folio

Product Introduction

1. Definition: Folio is a local-first, AI-integrated desktop application designed for the advanced analysis of unstructured data and large-scale document sets. It functions as a specialized "control plane" for AI agents, categorizing it as an AI-powered document review platform and an LLM (Large Language Model) orchestration environment. Unlike cloud-based SaaS tools, Folio operates locally on the user’s machine, allowing for secure interaction with thousands of files through a tabular user interface.

2. Core Value Proposition: Folio exists to bridge the gap between massive datasets and actionable insights by providing a steerable workspace for AI agents. It addresses the limitations of standard LLM chat interfaces—such as restricted context windows and privacy concerns—by offering a local-first architecture where users "bring their own AI key." Its primary goal is to amplify human judgment through automated data transformation, multi-step research pipelines, and grounded, hallucination-free reporting across diverse file types like PDFs, JSON, Markdown, and audio files.

Main Features

1. Tabular Document Review UI: Folio organizes unstructured data into a structured, spreadsheet-style interface. This allows users to view, filter, and sort thousands of files simultaneously. Each row represents a document, while columns represent AI-generated metadata, summaries, or extracted data points. This UI enables users to oversee agentic workflows and manually verify AI outputs directly alongside the source file.

2. Multi-Step Research Pipelines: Users can define and execute complex, sequential analysis workflows powered by LLMs. This feature allows for data transformation using natural language commands (e.g., "Extract contract clauses into a structured table" followed by "Flag compliance mentions"). The pipeline architecture ensures that each step is grounded in the specific document's context, allowing for high-accuracy scaling across massive document sets.

3. MCP (Model Context Protocol) Integration: Folio is built to be compatible with the Model Context Protocol, enabling it to connect with AI agents such as Claude Code, Cursor, and Codex. This allows agents to source data from various integrations, load it into the Folio workspace, and perform tasks like labeling, summarizing, or mapping data to a master taxonomy.

4. Privacy-Centric Local Processing: The application is designed for sensitive data environments. Files are never stored on Folio’s servers; they remain on the local desktop. LLM requests are heavily isolated, ensuring that every cell in the tabular UI only has access to its specific content and proactively injected context. This "separation of concerns" prevents prompt injection attacks from accessing the local file system or API keys.

5. Heavy Operation Parsing via Modal.com: For compute-intensive tasks such as parsing large PDFs or transcribing audio files, Folio integrates with the user's own Modal.com account. Data is processed in transit and is never indexed or stored permanently in the cloud, maintaining a strict trust boundary while providing the power of serverless infrastructure.

Problems Solved

1. LLM Context Window Limitations: Traditional LLM chat interfaces struggle with "lost in the middle" phenomena and strict token limits when processing hundreds or thousands of documents. Folio solves this by using per-document orchestration, ensuring that every file receives individual LLM attention without exceeding context constraints.

2. Data Privacy and Compliance Risks: Many AI tools require users to upload sensitive documents to the cloud, which is often a deal-breaker for legal, financial, and clinical sectors. Folio eliminates this risk by keeping the data local and using API providers only for transit-based processing.

3. Hallucination and Lack of Auditability: Generative AI often fabricates details when asked to summarize large batches of information. Folio provides a grounded research process where every insight is linked back to the source document, allowing for easy verification and auditing.

4. Target Audience:

• Clinical Researchers: Analyzing trial papers, conference abstracts, and patient population data.
• Legal Professionals: Reviewing case law, litigation strategy, and contract compliance.
• Financial Analysts: Processing SEC filings, earnings transcripts, and fundamental metrics.
• GTM & Support Teams: Tagging support tickets, classifying inbound leads, and analyzing call transcripts.

5. Use Cases:

• Trial Landscape Briefs: Automatically updating summaries of new medical studies.
• Litigation Strategy Memos: Synthesizing holdings and reasoning across filtered case search results.
• Company Briefs: Extracting narrative and metrics from raw financial filings into structured reports.
• Data Normalization: Mapping disparate product names or categories to a master taxonomy using natural language.

Unique Advantages

1. Steerable AI Orchestration: Unlike autonomous agents that operate in a "black box," Folio provides a surface to see, steer, and scale. Users approve and edit the analysis pipeline at every stage, ensuring the final output matches the specific research requirements.

2. Local-First Security Architecture: Folio is built on the principle of minimal trust. By keeping files local and isolating LLM requests, it provides a more secure environment than cloud-first AI platforms, specifically protecting against data leaks and hidden injection attacks embedded in external documents.

3. Broad Data Source Compatibility: Folio can ingest data from a wide array of sources—including PubMed, ArXiv, Lex Machina, PACER, and AlphaSense—and transform it into a unified workspace. It supports a variety of data types, from raw Markdown and JSON to complex PDFs and audio files, all within the same project.

Frequently Asked Questions (FAQ)

1. Is my data used to train AI models when using Folio? No. Folio uses a "bring your own key" (BYOK) model. Your data is sent to the AI API providers (like OpenAI or Anthropic) only in transit for processing. It is not stored by Folio, and most enterprise API agreements explicitly state that data sent via API is not used for model training.

2. How does Folio handle thousands of files without crashing? Folio is a desktop application optimized for local performance. It uses per-document orchestration and integrates with Modal.com for heavy lifting like PDF parsing. This distributed approach allows it to scale to massive datasets that would overwhelm a browser-based chat interface.

3. What is MCP and how does it work with Folio? The Model Context Protocol (MCP) is an open standard that allows AI agents to connect to data sources and tools. Folio acts as a host for MCP, meaning you can use agents like Claude Code or Cursor to pull data into your Folio workspace or perform complex transformations on the files you have already imported.

4. Can I export my research from Folio? Yes. Folio allows you to export all your work, including AI-generated insights, labels, and summaries, into a standard CSV file. This makes it easy to integrate your findings into other tools like Excel, BI platforms, or internal databases.