EnterpriseA private OpenData instance for your team

Your data has the answers.
Now anyone can ask.

OpenData's public catalog is public by design. A private instance brings the same engine to the data that can't live there: your spreadsheets, databases, and warehouse tables, cataloged like the 570+ public datasets but visible only to your org. Then anyone asks a question in Claude or ChatGPT and gets a charted answer spanning both.

Full public catalog includedYour data stays in your instanceWorks in Claude and ChatGPT
The public catalog comes standard
570+ datasets across 104 sources
How it works

From a pile of files to answered questions.

Your data goes through the same three steps the public catalog does, without ever becoming public. Then the whole team can ask it questions in chat.

01

Connect your data

CSVs, spreadsheets, Postgres, warehouse exports. The same ingestion pipeline that handles the public catalog runs on yours, inside your instance, where none of it is public.

02

It becomes a catalog

Every dataset gets a schema, a plain-English description, and joins to the public catalog through the knowledge graph. Searchable by anyone, not just the data team.

03

Ask it a question

Connect Claude, ChatGPT, or Cursor once. Questions come back as answers with charts, right in the conversation.

mcp · tryopendata.aiCONNECTED · 7 TOOLS

Simulated MCP conversation.

Simulated conversation. Caprock Foods is a fictional company.

Ask about your data, right where your team already works.

Analysts and non-analysts alike ask from Claude, ChatGPT, or Cursor: what a site cost last quarter, which vendors overlap, how a number moved. Answers draw on your private datasets and the full public catalog in one conversation, so a question about your numbers and a question about the market get the same treatment.

ClaudeDesktop · ready
ChatGPTConnectors · ready
CursorMCP · ready
  • Works with your data. Not just the public catalog. Your own spreadsheets, databases, and warehouse exports are ingested and searchable in the same chat.
  • Nothing new to install. Your team keeps using Claude, ChatGPT, or Cursor. Point it at your instance and both your data and the public catalog show up.
  • Tables and charts inside the chat. Answers come back as sortable tables and styled charts, not walls of JSON. Rename a title, export a PNG, keep going.
  • Scoped to your instance. Your keys, your usage, your datasets. Access stays inside your instance, and nothing is shared with other customers.
tools registeredsearch_datasetsget_dataset_infoquery_sqlopenchartApache 2.0Streamable HTTP

The people who turn a private instance into answers.

Four shapes of buyer, one instance. Each card is the question they would type.

Strategy & operations

See how every location is really doing

Regional performance lives in exports, dashboards, and analysts' heads, so the answer takes a week to assemble. With your data in the instance, anyone can compare stores against local demographics and the broader market in one question.

They ask

Which of our regions underperformed the market last quarter, and what do those areas have in common?

A sourced answer in the conversation, not a dashboard request
Consulting & research firms

Client data and public benchmarks in one place

Every engagement means stitching a client's spreadsheets to census, labor, and economic data by hand. Load both into one instance and the whole team can ask questions that span client numbers and public context.

They ask

How does this client's regional footprint compare to population and income growth in the same counties?

Findings for the deck without a data-prep sprint
Finance & FP&A

Answer the board question before the meeting

A board member asks something the current deck doesn't cover, and it turns into a two-day fire drill across finance and analytics. When your actuals sit next to public economic data, you get the answer and a chart while you're still on the call.

They ask

Are our margins tracking with inflation and regional wage trends where we operate?

Turnaround measured in minutes, with the source attached
Healthcare & regulated industries

Query the data that can't leave your walls

A hospital's claims, payer mix, and financials have no business on a public data platform, so every question waits on the analytics backlog. A private instance keeps that data in an environment you control and gives the same plain-English answers, inside the boundary your SOC 2 and HIPAA obligations draw.

They ask

How did our claim denial rates trend by payer this year, and which service lines got hit hardest?

Sourced answers from data that never touches a public platform

Your datasets join the same graph.

When your data lands in the instance, it gets crosswalked against the public catalog on the keys they share: state, county, year, industry. So a single question can span your numbers and the government's, and the answer comes back already joined.

How the assistant finds the join

When someone asks a question that spans datasets, this graph is how the assistant knows which ones connect and on what key. Your private datasets become nodes on the same graph, so they join in exactly the same way.

For your data team, when they want it

Everything the chat assistant does is an ordinary API underneath. The same instance speaks Python, REST, SQL, and MCP, so your engineers can drop down to code whenever they need to.

from opendata import client
od = client(api_key=os.environ["OD_KEY"])
# your private sales x public demographics
# ("caprock" is a fictional example company)
df = od.query(
"caprock/store-sales-weekly",
fields=["store", "region", "revenue",
"county.population", "county.median_income"],
join={"census": "county"},
where="quarter == '2025-Q1'",
).to_pandas()

Ingestion, search, graph joins, MCP, and the public catalog, in one instance.

Six capabilities shipped today, two on deck. Nothing tagged “Shipped” is a roadmap promise.

Your data, ingested like the catalog

CSVs, spreadsheets, Postgres, and warehouse exports run through the same pipeline as the public catalog and come out documented and queryable, without any of it becoming public.

Shipped

A catalog anyone can search

Every dataset gets a schema, a plain-English description, and sample queries automatically, so people can find what they need without asking an analyst.

Shipped

Full public catalog included

The 570+ datasets across 104 sources, Census to BLS to CDC, come standard with every instance.

Shipped

Knowledge graph joins

Your datasets are crosswalked against public data on the same graph, joined on state, county, year, and industry, so one question can span both.

Shipped

MCP for the whole org

Claude, ChatGPT, and Cursor connect to your instance. Anyone can ask a question and get an answer back with charts.

Shipped

Every interface your team expects

REST API, a SQL sandbox, and Parquet or CSV exports for the people who want to work with the data directly.

