Skip to content
DataTug.ai
GitHub

Give AI agents access to data.
Not unlimited power.

DataTug Perimeter is how your agents will get scoped access to your databases, see sensitive fields only in masked form, and change data only after you approve. Start today with the DataTug skills and the access policies in the CLI.

Evolving fast

Set perimeter for your agents

Paste this into your AI agent:

Set up DataTug using https://datatug.ai/install-skills.md Ask before changing shell settings or anything outside this setup.

Read the guide your agent gets The guide explains what your agent can set up and when it needs your approval. Setup needs no database passwords.

  • Free CLI · personal and business use
  • CLI v0.152.1

DataTug is a workspace where you and AI investigate connected data: in the web app, in the terminal, or inside your own coding agent with the DataTug AI skills. Answers across sources work today in the CLI. The evidence trail, shared meaning and memory are what we are building next.

The live demo in DataTug.app is in development. For now this opens a prototype of the investigation.

See how DataTug answers

Your AI agent has your database password. Now what?

Three problems every team with AI agents runs into.

DataTug answers each with one part of the same platform. They are stronger together.

  1. A painted harbour breakwater: the tug passes through the one open gate while the other channels are roped off with buoys.

    Uncontrolled access

    An agent with your database password can do anything.

    Perimeter is being built to put a boundary between the agent and the data: access scoped to what the task needs, sensitive fields masked, and changes only after you approve them.

    DataTug PerimeterEvolving fast

    How Perimeter works →

  2. A navigator studies a sea chart whose islands are joined by fine blue threads.

    No understanding

    Agents do not know what your data means.

    MeaningGraph is being built to record what tables and columns mean and how they connect. The DataTug skills already teach your agent the commands that read your data.

    MeaningGraph integrationPlannedSkills plugin for coding agentsAvailable

    How meaning works →

  3. At dusk the tug's searchlight picks out one stack of containers among many on the quay.

    Lost investigations

    Agents struggle to follow an issue across systems.

    Detective is being built to keep the case: the records followed, the queries run and what was found, so you and the agent work from the same evidence. The skills give the agent the commands.

    Detective CasesIn developmentSkills plugin for coding agentsAvailable

    How investigations work →

Where they meet

  • Perimeter decides what every skill and every investigation step may read.
  • MeaningGraph tells Detective how records in different systems relate.
  • The skills are how your agent reaches all three.

DataTug PerimeterEvolving fast

Who can reach your data, and why?

Before more people and AI agents query your data, you need to know who can reach it. Perimeter shows the whole picture, so access is decided and audited, not guessed.

Perimeter is evolving fast. Full capabilities are coming soon.

DataTug Perimeter — just enough access

  • See who can reach which data, and why: direct grants, groups and native database roles.
  • Grant temporary access, with approval.
  • Audit who did what, and when.
All DataTug films

One platform, three ways in.

Explore and chat in the browser, work from the terminal, or connect an AI agent. The status of each capability is shown here.

The whole platform: datatug.io

Start with the right fit

Explore for free. See cloud plans as your needs grow.

Free €0/ monthPro €19/ monthBusiness €199 base / month + usageEnterprise Custom
Start free with the local CLI. Compare cloud plans for individuals and teams. Bring your own AI provider or choose managed AI where available.
  1. Chinook · Invoice.BillingCountry"Ireland"
  2. MeaningGeo.CountryPlanned
  3. ISO 3166 codesIE · IRL
  4. World BankPopulation 5.48 million
  5. Answer8.3 per million

Sales per million people

Real demo run
  1. Ireland8.3
  2. Czech Republic8.3
  3. Finland7.4
  4. Canada7.3
  5. Portugal7.1
  6. USA1.5

The USA buys the most in total, yet ranks number 17 per person.

Ask a follow-up

Result: sales per million people

CountrySales, USDPopulation (2025)USD per million people
Ireland45.625,484,3678.32
Czech Republic90.2410,886,8788.29
Finland41.625,646,4367.37
Canada303.9641,651,6537.30
Sort and filter it

Investigation

  1. Works out what the question needsPlanned
  2. BillingCountry means a countryPlanned
  3. Finds population data elsewherePlanned
  4. 24 of 24 country names mappedReal run
  5. Three sources joined, metric computedReal run
Inspect any step

Swipe for the other views

  • 24countries
  • 3sources joined
  • 0AI tokens
  • 0.3 srun time

From the real demo run: one saved query, no model.

