Outputs are delivered in spreadsheet-style Matrix views with full source citations traceable to specific document passages. Core workflows include: M&A diligence (VDR analysis, covenant extraction, comp benchmarking), earnings call synthesis, credit agreement review, investment memo drafting, portfolio monitoring, and deal screening. The June 2025 acquisition of FlashDocs extended the platform to automated artifact generation — investment committee memos, board decks, and strip profiles output directly from the analytical pipeline, generating 10,000+ slides per day across the customer base. A 'Draft' feature allows report/presentation templates to be saved as reusable Agents for repeatable deliverables. Mobile app launched November 2025. GPT-5 integrated via Microsoft Azure AI Foundry (August 2025).

Architecture: Iterative Source Decomposition (ISD) — proprietary orchestration layer that decomposes queries, routes subtasks to best-fit models (text LLMs + vision models for charts/tables), and distills context between agents to reduce latency and maximize recall. Context Distillation Agent reduces inter-agent context by over 90%. Model-agnostic: runs OpenAI (GPT-5 via Azure), plus other frontier models selected per task type. Benchmarking: Hebbia built and published the Financial AI Benchmark for measuring model capabilities across finance-specific workflows. Data integrations: FactSet, S&P CapIQ, Preqin (via BlackRock Aladdin), Fitch Solutions, Third Bridge expert interview library, and EU regulatory filings. API access: Hebbia has an active API business (managed by FlashDocs co-founders post-acquisition); scope and documentation are enterprise-only. MCP: Hebbia consumes MCP servers from data partners as part of its 'Deeper Research' architecture, but does not offer a publicly documented outbound MCP server — it is a closed platform.

Not publicly disclosed. Enterprise-only, sales-gated — no self-serve trial or published tiers. Industry sources place per-user annual costs at $20,000+/user/year, comparable to a Bloomberg Terminal, though Hebbia has not confirmed figures. Contracts are multi-year, customized by firm size, user count, and data volume. Implementation and onboarding (delivered by a forward-deployed team of ex-bankers and lawyers) are bundled into enterprise agreements. ROI is typically framed in partner/analyst hours saved ($500–$1,500/hour) rather than seat cost.

Native data integrations: FactSet, S&P CapIQ, Preqin (via BlackRock Aladdin), Fitch Solutions, Third Bridge, EU regulatory filings. Document sources: VDRs, Google Drive, PDFs, Excel, PowerPoint, transcripts. Output formats: structured Matrix spreadsheet views, PowerPoint decks (via FlashDocs), Word/PDF reports. Mobile app (iOS/Android, launched November 2025). No published Slack, CRM, or calendar integration. Integration breadth outside financial data providers is narrow; the platform is designed as a standalone research environment, not a connective layer across a firm's existing toolstack.

Public Trust Center at trust.hebbia.ai. Commits to never training models on customer data. SOC 2 referenced in third-party and industry sources; exact Type I vs. Type II and ISO 27001 status are not explicitly confirmed in public documentation — prospective customers should request the SOC 2 report directly. Encryption at rest and in transit is standard. Adopted by the U.S. Air Force, validating the platform against government-grade security requirements. Customer base includes firms regulated by the SEC, FINRA, and state bar associations. DPA available for enterprise contracts.

Founded August 2020 in New York City by George Sivulka (sole founder and CEO), a Stanford PhD student in electrical engineering. Early team also credited Lukas Schmit and Tim Lupo. Launched Matrix in 2022. Total funding: $161M across three rounds — Seed (October 2020), Series A (June 2022, ~$31M pre-Series B), and Series B (July 2024, $130M led by Andreessen Horowitz at a $700M valuation). Series B investors include Index Ventures, GV, Peter Thiel, Eric Schmidt, and Jerry Yang; additional backers include Founders Fund and Social Capital. At Series B, ARR was reported at $13M with revenue having grown 15x over the prior 18 months. June 2025: acquired FlashDocs. Platform has processed over one billion pages. Notable customers: BlackRock, Carlyle, KKR, Centerview Partners, MetLife, Oak Hill Advisors, Fenwick & West, Seyfarth Shaw, U.S. Air Force. Over 40% of the largest asset managers by AUM are reported users.

Strong. Hebbia is the most directly applicable AI research tool for a growth equity firm conducting sector diligence, company screening, and investment memo preparation. The Matrix architecture maps precisely to core analytical workflows: ingesting CIMs, VDR documents, SEC filings, earnings transcripts, and proprietary research; cross-referencing structured data (FactSet, CapIQ, Preqin) against unstructured internal documents; and outputting IC memos and diligence summaries with full citations. The Preqin integration is particularly relevant for benchmarking private company targets against private markets comps. Reusable Agent templates for IC memos and diligence checklists allow a firm to encode its own process once and run it continuously. Government/defense customer base (U.S. Air Force) is a credibility signal for defense and advanced-tech theses. Primary limitation for AI-native ambitions: no public MCP server means Hebbia cannot be called from external LLM orchestration pipelines (Claude, Cursor), limiting composability with a firm's broader AI toolstack.

No publicly available MCP server — cannot be natively composed with external LLM agents or orchestration frameworks; limits AI-native workflow integration. No self-serve trial; full enterprise sales cycle required before evaluation. Pricing reportedly very high ($20K+/user/year), making it difficult to pilot for smaller teams. Standalone research environment — does not integrate with CRM, Slack, email, or calendar systems. API access exists but documentation and scope are not public. Narrow integration breadth outside financial data providers. Highly tailored to finance/legal document workflows; not designed for general firm operations or portfolio company support.