Shipped

SSO & audit trails

Sign-in via your identity provider with audit logs for every query and export. No timeline commitment yet, but talk to us if it's a blocker.

Coming soon

Federated queries

Query your warehouse where it lives instead of ingesting a copy. Currently in design.

Coming soon

Your data, answerable by anyone.

BI dashboards need an analyst to build them, and plain AI chat can't see your numbers, which were never going on a public platform anyway. A private instance lets the whole team ask questions and get sourced answers that span your data and the public catalog.

Capability
OpenData private instance
BI dashboards
AI chat alone
Anyone can ask a question in chat
Analysts build, others wait
Sees your company's data
—
570+ public datasets built in
—
Web knowledge, no source data
Private and public joined in one answer
Manual data prep
—
Charts rendered in the conversation
Separate dashboard
No live data behind them
New question to new answer
Minutes
Dashboard request queue
Instant but unsourced
Answers cite the underlying dataset
Varies
—
Data stays in your instance
Depends on plan

A dashboard answers the question it was built for. AI chat answers anything but has nothing behind it. And OpenData itself is public by design. A private instance brings the same engine to the data you could never publish, so anyone can ask a new question and get a cited answer from an environment you control.

A private instance, or just the public catalog.

Most teams want their own data in the mix, so the private instance leads. If you only need the public catalog, the team plan is there.

Private instance

Your own OpenData instance, run by us: your private data plus the full public catalog, answerable by everyone in the org.

from$5,000/ month
Fully managed · unlimited seats · scales with data and support
  • Hosted and operated by us: infrastructure, ingestion, and updates
  • Full public catalog (570+ datasets)
  • Private data ingestion: spreadsheets, databases, warehouse exports
  • MCP access for the whole org (Claude, ChatGPT, Cursor)
  • Founder-led onboarding
  • Priority support with direct founder access

Team plan

Just the public catalog, with every interface your team expects.

$99/ seat / month
5-seat minimum · $495/mo to start
  • Full public catalog, no private data
  • REST API, SQL sandbox, Parquet and CSV exports
  • MCP for Claude, ChatGPT, and Cursor
  • Knowledge graph joins across the catalog
  • Academic and journalism pricing: 50% off

What it saves you

Conservative estimate of the time your people spend waiting on exports, dashboard queues, and data prep. Most teams recover more than this once the whole org can just ask.

Monthly cost$990
Monthly value created$31,176
Net monthly savings$30,186

Pays for itself once your team recovers roughly 8 hours a month that would otherwise go to exports, dashboard queues, and data prep. Cost assumes whichever plan is cheaper at this team size: $99 per seat or the flat $5,000 instance.

Deployment & security

Your instance. Your data. Nobody else's.

OpenData's public platform is public by design. A private instance exists for everything that can't live there: claims, contracts, financials, customer records. It's a dedicated deployment, isolated from the public platform, and your data is ingested into it and lives nowhere else.

The engine that touches your data is Apache 2.0 open source. Your team can read every line, and self-hosting the core is always on the table.

SSO and audit logs are on the roadmap but not shipped yet. We're also exploring federated queries, where your warehouse data is queried in place instead of ingested. If either one is a blocker for you, say so on the call.

Public catalog570+ datasets104 sourcesYour dataprivate to your instancejoined in your instanceOne query layerChat · REST · SQL · MCP

The questions most teams ask before kickoff.

If yours isn't here, the booking link below is the fastest way to get one.

Where does my data live?

In your own dedicated instance, isolated from the public platform and from other customers. That isolation is the point: the public platform is public by design, and the private instance exists for the data that can't live there. We run on dedicated hardware we own rather than a major cloud, and where the instance is deployed is flexible, so if you have a location or residency requirement, tell us and we'll work it out.

How does our data get in?

The same ingestion pipeline that runs the public catalog. CSVs, spreadsheets, Postgres tables, and warehouse exports become cataloged datasets with schemas, descriptions, and sample queries. Once ingested, your data is queryable exactly like the public datasets.

Which AI tools does it work with?

Claude, ChatGPT, and Cursor connect over MCP, and so does anything else that speaks the protocol. People ask questions in the tool they already use and get answers back with charts.

Do we need a data team?

No. Anyone can ask questions through chat and get a sourced answer. Your engineers can still reach for REST or the SQL sandbox when they want to work with the data directly, but that's an option, not a requirement.

Is the public catalog included?

Yes. All 570+ public datasets across 104 sources come standard, and the knowledge graph joins them against your own data so a single question can span both.

What about security and access control?

Access runs on instance-scoped API keys with per-seat revocation, so you can cut off a seat without touching anyone else. SSO and audit logs are on the roadmap but not shipped yet. If SOC 2 or HIPAA reviews are part of your process, bring the requirements to the call and we'll walk through exactly what's in place today.

What does it cost?

A fully managed private instance starts at $5,000/month with unlimited seats; we host it, run the ingestion, and keep it updated, and the final number scales with your data sources, deployment requirements, and support level. If you only need the public catalog, the team plan is $99/seat with a 5-seat minimum. Prefer to run it yourself? The Apache 2.0 core is free to self-host, without the platform layer. Book a call and we'll size it to what you're doing.

What's open source here?

The Apache 2.0 core (the `opendata/` Python library) is the ingestion engine, query layer, and REST API. You can read the code that touches your data and self-host the engine yourself. The platform features (search, knowledge graph, accounts, the hosted MCP App) live in the proprietary backend we host.

Bring a spreadsheet. Leave with answers.

A 20-minute walkthrough. Bring one real file or export and watch it become something your team can question in Claude on the call.

No credit card neededMSA/DPA availableYour data stays yours