How DataTug assembles an answer. Open any step.

Not a chain of thought: a record of actions, decisions and evidence. First, exactly what the real run did. Then the investigation DataTug is being built to show.

What the real run did

Real demo run

The DataTug CLI executed one saved TugQL query. It joins three sources and calls no model.

Command

$ datatug query run sales/chinook-sales-per-capita

What it printed

query download: 412 rows chinook
query download: 24 rows geo
query download: 216 rows geo
query process: 24 rows
Chinook invoices read
412
country names in the alias table
24
equal the ISO name
21
matched by hand
3
World Bank population records in the snapshot
216
result rows, one per country
24
AI tokens
0

Cross-source TugQL joins in the CLIAvailable

Reproduce it: datatug/datatug-demo-projects, commit 0e5b98f, DataTug CLI 0.51.0.

What the product will show

In development

The same question as an investigation in DataTug.app. Every step is marked: it either happened in the real run, or it is behaviour we are building, with its status.

Investigation: music sales relative to populationPrototype

Which countries buy the most music relative to their population?

Steps: 7. In the real run: 4. Target behaviour: 3.

  1. Works out what the question needs.Music sales grouped by country: the Invoice table, its Total and its BillingCountry. In the real run a person chose these when writing the query. In the product, Jev, a decision model by TypeSafe AI, will propose them and the schema will check them.Target behaviourDecisions by Jev, a decision model by TypeSafe AIPlanned
    Table
    Invoice
    Rows
    412
    Measure
    Invoice.Total
    Grouped by
    Invoice.BillingCountry
    How it is decided
    Proposed by Jev, a decision model by TypeSafe AI, then checked against the schema
    Known from schemaAI suggested
  2. Recognises BillingCountry as a country.MeaningGraph will say this field means Geo.Country, so nobody has to guess from the column name.Target behaviourMeaningGraph integrationPlanned
    Field
    Invoice.BillingCountry
    Meaning
    Geo.Country
    Distinct values
    24
    How it is decided
    Looked up in MeaningGraph
  3. Finds population data outside Chinook.Chinook has no population. Automatic discovery by DataTug through the existing OVDB Directory is planned. In the real run the World Bank snapshot was already in the project.Target behaviourAutomatic Directory discovery in DataTugPlanned
    Source
    World Bank
    Dataset
    SP.POP.TOTL · 2025
    Rows
    216
    How it is decided
    Found through OVDB Directory
    Declared
  4. The sources spell countries differently.Chinook stores names such as "USA". The population data is keyed by ISO country. They cannot be joined as they are.In the real run
    Chinook says
    "USA"
    ISO country key
    us
    ISO name
    "United States"
    How it is decided
    A difference in the data the run read. The run itself compared nothing: the alias table bridges the names
    No modelNeeds reconcilingVerifiedInspect this value
  5. A table of country names bridges them.All 24 Chinook names map to an ISO country. Equal to the ISO name: 21. Matched by hand: 3.In the real run
    Reference data
    country_aliases
    Resolved
    24 / 24
    Equal to the ISO name
    21
    Matched by hand
    Czech Republic, Netherlands, USA
    How it is decided
    Looked up in reference data kept as files in the project
    No modelVerifiedHuman confirmedInspect this value
  6. Joins the three sources.Invoices to aliases to population, in one TugQL query. No model is involved.In the real runCross-source TugQL joins in the CLIAvailable
    Join path
    Invoice.BillingCountry → country_aliases.alias → population_wb.country
    Source
    Chinook + World Bank
    How it is decided
    One deterministic TugQL query
    From the saved query (TugQL)
    from:
      database: chinook
      name: Invoice
      alias: i
      joins:
        - from:
            database: geo
            name: country_aliases
            alias: a
            joins:
              - from: {database: geo, name: population_wb, alias: p}
                on: [{left: {field: country, source: a}, op: '==', right: {field: country, source: p}}]
          on: [{left: {field: BillingCountry, source: i}, op: '==', right: {field: alias, source: a}}]
    No modelVerified
  7. Computes sales per million people.One row per country, ready to sort, filter and chart. Zero AI tokens.In the real run
    Formula
    sum(Total) / population * 1000000
    Rows
    24
    AI tokens used
    0
    How it is decided
    Arithmetic inside the query, no model
    No modelVerifiedSee the result

Where knowledge comes from

  • Known from schema
  • Declared
  • Human confirmed
  • Verified
  • Observed
  • Inferred from data
  • AI suggested
  • Hypothesis

The trace view is a prototype of how DataTug.app will present an investigation. It is not yet in the product.

