# Folio: build a digital twin of your organization, without the platform

> Start modeling on your laptop in minutes. Grow it into an ecosystem of agent context your whole organization can rely on.

Source: https://semvia.io/blog/introducing-folio

Today we're releasing [Folio](https://semvia.io/download) in public beta for macOS, Windows and Linux. Folio is a desktop workspace where a subject matter expert and their agent build a domain model, bind it to the systems that already hold the data, and serve the result back for humans and agents to query: a digital twin of how your organization works.

## Folio in action

<video class="product-media block h-auto w-full rounded-xl border border-edge" poster="https://semvia.io/media/folio-model-a-database-poster.png" width="1440" height="810" preload="metadata" aria-label="An 84-second screen recording of Claude connecting Folio to a Postgres film-licensing database, designing a domain model, mapping the tables onto it, and querying for licenses that expire by March 2027." muted playsinline controls><source src="https://semvia.io/media/folio-model-a-database.mp4" type="video/mp4" /><source src="https://semvia.io/media/folio-model-a-database.webm" type="video/webm" /></video>

In the video, Claude Code running Claude Sonnet 4.5:

1. Connects Folio to a Postgres database of film licenses
2. Creates a basic domain model
3. Maps the tables onto that domain model
4. Queries for the licenses that expire by March 2027

None of that data is copied into the workspace. The licenses stay in Postgres, and the same domain model can bind a second source, like a [public API](https://semvia.io/docs/federation/query-a-json-api), and join across both in one query.

## The search for a context layer

The largest data and AI companies are all building the same thing: a layer of domain knowledge between their data and their agents.

- **Anthropic's data team** runs about 95% of its internal analytics questions through Claude at about 95% accuracy. Without curated context, accuracy sat at or below 21%. ([source](https://claude.com/blog/how-anthropic-enables-self-service-data-analytics-with-claude))
- **NVIDIA** launched its [Agent Knowledge Foundation](https://blogs.nvidia.com/blog/enterprise-platforms-agent-toolkit/) on the premise that "agents are only as useful as what they know."
- **Microsoft, Google, Databricks and Snowflake** ship semantic layers their agents ground on. Snowflake, Salesforce, dbt Labs and BlackRock started the [Open Semantic Interchange](https://snowflake.com/en/news/press-releases/snowflake-salesforce-dbt-labs-and-more-revolutionize-data-readiness-for-ai-with-open-semantic-interchange-initiative) so a metric's meaning can move between tools.

The vocabulary differs (semantic layer, ontology, knowledge graph, context layer), and so do the answers. Most either live inside one platform, describing that platform's data in its vendor's format, or attach descriptions to tables and columns: text an agent can read, but not structure it can query or build on.

## The hard part

Context is a socio-technical problem. It only has value if people contribute to it, and people only contribute when the barrier to entry is low and the payoff is immediately felt. Even when an organization has the will to capture context, how does it do it? If a subject matter expert has the time and the mandate to capture the domain they know, what do you have them do? How do you capture their understanding in a way that multiplies their impact rather than wasting their time?

## How Folio helps

Many platforms want you to build a semantic layer. Folio helps you build a semantic foundation: a domain model your experts and agents build on, rather than descriptions laid over your data.

In Folio, a subject matter expert and their agent build a domain model together in plain text files: the vocabularies, taxonomies, and entities in the domain, how they relate, and how they map to existing data systems. The expert brings what they know: what a customer means, which revenue figure is the right one, and when a license counts as expired. The agent accelerates the modeling and mapping work that captures that knowledge, proposing classes and properties, writing the bindings to data sources, and validating the result. Every step is visible to the expert, who can query and test it as they go and tweak it by hand as needed. The expert explains what a customer means once, and every agent and every query from then on gets it right, so nobody has to ask them again. Publish that domain model as a package, and every team that installs it gets it right too. Because the domain model is a graph, the context grows by connecting to what is already there: a new class links to existing ones, and a new source binds to the classes it shares. Agents query and traverse it, and Folio validates it, rather than reading one more description and guessing.

## No platform required

Starting a knowledge graph project usually means picking a vendor, standing up a graph database, building a pipeline to copy data into it, adding a validator and an inference engine, and finding someone fluent enough in all of it to glue the pieces together, all before the first useful query. And once the data is in, it's a copy, falling behind its source systems from day one. Most teams, and most domain experts, never get that far.

[DuckDB](https://duckdb.org/) changed this for analytics. Before it, serious analysis meant a warehouse. After it, you could point a query at a Parquet file on your laptop and have an answer in seconds.

Folio is that idea applied to a knowledge graph. It's a single application you install and run on your own computer: no server, no graph database, no pipeline. Everything lives in one workspace, already fitted together, and the agent working beside the expert is fluent in the standards so the expert doesn't have to be. It takes minutes to go from download to a query over your own data.

