CtrlK
Hebbia is an American technology company that builds AI software for knowledge-intensive professional work, primarily in financial services, law, and consulting. The company's flagship product, Matrix, enables analysts, lawyers, and researchers to query large collections of documents in plain language and receive structured, cited answers. Hebbia was founded in August 2020 by George Sivulka, a former Stanford University PhD student, and is headquartered in New York City.
As of mid-2024, Hebbia had raised $161 million in total funding, including a $130 million Series B round led by Andreessen Horowitz at a roughly $700 million valuation. The company reported annual recurring revenue of $13 million and had reached profitability at the time of the Series B. More than a third of the top 50 asset managers by assets under management were using the platform, and its customer base collectively managed $14 trillion in AUM.
Hebbia occupies a distinct position in the enterprise AI landscape: it focuses specifically on complex, multi-document reasoning tasks that general-purpose AI assistants and traditional enterprise search tools handle poorly. Its core differentiation is a proprietary architecture called Iterative Source Decomposition (ISD), which decomposes user queries into sub-tasks, runs multiple AI agents in parallel across full documents, and returns every answer with clickable citations to the exact source passage.
George Sivulka grew up showing an unusual aptitude for applied science and mathematics. By age 12 he was building lasers capable of lighting fires. At 16, he interned at NASA's Goddard Institute, where he developed multiphysics simulations using COMSOL Multiphysics to model electromagnetic wave propagation. In 2016, at 18, he enrolled at Stanford University, where he completed a Bachelor of Science in Mathematics in roughly 2.5 years, graduating with distinction. He then pursued a Master of Science in Applied Physics as a Threshold Venture Fellow before starting a fully funded PhD in Electrical Engineering focused on machine learning and theoretical neuroscience.
While working in a Stanford computational neuroscience lab, Sivulka encountered transformer-based language models and became interested in how their attention mechanisms resembled human memory retrieval. Simultaneously, he was watching friends take analyst roles at firms like Morgan Stanley and Goldman Sachs and found them visibly unhappy with the manual research burden those jobs involved. The insight that motivated Hebbia was simple and practical: analysts spent most of their time reading dense documents, extracting specific facts, and organizing those facts into tables and memos. That work was exactly the kind of pattern-matching and synthesis that large language models were beginning to do well.
Sivulka built a rough prototype as a Jupyter notebook, a neural information retrieval model he designed to help analysts search and summarize documents. He walked it into Morgan Stanley's Menlo Park office, showed it to bankers, and found more interest than he expected. In August 2020, at age 22, he took a leave of absence from the PhD program and founded Hebbia.
Hebbia's first product, launched in October 2020, was a Google Chrome extension. It worked as a semantic search plugin for in-page document reading: rather than matching exact keyword strings, it highlighted contextually relevant passages based on the user's intent. Law students, financial analysts, and Stanford researchers adopted it for in-browser research, and the organic interest confirmed a real demand for AI-assisted document work.
In 2020, the company received early funding from Peter Thiel and Floodgate Fund. The initial backing was small but gave Sivulka enough runway to iterate.
Hebbia raised a $30 million Series A in July 2022, led by Mike Volpi at Index Ventures, with participation from Radical Ventures. A noteworthy group of individual investors joined alongside the institutional round, including Jerry Yang (Yahoo co-founder), Ram Shriram (one of Google's earliest investors), Marty Chavez (Alphabet board member), Stanley Druckenmiller, Kevin Warsh, Julius Genachowski, Alexandr Wang, Pieter Abbeel, Henrique Dubugras, and Raquel Urtasun. Naval Ravikant also participated.
The Series A funded the buildout of Hebbia's engineering team and accelerated development of what would become Matrix, the company's primary product.
Hebbia launched Matrix in 2022, initially as a document analysis interface for financial services workflows. The product took a different form from the browser plugin: instead of helping users navigate a single document, Matrix let them upload large collections of files and ask complex questions across all of them simultaneously. Results appeared in a grid layout, with rows representing documents and columns representing questions or extracted data points.
Adoption among asset managers and investment banks grew steadily. By the time Hebbia was pitching investors for the Series B in early 2024, the company had grown revenue 15 times over the prior 18 months and had quintupled its headcount. It had also reached profitability, a relatively unusual position for a venture-backed AI startup at that stage.
In April 2024, Forbes named Hebbia to its AI 50 list of the most promising AI companies. Hebbia and Harvey were the only two professional services AI companies on the list that year.
In July 2024, Hebbia closed a $130 million Series B led by Andreessen Horowitz, with participation from Index Ventures, GV (Google Ventures), and Peter Thiel. The round valued the company at approximately $700 million, which worked out to roughly 54 times ARR. An additional $30 million was raised shortly after the initial close.