The answer is data you can keep working with.

Sort it, filter it, resize it, select a bar to find the row. The grid and the chart are two views of the same result.

Cross-source TugQL joins in the CLIAvailableCross-source TugQL joins in the browserIn development

Result: sales per million peopleReal demo run
24 of 24 countriesTry sorting by Sales: the USA buys the most, yet ranks number 17 per person.
Skip the table
Sales and population by country
CountrySales, USDPopulation (2025)Sales per million people
Ireland45.625,484,3678.32
Czech Republic90.2410,886,8788.29
Finland41.625,646,4367.37
Canada303.9641,651,6537.30
Portugal77.2410,804,8717.15
Norway39.625,610,8707.06
Denmark37.626,009,1696.26
Hungary45.629,514,2514.79
Austria42.629,208,1634.63
Sweden38.6210,596,6203.64
Belgium37.6211,941,7813.15
France195.1068,720,3372.84
Chile46.6219,859,9212.35
Netherlands40.6218,087,6332.25
Germany156.4883,491,2491.87
United Kingdom112.8669,487,0001.62
USA523.06341,784,8571.53
Australia37.6227,614,4111.36
Poland37.6236,435,8611.03
Brazil190.10212,812,4050.89
Argentina37.6245,851,3780.82
Spain37.6249,355,1430.76
Italy37.6258,915,6560.64
India75.261,463,865,5250.05

Loading the interactive grid…

End of the table

Sales per million people

USD per million people

  1. Ireland8.3
  2. Czech Republic8.3
  3. Finland7.4
  4. Canada7.3
  5. Portugal7.1
  6. Norway7.1
  7. Denmark6.3
  8. Hungary4.8
  9. Austria4.6
  10. Sweden3.6

Showing the top 10 of 24 rows in the current order

From the real demo run: Chinook sales joined to World Bank population (SP.POP.TOTL, 2025). A recorded result, not a live query.

A follow-up question Prototype

What can you say about this data?

Three things stand out. Each one is computed from the rows it points at.

  • The USA has the largest total ($523.06) but ranks number 17 of 24 per person: 1.53 per million.Rows: 1
  • Ireland and Czech Republic lead per person at 8.32 and 8.29, almost tied.Rows: 2
  • 9 of the top ten have fewer than 11 million people. Canada is the exception, with 41.7 million.Rows: 10

Hover or select an observation to highlight its rows in the grid and the chart. Observations grounded in your dataIn development

The grid uses the DataTug AG Grid theme from the site kit. The CLI runs this cross-source query today; running the same query in the browser is in development.

Three ways to work with DataTug AI can help in each.

Use the browser workspace, terminal UI and CLI, or integrate DataTug into an AI agent. AI chat is part of the first two; statuses keep each integration clear.

  1. In your browser

    Explore projects, inspect schemas and chat over your data in DataTug.app. Use a model key you provide, and carry the same investigation into the terminal.Open DataTug.app

    DataTug.app · AI chatIllustrative

    Session: chinook-countries

    Which countries buy the most music relative to their population?

    Joined Chinook sales with World Bank population.

    • Ireland
    • Czech Republic
    • Finland

    datatug chatDataTug.appContinue this investigation in the terminal

    Start withDataTug.app

    • DataTug.app, the web appAvailable
    • Browser chat with your own API keyAvailable
    • Managed AI in the web appAvailable
    • Terminal-to-web session hand-offAvailable
  2. In your terminal

    Explore with the keyboard-driven terminal app, chat with AI, or run repeatable commands in scripts and pipelines.Read the terminal app guide

    Terminal UIIllustrative

    › datatug chat

    › Which countries buy the most music per person?

    Joined Chinook sales with World Bank population.

    ✓ 24 countries · 3 sources

    ›

    Start withdatatug chat

    • Terminal app (TUI)Available
    • Chat in the CLI and TUIAvailable
    • Managed AI in the CLIAvailable
  3. Integrated into your AI agent

    Claude Code skills support project setup, discovery, schema scans, and validation. Connect MCP tools for schemas and saved queries. The MCP Apps explorer developer preview adds table and record views, sorting, paging, and parameter reruns for saved SQLite queries. Tested with the contributor reference host; follow the MCP setup guide for supported scope.See the skills and how to install them Commands an agent can run

    Your agentIllustrative

    › /plugin install datatug@datatug

    ✓ Skills for the datatug CLI are installed.