## Maintain it like code

Starting is half of the challenge. The other half is keeping it up to date. [Anthropic's data team](https://claude.com/blog/how-anthropic-enables-self-service-data-analytics-with-claude) watched accuracy on their test questions drift from about 95% at launch to about 65% in a month, because their data model changed daily and the context describing it didn't keep up. Context rots. What fixed it for them was treating it like engineering work: corrections arrive as pull requests, and the person who owns a definition reviews and merges the change.

Software teams have kept large shared codebases up to date this way for decades, and Folio uses the same practices. A domain model in Folio is a set of plain text files, so you can diff it, review it and keep it in source control, alongside the code for the systems it describes. Validation flags a breaking change the way a failing test does. Semantic packages use [semantic versioning](https://semver.org/), like software libraries. Because all of this follows the usual development workflow, your agent already knows how it works. The agent proposes a change; the expert who knows that part of the domain reviews it before it lands.

## What's in Folio

Folio brings together open standards, time-tested development practices, and modern agents over the [Model Context Protocol (MCP)](https://modelcontextprotocol.io/specification):

- **Self-describing knowledge graphs.** Your domain model lives in an [RDF](https://semvia.io/blog/what-is-rdf)-based knowledge graph, built from the workspace's Turtle, Markdown and Excel files plus any packages it depends on.
- **[Graph-based inference](https://semvia.io/docs/modeling/inference).** Declare a solar installation a [subclass](https://semvia.io/docs/modeling/inference#subclasses) of power plant, and every solar installation answers as a power plant, without spending a token.
- **Real-time editing**, with completion that knows your domain model.
- **Real-time validation**, so you know the moment you or your agent makes a mistake.
- **Real-time collaboration** with any agent that supports MCP.
- **Industry-standard vocabularies** for describing a domain and mapping it to existing systems: [SHACL](https://www.w3.org/TR/shacl/) and [RDFS](https://www.w3.org/TR/rdf11-schema/) for classes and properties, [RML](https://rml.io/specs/rml/) for mappings, [SKOS](https://www.w3.org/TR/skos-reference/) for controlled vocabularies and taxonomies, and [WoT Thing Descriptions](https://www.w3.org/TR/wot-thing-description11/) for APIs.
- **SPARQL** for querying the systems you already run, through the domain model, with SQL views where a mapping needs to reshape the data first:
  - [Databases](https://semvia.io/docs/federation/query-a-database), through [Apache Arrow ADBC](https://arrow.apache.org/adbc/) drivers for Postgres, Snowflake, Databricks, BigQuery and [more](https://docs.columnar.tech/dbc/guides/finding_drivers/)
  - [Web APIs](https://semvia.io/docs/federation/query-a-json-api)
  - Data lakes, as [Apache Iceberg](https://iceberg.apache.org/) and [Delta Lake](https://delta.io/) tables
  - [CSV, Parquet and Excel files](https://semvia.io/docs/federation/query-data-files), in your workspace, a cloud bucket or at a URL
- **Visualizations** that make a domain legible at a glance. Agents love text, but humans love pictures.
- **Package management.** Semantic packages bundle part of a knowledge graph so others can install and reuse it, [published](https://semvia.io/docs/quickstart/publish-and-install-a-package) to an S3-compatible bucket or a git repository.
- **Secrets management** that keeps credentials out of your files.

## Start with one process

Nobody models an enterprise in one sitting, and you shouldn't try. [Start where you are](https://semvia.io/blog/data-products): one process, one system, one question you're tired of answering by hand. Bind the sources that question needs. The next question reuses what the first one built, and the same workspace grows toward the whole enterprise one binding at a time.

## Free for individuals and smaller organizations

Folio is free for:

- **Individuals**
- **Companies** with under $10M in annual revenue and under $10M in total funding raised
- **Nonprofits and governments** with an annual operating budget under $10M

Larger organizations get a 30-day evaluation, and a [commercial license](https://semvia.io/#contact) after that.

Folio is closed source for now. We're a small company, and keeping the core closed is how we stay around long enough to keep building it. Folio stands on open standards and open-source libraries, and everything you write in it stays in those standards, in plain files you own.

## Try it

- **[Download Folio](https://semvia.io/download)** for macOS, Windows or Linux.
- **[Read the docs](https://semvia.io/docs)**, starting with [Welcome to Folio](https://semvia.io/docs/quickstart/welcome-to-folio) and [Connect your agent](https://semvia.io/docs/quickstart/connect-your-agent).
- **Talk to us** on [Discord](https://discord.gg/rfTfQg3zx), and use Folio's in-app link to report what breaks. We read every report.

Folio is in beta, so features may change between releases. Keep backups of your workspace, and tell us if you run into any issues.

We're also looking for a small number of design partners: teams with a domain spread across databases, APIs, data lakes and spreadsheets who want to model it with us. If that's you, [get in touch](https://semvia.io/#contact). Tell us what your organization knows, and what it keeps forgetting.