At the time of the Series B announcement, ARR was $13 million and the company was profitable. The customer base covered more than 30% of the top 50 asset managers and included several Tier 1 investment banks. Named customers disclosed around that period included investment bank Centerview Partners, private equity firm Charlesbank, American Industrial Partners, Towerbrook Capital Partners, and Crestline Investors. Legal firm Fenwick was also an early customer (Fenwick later represented Hebbia in a 2025 acquisition). The US Air Force was among government customers. Three major consulting firms were disclosed as clients without naming them publicly.
Alex Immerman, the Andreessen Horowitz partner who led the investment, described Hebbia as providing "an interface that reflects the way knowledge workers work" and compared its potential to Excel's transformative role during the personal computing era. The investment thesis framed a broader transition from Software-as-a-Service toward what Immerman called "Service-as-a-Software," where AI agents handle entire professional workflows rather than providing incremental assistance.
Following the Series B, Hebbia began expanding into legal services in earnest. The company hired Ryan Samii, a former M&A lawyer at Paul Hastings and founder of legal contract automation startup Standard Draft, as head of legal. Fisher Phillips was an early legal sector customer. Pharmaceutical companies were also identified as a new expansion vertical.
In June 2025, Hebbia acquired FlashDocs, a startup founded in 2024 by Morten Bruun and Adam Khakhar that specialized in generative AI slide deck creation. FlashDocs was automating more than 10,000 slides per day for enterprise customers at the time of the acquisition. The deal terms were not disclosed.
The acquisition extended Hebbia's platform from document analysis and information extraction into content generation. A common knowledge-work workflow ends with a deliverable (an investment memo, a deal summary, a board presentation), and the acquisition let Hebbia address that last step. The FlashDocs team took responsibility for Hebbia's API business and artifact generation capabilities.
Matrix is Hebbia's core product. It is a software platform that accepts documents in any format (PDFs, spreadsheets, presentations, emails, images) and responds to user queries in a tabular interface. A user might upload 200 agreements from a data room, type a question such as "What are the termination clauses in each contract?", and receive a grid where each row is a document and each column holds the extracted answer with a clickable citation to the exact source passage.
The tabular format is a deliberate design choice. Financial and legal analysis almost always involves comparing information across many sources: comparing companies across SEC filings, comparing contracts across a transaction, comparing investment opportunities across fund documents. The spreadsheet-like output maps directly to how knowledge workers already organize and present their findings.
Users can click any cell to see the precise passage from the source document that the AI used to generate the answer, along with a step-by-step explanation of the reasoning. This transparency was specifically designed for regulated industries where outputs must be auditable and every claim must be traceable to a primary source.
Matrix also supports user-created templates. Teams can save question sets, share them across an organization, and build institutional workflows on top of the platform. This creates a network effect within large enterprises: the more a firm uses Hebbia, the more tailored templates it accumulates, making the platform progressively harder to replace.
In 2025, Hebbia introduced Deeper Research, an enterprise-grade research agent that operates on a multi-agent architecture. Where Matrix works primarily with documents a user uploads, Deeper Research combines access to proprietary document libraries with external data sources including SEC filings, PitchBook profiles, S&P Capital IQ financial data, sector news, and real-time web information.
The system coordinates seven specialized agent types: an orchestrator that directs workflow, a planning agent that breaks research questions into steps, a retrieval agent that locates relevant data, a document analysis agent that extracts insights from unstructured files, a distillation agent that compresses context to fit within model constraints, a reasoning agent that refines the research process dynamically, and an output agent that synthesizes findings into final reports.
Hebbia describes the agent's strategy as "explore-exploit": it first decomposes a research question broadly to map the information landscape, then iteratively focuses on the most promising areas based on what early retrieval reveals. Single research runs can consume millions of tokens; the distillation step reduces intermediate context by more than 90% to keep the process efficient. Every claim in the output links back to an original source quotation through the ISD system.
Hebbia's core technical architecture is called Iterative Source Decomposition, or ISD. The company developed this approach as a direct response to the limitations of Retrieval-augmented generation (RAG), which was the dominant pattern for grounding large language model outputs in external documents at the time Hebbia was building its first product.
Standard RAG systems work by chunking documents into small text segments, embedding those chunks as vectors in a database, and retrieving the most semantically similar chunks when a user poses a question. The retrieved chunks are passed to the language model as context, and the model generates an answer based on them. This approach works well for simple factual lookups but breaks down on complex queries that require reasoning across many parts of a document, comparing information from different documents, or following conditional logic.