    › Which countries buy the most music relative to their population?

    Running datatug query run sales/chinook-sales-per-capita

    Ireland           8.32
    Czech Republic    8.29
    Finland           7.37

    ✓ Done. The result is back in the conversation.

    ›

    Start with/plugin install datatug@datatug

    • Skills plugin for coding agentsAvailable
    • MCP serverAvailable
    • MCP Apps explorer (developer preview)Preview

Give AI meaning, not just schema.

A schema says BillingCountry is text. A meaning says it is a country, which other systems know as IE or IRL, and how sure we are of each fact. That is the job of MeaningGraph. Connecting it to DataTug is planned.

MeaningGraph integrationPlanned
  • Knowledge has a source

    Each fact says whether it came from the schema, a person, the data, or a model.

  • It goes both ways

    DataTug will read MeaningGraph before it asks a model, and write back what you confirm.

  • Meaning crosses systems

    The same country can be a name, an alias, an ISO code. Meaning is what connects them.

One value, followed across three sourcesReal values, concept view
Choose a country as Chinook stores it.
Chinook value"Ireland"Invoice.BillingCountry
MeaningConceptGeo.Countrywhat this value is
Alias row"Ireland" → iecountry_aliases
ISO 3166 codesIE · IRLcountries
World Bankiepopulation_wb · Population 5.48 million

Evidence

Alias row: "Ireland" → ie

  • VerifiedThe alias "Ireland" equals the ISO country name and maps to the key ie. Aliases like this: 21 of 24.

Select a node to see its evidence.

The values, the alias table and the population figures are from the real run. The Geo.Country meaning node is a concept: MeaningGraph integration is planned.

It learns while you work, and checks before it believes.

Queries, joins and the way you navigate leave evidence. The idea: DataTug turns repeated evidence into knowledge you can confirm, then reuses it without asking a model again.

MeaningGraph learning loopExploring
A relationship DataTug noticedConcept

Invoice.BillingCountry → Geo.Country

ObservedHuman confirmed
  • Used as a country in every saved query that touches it
  • Values checked against the reference list. Matches: 24 of 24
  • Not yet confirmed by anyone

Confirmed by you. The next investigation looks this up instead of asking a model.

Infer once. Verify. Remember. Reuse.

  1. InferA pattern or a model proposes what something means.
  2. VerifyDataTug checks the proposal against the real data.
  3. RememberWhat holds is saved, with its evidence, in the project.
  4. ReuseThe next investigation starts from what is known.

Every investigation makes the next one smarter.

The design target for a question asked again and again:

  1. First timeA model helps plan. A person confirms the mapping.Model calls: several
  2. Second timeThe mapping is looked up. No model for that step.Model calls: fewer
  3. Known pathEvery step is known. This is what the real demo run already is.Model calls: none

Learning from your work is an idea we are testing. The loop is the design target, not shipped behaviour.

Not every step needs a large model.

DataTug is designed to reach for the lightest dependable mechanism first, and to say which one it used: what is already known, then a small decision model, then a large model, then you.

Decisions by Jev, a decision model by TypeSafe AIPlanned
  1. Deterministic knowledge

    Schema, confirmed meanings, reference data, saved queries. Free, instant, repeatable.

  2. A lightweight decision model

    Jev, a decision model by TypeSafe AI, ranks candidate tables, fields and join paths. The schema then checks the pick.

  3. Generative AI

    New questions, plans nobody has seen, explanations in plain language.

  4. A person

    You confirm what will be trusted from then on.

Which tables are relevant to this question?Concept

Which countries buy the most music relative to their population?

  1. InvoiceChosen
  2. InvoiceLine
  3. Customer
  4. Track
  5. Album

Ranked by Jev, a decision model by TypeSafe AI. The chosen table is then validated against the schema.

Humans and AI build understanding together.

Not all knowledge deserves the same trust. DataTug is designed to keep the difference visible on every claim, so you can tell what was declared from what was guessed.

Authoritative

Reused without asking.

  • Known from schemaDeclared by the database itself: types, keys, constraints.
  • DeclaredWritten down by a person or an API description.
  • Human confirmedSomeone on the team checked it and said yes.
  • VerifiedChecked deterministically against the actual data.

