Hebbia

A one-sentence summary

Hebbia Matrix is an AI platform designed for financial institutions and companies that engage in highly knowledge-intensive work; it enables the organization of large volumes of private documents, publicly available information, and financial databases into structured, trackable analyses as well as continuous intelligent workflow processes.

Tool Introduction

Hebbia is operated by Hebbia Inc., headquartered in New York. The company’s official website describes its product as an Institutional Intelligence platform intended for investors, bankers, consultants, and large enterprises. It emphasizes the use of financial context, cross-team collaboration, large-scale analysis, and automated processes.

Matrix is not a chatbot that merely returns a single answer; instead, it breaks down documents, questions, analysis steps, and results into rows and columns. Users can view the location of the materials related to each conclusion, receive suggestions for modifications, add additional analysis columns, and save the entire process as a reusable agent.

Current product composition

Main functions

Matrix-based document analysis

Matrix uses documents or companies as rows, and research questions and extracted fields as columns, enabling the model to process large amounts of material in parallel. Users can view the conclusions, citations, and analysis processes on a per-cell basis, which makes it easier to review and compare compared to a single long response.

Natural language-based analysis process creation

The creation wizard allows users to first select a document, after which it automatically breaks down complex research objectives into multiple columns of questions. Users do not need to write prompts repeatedly for each file, but they should still check whether the definitions in those columns cover their own criteria.

Multiple steps and subsequent columns

Follow-Up Columns allow subsequent questions to reference the results from the previous column, creating a sequential process of filtering, extracting, calculating, and making decisions. This makes the research methods more explicit, but errors in earlier steps can also be transmitted to subsequent columns.

Matrix Agent and multi-agent execution

Matrix Agent can retrieve additional materials, create analysis columns, and consolidate the entire table. Hebbia also breaks down complex requests into multiple sub-tasks to be processed in parallel, thereby supporting financial research involving large volumes of documents and numerous steps.

Purpose-Built Agents

Users can save frequently used prompts and organizational processes as Agents, share them in the organization database, and make ongoing modifications to them. Reference documents can also be converted into Agent format; once new documents are incorporated into the process, they can be analyzed automatically using the same fields.

Scheduled execution and email delivery

The Agent can operate on a daily, weekly basis or according to a custom schedule; it retrieves the connected data and generates summaries, reports, or slides. Once the task is completed, notifications can be sent via email or the results can be delivered, making it suitable for continuous monitoring and routine research.

Company search and financial research

Global Company Search is used to identify corporate entities around the world and to aggregate information from publicly available sources, expert interviews, and authorized databases. Company Matrix allows public and private companies to be compared in a single table across financial, transactional, and industry-related aspects.

Document understanding and multimodal analysis

Matrix can handle complex materials that contain text, tables, charts, images, and watermarks, and it can choose between different processing methods based on text models or visual models. Improvements in parsing help reduce the need for manual editing, but the quality of scanning and complex layouts can still lead to errors in the data fields.

Quote preview and traceable results

In Chat, Matrix, and Draft, the conclusions drawn can display the relevant sections of the text, allowing users to preview the context without leaving their current workspace. Citations facilitate verification, but they do not guarantee that the inferences, calculations, and selections of material will be accurate.

Charts, Excel, and presentations

Hebbia can generate charts from company financial data, private documents, or CRM data, and insert these charts into slides. The platform also allows for the creation of editable PowerPoint tables, the use of corporate templates, and the export of Matrix data as branded Excel files.

Team Projects

Projects puts the documents, analyses, and deliverables related to the same transaction or research project in a shared space, enabling team members to build upon the existing context. This feature will be made available to certain customers at the beginning of 2026; the actual availability will depend on the workspace in question.

Which users are it suitable for

• Asset management and hedge funds: Analysis of financial reports, conference calls, industry materials, and portfolio risks.
• Private equity and venture capital teams: screening companies, managing VDRs, and preparing materials for due diligence and the investment committee.
• Investment banking: Prepare company profiles, analyses of comparable companies, lists of potential buyers, and presentation materials for transactions.
• Private credit and credit research team: Analyzes debt terms, contracts, maturities, and downside risks.
• Wealth management and advisory firms: organize client information, conduct market research, and provide insights across different asset classes.
• Corporate strategy and business teams: Assess RFPs, sales opportunities, patents, meeting minutes, and market trends.
• Knowledge Management and Innovation Teams: Package organizational processes, databases, and best practices into shared agents.

Typical use cases

• Batch financial report analysis: Compare the drivers of growth and risks across hundreds of earnings calls or disclosure documents.
• VDR due diligence: Identifying anomalies and gaps in contracts, financial records, customer information, and operational documents.
• Credit agreement review: extraction of credit limits, interest rates, amortization, protection clauses, and incremental debt capacity.
• Company screening: A list of potential companies is created by taking into account financial indicators, industry, region, and transaction data.
• Expert interview study: Organizes the interview records from various companies for comparison using unified fields and themes.
• Investment Memorandum: Organize research tables, citations, charts, and conclusions into a draft ready for review.
• Regular monitoring: Have the Agent check for newly released information, market data, or internal data on a scheduled basis and send summaries.
• Client deliverables: Convert the analysis into Excel files, reports, and presentations featuring the organization’s brand.