Hebbia's internal analysis found that standard RAG systems failed on approximately 84% of real-world user queries in their target domains. Chunking loses document structure and introduces context gaps; cosine similarity retrieval favors surface-level text matches over semantic relevance; and single-pass generation cannot iterate on partial answers.
ISD addresses these failures through a five-stage process:
For complex queries, the system deploys multiple AI agents in parallel, each responsible for a subset of the question or a subset of the documents. Results are then reconciled by an orchestrating layer. This parallel agent structure extends the effective context window of the system well beyond what any individual model call can handle, which Hebbia describes as achieving an "infinite effective context window" across arbitrarily large document sets.
The ISD design is specifically motivated by auditability requirements in regulated industries. A financial analyst presenting a due diligence finding to a portfolio manager, or a lawyer citing a clause to a client, needs to be able to show exactly where the information came from. Every output cell in Matrix links to a source passage and a reasoning trace, giving users a complete audit trail.
Hebbia's primary market is financial services. The company reported that more than a third of the top 50 asset managers by AUM were using Matrix as of July 2024, and that the combined AUM of its financial services customer base exceeded $14 trillion. Named customers in financial services include Centerview Partners, Charlesbank Capital Partners, American Industrial Partners, Towerbrook Capital Partners, and Crestline Investors.
The legal sector became a formal expansion target in 2024, with the company hiring dedicated legal sales staff and a head of legal with M&A practice experience. Fisher Phillips was an early law firm adopter. Law firms use Hebbia primarily for contract review, M&A due diligence, clause identification, and regulatory filing analysis.
Government and defense customers are a smaller but notable part of the base. The US Air Force has been identified as a customer. Pharmaceutical companies were identified as a future expansion vertical in the Series B announcement.
The company does not publish pricing publicly and sells through an enterprise sales model requiring a demo and custom contract negotiation. Pricing information that has appeared in secondary sources suggests annual per-user licenses at approximately $15,000 each, with volume discounts, though these figures have not been confirmed by Hebbia.
Hebbia competes primarily against enterprise search platforms, other document analysis AI tools, and market research aggregators. The competitive landscape depends on customer use case: for internal document analysis, Glean is the closest general-purpose competitor; for market intelligence combining internal and external data, AlphaSense is the dominant incumbent; for legal-specific work, Harvey is a direct rival.
| Company | Focus | Primary strength | Limitation vs Hebbia |
|---|---|---|---|
| Glean (company) | Enterprise-wide search across internal systems | Broad integration with SaaS tools; fast setup | Not designed for complex multi-document reasoning or regulated industry workflows |
| AlphaSense | Market intelligence with curated external content | Large library of earnings calls, SEC filings, broker research | Primarily external data; weaker on user's own document sets |
| Harvey | AI for legal work (contracts, litigation) | Deep legal workflow integration; document drafting | Legal-only; no finance capability |
| Microsoft Copilot | General productivity AI within Microsoft 365 | Native integration with Office tools; broad enterprise reach | General purpose; no specialized finance/legal reasoning |
| IBM Watson Discovery | Enterprise document search API | Established vendor relationships; enterprise support infrastructure | API-based, requires integration work; less purpose-built for complex reasoning |
The clearest distinction between Hebbia and Glean is the depth-versus-breadth tradeoff. Glean is designed to help any employee in a large organization find information quickly across all connected systems, from Slack messages to code repositories to sales CRM records. It typically takes under two hours to set up and requires no engineering configuration. Hebbia is designed for a narrower set of users (financial analysts, lawyers, researchers) who need to extract structured insights from dense documents and be able to audit every claim in the output. Glean's AI is not designed for multi-document reasoning or compliance-grade citation; Hebbia's platform is specifically built around those requirements.
AlphaSense and Hebbia serve overlapping customers in financial services but address different parts of the research workflow. AlphaSense aggregates public market intelligence: earnings call transcripts, SEC filings, broker research reports, and news. It is strong for monitoring a universe of public companies and searching curated external content. Hebbia analyzes documents that users bring to the platform: data room files, internal research, fund documents, contracts. The two tools are sometimes used together by large asset managers, with AlphaSense covering external research and Hebbia covering internal document analysis.
Harvey and Hebbia overlap in legal services but differ in their scope. Harvey is built specifically for legal professionals and integrates more deeply with law firm practice management systems. It also offers document drafting capabilities that Hebbia did not have natively before the FlashDocs acquisition. Hebbia's advantage in the legal market is that law firms with existing deployments in financial services can use a single platform across both practice areas.
Mergers and acquisitions due diligence is Hebbia's most cited use case. In a typical deal, a target company uploads thousands of documents to a virtual data room: financial statements, contracts, employment agreements, IP filings, regulatory correspondence. Acquirer-side analysts must review these materials under time pressure, often in days rather than weeks. Before AI tools, this required teams of junior analysts working through the night.