Corroborated

Reused, with the evidence in view.

  • ObservedSeen repeatedly in queries and navigation.
  • Inferred from dataPatterns in the values point to it.

Suggested

Shown, never trusted silently.

  • AI suggestedA model proposed it. It still needs checking.
  • HypothesisA guess worth testing, never reused silently.
orders.customer_ref → customers.idExample
  1. ObservedSeen in repeated queries
  2. VerifiedValues verified against data
  3. Human confirmedConfirmed by the team

Reuse then needs no model call.

Investigate, don't just generate SQL.

A query is one move in an investigation. DataTug is built around the whole thing: the question, the path you took, the evidence, and what you learned along the way.

A database client

  1. You
  2. SQL
  3. Database

An AI database client

  1. You or an agent
  2. Generated SQL
  3. Database

DataTug

  1. You + AI
  2. Investigation
  3. Meaning
  4. Connected data
  5. Evidence
  6. Reusable knowledge

Ways in

  • Natural language
  • SQL
  • TugQL
  • Navigation
  • Grids
  • Charts
  • CLI
  • TUI
  • Agents

They complement each other. Use whichever is quickest for the next step, whether or not you write SQL.

What is different

  • The agent sees meaning

    Not only table and column names, but what they stand for.

    Planned
  • It works across sources

    Join a database with an outside dataset in one query.

    Available
  • It remembers between sessions

    What you confirmed is there next time, with its evidence.

    Exploring
  • One session, three interfaces

    Web, terminal UI and CLI work on the same session.

    Available

Multi-tenant reconciliationIn development

One application, many databases. Do they still agree?

Many companies run one application as many databases: one per factory, region or tenant, each with production and test copies. DataTug recognises that they share one data model and reconciles a dataset across all of them in one pass, separating the differences you expect from the ones you don’t.

Five factories, production databases
DatasetFactory AFactory BFactory CFactory DFactory E
CountriesMust matchSameSameUnexpected differenceSameInconclusive
CurrenciesMust matchSameSameSameSameInconclusive
Local pricesMay differExpected differenceExpected differenceExpected differenceExpected differenceInconclusive
Factory C stands apart: its countries list gives France the currency FRF, while A, B and D say EUR. Factory E could not be read, so it is inconclusive, never “the same”.
  • Rules say what must match: reference data such as countries and currencies everywhere, while a local price may differ on purpose. AI can suggest a rule; it applies only once you confirm it.
  • See which instance is the outlier, then drill down to the record and the field. Compare against an instance you name as the authority, or see how the instances split, without assuming the majority is right.
  • Every database is read through DataTug Perimeter, with your own access. A database you cannot read is reported inconclusive, never “the same”.
  • Read-only: DataTug finds and explains differences and never changes your data. Save the rule as a Check that runs on a schedule.

Coming to the CLI

datatug reconcile \
  --query countries \
  --key code \
  --family factory \
  --kind production

Don't make AI guess what it can look up.

Chinook says "USA", the population data says "us", and ISO calls the same country USA and "United States". That is a fact to look up, not something for a model to work out again on every question. So keep it as a table: stored once, versioned, reviewed like code.

Reference data: country_aliasesReal demo data
aliascountryhow it was matched
Czech Republicczmatched by hand
Netherlandsnlmatched by hand
USAusmatched by hand
Irelandieequals the ISO name

The three names that needed a person, and one that did not. The mapping is first-class data, not something a model re-infers on every question.

The same row, as a file in GitReal file
# data/geo/country_aliases/$records/usa.json
{
  "alias": "USA",
  "country": "us",
  "source": "chinook-database Invoice.BillingCountry; manual match: Chinook spells it 'USA'; countries.names.en is 'United States'"
}

Four kinds of durable investigation data

  • Reference data

    Countries, currencies, internal ids and standards, status mappings, lookups.

  • Snapshots

    Point-in-time copies for incidents, test failures, migration rehearsals and regression fixtures.

  • Cached data

    Outside or expensive data kept so it is fetched once, not on every run.

  • Session data

    Selected records, suspicious transactions, intermediate joins and temporary mappings.

Where it lives

inGitDB

Available

Persist and version data

inGitDB is live today: human-readable files in Git that are branchable, reviewable in a pull request, and reproducible in CI. Managing project data this way from inside DataTug is planned.

inGitDB

OVDB

Available

Access and share

OVDB and its public Directory are live today, providing a common way to discover and describe very different database sources. Automatic source discovery inside DataTug through the Directory is planned.

