Technical Guide: Automating M&A Due Diligence with Due Diligence Agents

Practical Summary and Use Case

Due Diligence Agents is an open-source tool designed to automate forensic M&A due diligence by utilizing 13 AI agents to analyze data rooms. It is intended for corporate development teams, private equity firms, and legal professionals who need to synthesize findings across nine specialist domains: Legal, Finance, Commercial, ProductTech, Cybersecurity, HR, Tax, Regulatory, and ESG. The tool connects siloed information, cross-references findings, and provides citations to exact pages and quotes, aiming to accelerate the review process while maintaining an audit trail for downstream decision-making.

Prerequisites and Environment Setup

To utilize the tool, you must have Python 3.12+ installed. The system is provider-agnostic, supporting Anthropic API, AWS Bedrock, Google Vertex AI, or any Anthropic-compatible gateway. You must configure your environment with the necessary API keys. For temporary use, you can export the key as an environment variable, though a .env file is recommended for persistence. Before running a full analysis, you should verify your configuration using the dd-agents doctor command to ensure provider and model routing are correctly set up.

Preparing the Data Room

The tool requires a structured data room to function effectively. You should organize your contracts into folders by subject or counterparty. The system supports various file formats, including PDFs, Word documents, Excel spreadsheets, PowerPoint presentations, and images, with scanned PDFs processed via OCR. A sample structure is provided below:

data_room/
  SubjectGroup_A/
    Acme_Corp/
      master_agreement.pdf
      amendment_2024.pdf
    Beta_Inc/
      license_agreement.pdf
  SubjectGroup_B/
    Gamma_LLC/
      services_contract.docx
  _reference/
    buyer_overview.pdf

Configuring and Running the Pipeline

The analysis is driven by a deal-config.json file. You can generate an initial configuration automatically using dd-agents auto-config "Buyer Corp" "Target Inc" --data-room ./data_room. The configuration must include at least one focus area in the deal.focus_areas field. Once configured, the full pipeline is executed with the command dd-agents run deal-config.json. This process runs nine workstreams in parallel, applies five blocking quality gates, and produces a comprehensive set of outputs including an interactive HTML report, an Excel file, and per-subject JSON findings.

Interpreting the Output

The primary output is an interactive HTML report located at {data_room_path}/_dd/forensic-dd/runs/latest/report/dd_report.html. This report provides an executive narrative, a Go/No-Go verdict, and progressive disclosure of findings, allowing users to drill down from high-level decisions to specific domain details and evidence. Additionally, the tool generates a 16-sheet Excel report for audit trails and modeling. Users can also perform targeted searches using dd-agents search or engage in multi-turn chat sessions with the report data using dd-agents chat to query specific findings or P0 risks.

Customization and Agent Control

Specialist agents are configuration-driven, allowing users to tailor personas, focus areas, and severity calibration without modifying the core Python code. You can inspect available agents using dd-agents agents list and view an agent's specific persona and safety floor with dd-agents agents describe --agent legal. Customizations are applied by placing markdown files in a dd-config/ directory. This allows for deal-specific overrides while ensuring that safety rules remain intact. The tool also provides validation commands to lint these customizations before execution.

Security and Privacy Considerations

Due Diligence Agents is designed for local execution. All analysis runs on the user's machine, and documents are only transmitted as API calls to the configured LLM endpoint. The tool does not perform telemetry, collect usage data, or store persistent credentials; API keys are read from environment variables or .env files. Furthermore, the tool operates in a read-only mode, ensuring that no files within the data room are modified. All output artifacts are written to a separate _dd/ directory, maintaining a clear separation between source data and analysis results.

Limitations and Professional Oversight

It is critical to note that this tool does not replace professional advisors. Legal, financial, and regulatory conclusions must always be made by qualified professionals. The tool is intended to accelerate the work of teams and advisors, not to provide final legal or financial opinions. The pipeline includes quality gates to halt execution if reliability thresholds are not met, but users should treat the output as a starting point for human review. The tool is documented at https://github.com/zoharbabin/due-diligence-agents#readme.