Basic usage tutorial

1. Contact sales and request a demonstration using a real-world business scenario, to clarify the requirements regarding the number of users, data connections, security, and deployment.
2. The administrator creates the organizational workspace and configures identity logging in, permissions, data connections, single-tenant settings, and retention policies.
3. Create a Project or directly set up a Matrix, and select a financial database that is public or approved for use.
4. Describe the research objectives in natural language; let the creation wizard break down the columns, and then manually adjust the fields, variables, and criteria.
5. Run the analysis and examine each column for citations, calculations, time ranges, units, and missing values; add additional columns as necessary.
6. Save the matured process as an Agent, and set the sharing scope, execution frequency, output format, and delivery recipients.
7. Export Excel files, charts, reports, or slides for final approval by analysts and supervisors.

VDR due diligence workflow

1. Create isolated projects for each transaction, and ensure that team members and external consultants can only access the materials that have been authorized.
2. Import the VDR file to check for duplicates, scanning quality, folder structure, and records of failed uploads.
3. Set the document type, date, key terms, risks, opportunities, and issues to be verified as Matrix columns.
4. Use the subsequent columns to convert the initially extracted information into risk levels, supporting evidence, and review priorities.
5. Open the sections of material for each high-risk conclusion, and check the page numbers, context, version, and data consistency.
6. Organize the confirmed results into a list of issues, an investment memorandum, an Excel model, or a presentation.

Data connection and support platforms

The connections listed on the official website currently include those to cloud storage, data warehouses, financial databases, and publicly available resources. Actual data access permissions, geographic coverage, field completeness, and any additional subscription fees are determined by the customer’s own contract as well as the Hebbia deployment setup.

Model and output capabilities

Hebbia employs multi-model and multi-modal routing, selecting either text or visual models depending on the task at hand. Official updates have introduced new models from companies such as OpenAI, Anthropic, and Google, but the specific models available may vary over time, by region, and depending on customer settings.

Prices and packages

Hebbia does not list the price of standard seats, fixed monthly fees, a standard trial period, or any free usage quota on its public pricing page. Matrix adopts a business-oriented sales approach; its quotes are determined based on core user licenses, deployment, security, connectivity, APIs, and the level of usage.

It needs to be confirmed at the time of purchase.

• Minimum number of named users, contract duration, renewal procedures, and methods for price adjustments.
• Whether it includes Chat, Matrix, Agents, Projects, mobile versions, and Artifacts.
• Is additional licensing or extra costs required from customers to access third-party financial data?
• Are there any restrictions on APIs, automation, the number of executions, the amount of documentation, and concurrency?
• Costs for single-tenant, data zones, security auditing, and implementation services.
• Time limits for exporting data after termination, deleting it, and handling unused fees.

Licensing and Fair Use Rules

The Core User License must be assigned to an individual with a specific name who uses the platform on a continuous basis; it cannot be shared among multiple people, rotated among them, or used as a service account. High-volume automation, batch processing, and the resale of outputs require prior written approval, and the API cannot be used to bypass user licensing requirements.

If usage exceeds the agreed limits significantly, the platform may impose restrictions on access, require an upgrade, or suspend access. The official fair use policy generally provides ten working days to correct non-compliant behavior, but serious or repeated violations can lead to suspension or termination of access.

• It is not permitted to hold multiple positions or undertake tasks that should be authorized separately by different departments through a single role.
• Named seats must not be reallocated frequently in order to avoid purchases.
• Scripts, bots, or APIs must not be used to simulate multiple users in order to bypass the business structure.
• It is not permitted to carry out batch processing in abnormally large volumes or to resell results without written approval.
• The platform must not be used in scenarios involving high-risk systems as defined by the EU AI Act.

APIs and developer capabilities

The official rules regarding fair use explicitly cover APIs, integrations, and other forms of programmed access, indicating that enterprise customers can negotiate the relevant capabilities. At present, there are no complete interface documents available for the general public, no fixed prices for APIs, no option for registering public keys, and no universal SDKs; therefore, Hebbia cannot be considered a self-service developer API.

Teams that require programmed access should specify in the contract the endpoints, authentication methods, data transfer rates, batch size limits, data connections, auditing requirements, and permissions for output. The availability of API capabilities does not mean that the platform is open source; the scale of automation is still subject to the permissions granted by core users and to rules regarding fair use.

Suggestions for enterprise integration

1. Submit to sales the specific system, volume of data, number of users, and frequency of automation, to determine whether an API or an existing connection is required.
2. Have the security, legal, and data procurement teams work together to review third-party data licenses, DPA agreements, and sub-processors.
3. Use non-sensitive materials to verify authentication, permissions, rate limiting, error handling, and result traceability.
4. Establish service accounts, key rotation, logging, manual approval processes, and procedures for abnormal shutdowns.
5. Expand batch tasks only after obtaining written permission, to avoid systematic use that violates seat allocation or fair usage rules.