OVDB

One intelligence, many interfaces.

The session belongs to DataTug, not to the interface. Start in the terminal, open the grid in the browser, and come back with the same context. Or the other way round.

Terminal-to-web session hand-offAvailable
TerminalIllustrative

$ datatug

Session chinook-countries ● synced

› Which countries buy the most music relative to their population?

✓ Joined Chinook with World Bank population.

o open grid in browser

DataTug.appIllustrative

Session: chinook-countries

Continued from your terminal

  • Ireland
  • Czech Republic
  • Finland

Return to the terminal anytime; it is the same session.

Scripts and embedded views

  • CLI

    One-shot commands for pipelines and scripts, with managed AI built in.

    Managed AI in the CLIAvailable
  • Embeds

    Put a live grid or chart on any page.

    <datatug-grid> and <datatug-chart> embedsAvailable

Your model, or ours.

DataTug AI is not tied to one provider.

  • Bring your own key

    Use your own API key with the provider you already trust, in the browser or the CLI.

    Browser chat with your own API keyAvailable
  • Managed AI

    DataTug-managed AI with no key to set up: in the CLI and the web app. Your prompts, the schema and the query text go to the cloud. Row values do not.

    Managed AI in the CLIAvailableManaged AI in the web appAvailable

When an answer needs more than text, pop a Bubble.

A Bubble is a small interactive surface, such as a grid, a chart or an inspector, that appears inside the conversation and composes with others into richer tools. UIBubbles is the open standard for them.

UIBubbles runtimePlanned
Chat with BubblesConcept

Show the top countries as a grid.

Here are the countries, sorted by sales per million people.

CountryPer million
Ireland8.3
Czech Republic8.3
Finland7.4
Canada7.3

Chart that.

Charted, and linked to the grid.

Why does Ireland map to IRL?

Here is the evidence for that mapping.

  • Mapped from "Ireland" through project reference data
  • Names checked: 24. Resolved: 24.
Pop a Bubble

This is a hand-built concept of the experience. The Bubble runtime is planned. About UIBubbles

Knowledge should not reset at every stage of development.

What one stage learns about the data should be there for the next. The aim of DataTug and MeaningGraph is to carry it from first idea to day-to-day support.

  1. Envision

    Explore the data and domain knowledge that already exist.

  2. Design

    Define entities, meanings, identifiers, relationships and mappings.

  3. Develop

    Developers and agents use what is known, and feed discoveries back.

    Incident possible
  4. Test

    Investigate failures, compare environments, keep the edge cases you find.

    Incident possible
  5. Operate

    Support and agents investigate real behaviour with everything learned so far.

    Incident possible

Incident is a state, not a sixth stage.

It can happen while you develop, in test, in staging or in production. An early incident is a gift: it shows a problem before customers do, while it is still cheap to fix.

Each one starts a focused investigation, and what it teaches flows back into the work ahead.

The best production incident is the one already caught in test.

Pressure reveals truth. Keep it.

Incidents are high-pressure moments when important truths about systems and data surface. Preserve those truths instead of losing them when the incident ends.

Incidentius integrationPlanned

  • DataTug

    Helps you investigate while it is happening.

  • MeaningGraph

    Preserves what was established: meanings, relationships and diagnostic knowledge.

  • Incidentius

    Holds the incident itself: the event, the context and the response.

The idea: discoveries keep their link back to the incident that taught them, and come back in future incidents, in development, in regression tests and in prevention.

Visit Incidentius.com

Control comes first.

Pointing AI at production data should be safe by design. This is exactly what holds today and what is still ahead, claim by claim.

  • Local execution

    Run DataTug on your own machine, next to your data: queries run locally. With managed AI, your prompts, the schema and the query text go to the cloud. Row values do not.

    Available
  • Access policies

    Scope what a role may read through policy-enforced saved queries, on SQLite and inGitDB sources today. With no policy set, access is unrestricted, and the terminal database viewer does not apply policies yet.

    Available
  • Read-only connections

    Connect to a source as read-only, so an investigation cannot change it.

    Planned
  • Approve an investigation step

    Approve an investigation step before it runs, instead of trusting a whole plan at once.

    Planned
  • Meaning-driven masking on the server

    Mask sensitive values on the server, driven by what a field means rather than by its name.