With Matrix, a private equity or investment banking team can upload the entire data room, run a standardized question set across all documents simultaneously, and receive a populated grid in minutes. A question like "Does this contract contain a change-of-control clause?" returns answers from every contract in the data room with citations. A user reported that data room synthesis that previously required ten hours was taking approximately two hours with Hebbia; the a16z announcement noted that analyses previously taking "2-3 hours were now taking 2-3 minutes" for some tasks.
Hebbia's customer base managing $14 trillion in AUM is primarily using the platform for this workflow. The 30% adoption rate among the top 50 asset managers reflects how concentrated Hebbia's penetration is in private equity, hedge funds, and asset management.
Asset managers use Matrix to analyze fund documents, limited partnership agreements, earnings call transcripts, equity research reports, and regulatory filings. A portfolio manager might ask Matrix to pull specific risk disclosures from ten years of 10-K filings for a group of portfolio companies, or to compare revenue recognition policies across five prospective investments.
Hebbia's Deeper Research agent extends this use case to include external data sources. A researcher can pose a question that spans both proprietary internal documents and public information (SEC filings, financial databases, news), receiving a synthesized report with sources from both domains.
Law firms use Matrix for large-scale contract review, both in transactional contexts (reviewing all contracts in a deal data room) and in standing legal matters (reviewing a client's entire contract portfolio for a specific clause type). Clause identification across thousands of documents is particularly well-suited to the Matrix grid interface, where each row is a contract and each column is a specific clause or provision.
E-discovery is another legal application. Litigation teams often must review large volumes of documents for relevance and privilege determinations under strict deadlines.
Regulatory professionals use Hebbia to process large bodies of regulatory text, proposed rules, and comment letters. Government clients have used it to analyze structured submissions and supporting documents. The citation-first design is particularly useful in regulatory work, where every analytical finding must be tied to the specific statutory or regulatory text that supports it.
Hebbia has received generally positive attention from the investment and analyst communities covering enterprise AI, primarily because it achieved profitability and rapid revenue growth in a segment where many AI startups were burning capital aggressively. The 15x revenue growth over 18 months prior to the Series B was frequently cited as evidence that the product had genuine enterprise fit.
The Forbes AI 50 listing in April 2024 brought additional mainstream attention. Within professional services AI, the parallel inclusion of Hebbia and Harvey on the list positioned them as the two most credible pure-play AI tools for finance and law respectively.
User reviews from practitioners are more mixed. On review platforms, some users describe Matrix as transformative for their workflows, particularly for data room synthesis and investment memo analysis. Others report reliability problems, noting that outputs sometimes require manual verification against source files and that the AI layer can occasionally produce answers that are less useful than a direct query to a general-purpose model. One criticism that appeared on professional review sites was that integration with external tools (SharePoint, Excel, Google Drive) was incomplete or cumbersome in 2024. Hebbia has described integrations as ongoing development priorities.
The pricing model has attracted criticism for being opaque and inaccessible to smaller firms. There is no public pricing page, no free trial, and no self-serve onboarding. The enterprise-only sales motion with long contract cycles limits the addressable market to large institutions with dedicated procurement processes. Smaller law firms and boutique investment managers that might benefit from the product are effectively excluded from the current distribution model.
Several limitations are recognized by observers and users of Hebbia's platform.
Accuracy and hallucination risk remain a concern despite the ISD architecture's validation steps. While every answer includes citations, users have reported cases where the cited passage does not fully support the generated claim, requiring manual verification. The platform's own guidance acknowledges that human review of outputs is best practice for high-stakes decisions.
Hebbia is a static document analysis tool, not a real-time data platform. It does not provide live market data, streaming news, or integration with live databases. In financial services, this limits its use for workflows that depend on current prices, recent filings, or time-sensitive news. The Deeper Research agent adds some real-time web search capability but is not designed to replace dedicated market data providers.
The platform is not built for workflow automation beyond research and document generation. It cannot take actions in other systems, execute trades, file documents, or trigger downstream processes. Competitors focused on agentic workflow automation occupy a different space.
Integration with enterprise software stacks was limited through 2024. Uploading documents from SharePoint, Salesforce, or other enterprise systems required manual work in some cases. The company has indicated that native integrations are under active development.
Hebbia has not established the kind of cloud marketplace partnerships that competitors like Glean have built with Amazon Bedrock and Google Cloud. This may limit its ability to reach customers through cloud procurement channels that large enterprises increasingly use to consolidate software spending.
The company's growth has been heavily concentrated in US financial services. International expansion and penetration into consulting and pharmaceutical verticals remained early-stage as of the Series B.