Data privacy and security

Hebbia states that it will not use customer data to train models; the data is encrypted with AES-256 when stored statically, and with TLS 1.3 during transmission. The platform lists the SOC 2 Type I and Type II certifications as well as the measures related to GDPR, and the DPA specifies the responsibilities regarding data processing, security, deletion, and auditing.

Differences in the wording between the privacy policy and the DPA

The website’s privacy policy states that personal data may be stored in U.S. databases for a period of time after the business relationship ends, for operational, historical, and archival purposes. The DPA regarding data submitted by customers requires that such data be returned or deleted at the customer’s request after the relationship ends, with exceptions allowed in cases where legal requirements dictate retention; therefore, it is necessary to distinguish between the data of website visitors and the data of customers based on the applicable contracts when making purchases.

Restrictions on sensitive data

The standard handling guidelines for DPA prohibit customers from submitting certain types of sensitive personal data, including information related to health, biometric data, political opinions, and religion. Financial or legal documents may contain such information; customers should first check whether the contract specifies otherwise and carry out data anonymization in such cases.

Product advantages

• A tabular interface is suitable for batch comparison, allowing each conclusion and citation to be reviewed individually.
• It is able to connect document retrieval, structured extraction, multi-step reasoning, and the generation of output documents.
• It supports unified research across large-scale private datasets, publicly available data, and mainstream financial databases.
• Follow-Up Columns and Agents can transform institutional methods into shared, sustainable processes.
• The capabilities related to charts, Excel, and PowerPoint are more in line with the final outputs of the investment and banking teams.
• Enterprise security, data processing protocols, single-tenant options, and audit mechanisms are suitable for evaluation by regulatory authorities.

Usage restrictions and precautions

• There are no fixed public prices or a general free version; the procurement timeline and total cost require evaluation by sales staff.
• The concept of unlimited context promoted on the official website is part of the product’s marketing message; it does not mean that there are no limits regarding capacity, time, concurrency, or fair usage.
• AI may still fail to read files properly, misinterpret charts, select the wrong time range, make calculation errors, or produce incomplete conclusions.
• Chain-like subsequent columns can amplify errors from earlier stages; therefore, critical decisions must be reviewed against the original text and data.
• Connections to financial databases are usually subject to the customer’s existing licenses, and not all data access rights are included by default.
• The models, Projects, mobile options, single-tenant capabilities, and programmatic features available to different customers may vary.
• Named seats cannot be shared by multiple people; abnormal automation and large-scale batch processing may trigger rate limits or require upgrades.
• The standard DPA imposes restrictions on certain specific types of sensitive information; classification and contract approval are required before such information can be uploaded.
• Investment research, valuation, and trading recommendations generated by AI cannot replace professional judgment, nor do they constitute any guarantee regarding investments.

Is it open source?

Hebbia Matrix is a closed-source commercial platform; there is no official open-source repository, model weights, or public SDK available for hosting a complete version of it. Other projects on the internet that are named Matrix or Hebbian have no connection to Hebbia’s product.

Enterprise APIs and data connections represent integration capabilities that are subject to contractual constraints; they do not equate to open-source code. If official developer tools become available in the future, it is necessary to verify the ownership of the repositories, the licenses in use, their maintenance status, and whether they cover the core platforms.

Basic information

Frequently Asked Questions

Is Hebbia Matrix free?

No public free version, fixed trial quota, or individual subscription price was found. Organizations need to contact sales to request a demonstration and obtain a quote based on their requirements regarding seats, modules, connections, and deployment.

What is the difference between Matrix and regular AI chat?

In ordinary conversations, the results are usually provided within a single response, whereas Matrix presents documents and analyses in tables. Users can examine each element of the table, make modifications, and save the process as an Agent.

Is it only suitable for financial teams?

The current official website focuses primarily on finance, but its tools can also be used for handling tasks related to law, corporate strategy, knowledge management, and other document-intensive activities. Teams that are not in the finance sector should still verify through demonstrations whether the industry-specific data, workflows, and costs are appropriate.

Does Hebbia train customer data?

According to the official security guidelines, customer data will not be used to train models. Companies should still check the DPA and relevant orders to understand the specifics regarding downstream models, optional functions, data retention and deletion, as well as cross-border data processing.

Does Hebbia have a public API?

The official rules mention APIs and programmed access, but there is no developer console available for public registration, no complete documentation of the interfaces, nor any fixed prices. Such capabilities must be requested through sales and included in the corporate contract.

Can Hebbia completely eliminate hallucinations?

No AI platform can be described as being error-free. Even with citations and visualizations of the processes, users still need to examine the original documents, data licenses, as well as apply their own computational skills and professional judgment.

Summary

Hebbia Matrix is suitable for institutions that need to conduct repetitive research on large amounts of financial and corporate data. Its distinguishing features include grid-based tables, citation per cell, multi-step agents, data integration, as well as a complete workflow that allows results to be exported to Excel and presentation files.

When making purchases, it is important to first verify the authenticity of the data sets, the rules related to assigned seats and APIs, any custom quotes provided, the security aspects of the contracts, as well as the costs associated with manual verification. Only by properly designing the permissions, the quality of the materials used, and the approval processes can large-scale automation be used without increasing the risks of errors and non-compliance.