    Planned
  • Control over what reaches a model

    Decide what reaches an external model, down to field level.

    Planned
  • Audit trail merged with provenance

    An audit trail that is the same record as provenance: who or what did this, from which evidence.

    Planned

How the pieces fit.

You met them one at a time above. Each has one job, and none of them is required to start.

  • Investigate and learn

    DataTug

    The workspace where you and AI work through connected data.

    The workspace
  • Understand

    MeaningGraph

    What fields, values and relationships mean, with the evidence for each.

    meaninggraph.io
  • Access and share

    OVDB

    One way to reach and publish data across very different sources.

    openvaultdb.com
  • Persist and version data

    inGitDB

    Reference data and snapshots as readable files in Git.

    ingitdb.com
  • Interact and extend

    UIBubbles

    Grids, charts and inspectors that appear right inside an answer.

    uibubbles.org
  • Keep what incidents teach

    Incidentius

    Incident context and response, so the learning outlives the incident.

    incidentius.com

Who it is for today, and who should wait.

DataTug is early. Here is who it already serves well, and who should wait.

Best fit today

  • You investigate data that lives in more than one place.
  • You switch between terminal and browser and want the same session in both.
  • You want to bring your own model key, or use managed AI in the CLI or browser.
  • You like tools you can inspect, script and run yourself.
  • You would rather shape meaning and provenance with us than wait for them.

Not the best fit yet

  • You need AI that already knows what your fields mean. MeaningGraph integration is planned.
  • You need a step-by-step evidence trace in the app today. It is in development; this page shows the prototype.
  • You need a dashboard builder for many teams. DataTug investigates; it is not a BI suite.
  • You want one-click setup for every agent. The local MCP server needs a running DataTug daemon and a stdio client configuration.
  • You need server-side masking and approval gates for production data. Both are planned.

What works today, and what is still ahead.

This list tracks DataTug capability status. The availability of ecosystem projects is described separately.

Available

Shipped. You can use it today.

  • DataTug.app, the web app
  • Terminal app (TUI)
  • Query workspace in the web app
  • Related records in the web app
  • Cards
  • Dashboards
  • Column meanings
  • Chat in the CLI and TUI
  • Browser chat with your own API key
  • Managed AI in the CLI
  • Terminal-to-web session hand-off
  • <datatug-grid> and <datatug-chart> embeds
  • Skills plugin for coding agents
  • Skills plugin for DeepSeek Harness
  • MCP server
  • Cross-source TugQL joins in the CLI
  • Managed AI in the web app
  • Local execution
  • Access policies
Preview

Works from start to finish, but not everywhere yet.

  • Schema browsing in the web app
  • Log viewer
  • MCP Apps explorer (developer preview)
Evolving fast

Perimeter is evolving fast. Full capabilities are coming soon.

  • DataTug Perimeter
In development

Being built now. Do not plan on it yet.

  • HTTP queries
  • Public data lookup in DataTug
  • Incidents in DataTug projects
  • Cross-source TugQL joins in the browser
  • Inspectable investigation trace
  • Detective Cases
  • Reusable Checks
  • Context Bag
  • Observations grounded in your data
  • Multi-tenant reconciliation
  • Anonymous web demo
Planned

Designed and scheduled, not started.

  • AI assistant in the query builder
  • AI assistant in schema discovery
  • Decisions by Jev, a decision model by TypeSafe AI
  • MeaningGraph integration
  • Automatic Directory discovery in DataTug
  • inGitDB project data management in DataTug
  • UIBubbles runtime
  • Incidentius integration
  • Read-only connections
  • Approve an investigation step
  • Meaning-driven masking on the server
  • Control over what reaches a model
  • Audit trail merged with provenance
  • Agent changes you approve
  • Clone the demo project
Exploring

An idea we are testing. It may change or be dropped.

  • MeaningGraph learning loop

See it first. Install when you are convinced.

The live demo is in development. Today you can walk through the prototype on this page, or install the CLI and run the real query yourself.

Star on GitHubThe CLI is free for personal and business use within the Free plan limits. See pricing. The AI skills are open source (MIT). Skills on GitHub.

What happens next

  1. AskSoon: run a question in the live demo, with no sign-up. Today: the prototype on this page.In development
  2. InspectOpen any step of the investigation and see the result as a grid and a chart.In development
  3. Keep itKeeping an investigation, cloning the demo project and connecting your own data are planned.Planned
See how DataTug answersLive demo in development