Platform overview

Data engineering agents that work from your context

VaultSpeed is an agentic data modeling platform. Agents generate, convert and migrate in the methodology you choose, the context store keeps them right, and every change is predictable and reviewed.

Book a working session Explore the context store

Agents analyze and model. The generator writes the code from the graph the same way every run, wherever the pattern allows. Every change is reviewed in Git.

Inside and around the platform

Three components on one shared context

Everything on the left ships with VaultSpeed and runs in your cloud. Everything on the right is yours already. The context store speaks MCP, so both sides work from the same data model, lineage and logic.

Inside VaultSpeed

Git workspace

Skills, agents and every generated artifact, versioned

Your work is versioned and reviewable.

A persistent workspace in your Git: models, code, documents, analysis and lineage, every change tracked and reversible, built to be extended.

Platform skillsConnectivity, security, scheduling, skill creator
Your own skillsBuilt by your team for what is unique to you
Skill marketplaceTested and categorized before release

Agent metaharness

One session, several agents, one set of rules

People and agents work one task on one context.

Live memory for the task at hand. Runs specialist agents from several harnesses in one session, routes context between them and enforces your policies.

ComposeSpecialist agents, from several harnesses
ControlCost budgets and security policies
SandboxesEphemeral per run, live session sharing

Enterprise context store

Permanent organizational knowledge, as a graph

Change the data model once, everything downstream follows.

Structured, versioned and accumulated over time on open source graph infrastructure. Everything downstream is derived from it.

BusinessEntities, relationships, harmonization
↔
ImplementationPhysical objects, loading logic, lineage
↔
Data productProducts, contracts, semantic views
MCPthe shared language

What it connects to

Existing models and governance

What the organization already wrote down

Pulled into the store as the organization wrote them. Governance tools stay the master.

Modeling

ErwinER/StudioPowerDesignerSqlDBM

Governance

CollibraAlationAtlanUnity CatalogHorizon Catalog

Documentation

DocumentsSpreadsheetsLegacy ETL and dbt projects

Other IDEs and harnesses

The same context, in the tools your team already uses

Marketplace skills install in these harnesses too.

Claude CodeCodexCursor and other IDEsSnowflake CoCoDatabricks GenieAgents your team builds

Data platforms

Code, semantic views and metrics land where the data lives

Cloud

SnowflakeDatabricksMicrosoft FabricBigQueryRedshift

On premises

OracleSQL ServerPostgreSQL

Code as

SQLdbt modelsSpark SQL
What agents do

Agents do the analysis and propose the data model

Agents profile sources, read legacy code and interview transcripts, and propose the business model, the mappings and the vault structure. That work happens in the open, in Git and in shared sessions, where people and agents review each other's proposals and the context carries from one piece of work to the next.

Timeline: reason over the business model, profile the source and map it to the vault, generate the vault, design the data product

Reason over the business model

Agents identify the core business concepts and their keys, flag ambiguities, and write the conceptual model into the context store.
Reason over the business model

Profile the source and map it to the vault

Profiling turns a business key from an assumption into a decision that can be defended. The proposed plan is reviewed before anything is generated.
Profile the source and map it to the vault

Generate the code

DDL, transformation code and workflows, native to the target. No language model takes part in this step.
Generate the code

Design the data product

A semantic view over the vault, aligned to the Open Semantic Interchange, defined once and portable across the tools that consume it.
Design the data product
A VaultSpeed agent session: the layers inside the context store, then the lineage of one table traced downstream
What holds it together

The context store, a metadata graph

The context store is the data model made machine readable, with its lineage and logic, in one graph. Business context holds the definitions the organization agreed on, implementation context the physical objects, loading logic and lineage across bronze, silver and gold, and data product context the products, their contracts and semantics.

A versioned graph

Every version of the data model is a version of the graph. A change is a proposal, whether an agent or a person made it, and the diff shows which entities, objects and products change. A locked version is what the generator builds from.

Served over MCP

The modeler builds the graph, and engineers query it from Claude Code, Cursor, Snowflake CoCo, Databricks Genie or any agent over MCP. Whoever asks reads the same governed graph and gets the same answer. Governance tools stay the master; the store reads from them and never writes back.

Works from the store

VaultSpeed agentsClaude CodeCodexCursorSnowflake CoCoDatabricks GenieYour own agents

How modelers build and hand over the graph

From data model to code

The generator writes the code the same way every run

Most data platforms get built twice, once as a data model and once as the code that moves the data. VaultSpeed generates the code from the data model: DDL, transformation code as SQL or dbt models, and the workflows that load it, native to Snowflake, Databricks, Fabric, BigQuery or Redshift. Rule based generators or decision models such as JEV produce the same code from the same data model on every run, with no language model in the code path. Where no rule set exists, an LLM writes the code, grounded in the context store and reviewed on every run.

Rule based generators

The Data Vault Generator turns the data model into DDL, transformation code and workflows through rules and templates. No LLM in the code path, so the same data model gives the same code on every platform.
Rule based generators

Decision models

Decision models such as JEV encode the pattern choices explicitly. Same input, same decision, same code, every run.
Decision models

Other methods run as skills

Pattern libraries are embedded in skills. Install them from the marketplace or build your own skills and plugins. An LLM generates from the skill and the context store, and the output goes through Git review like any other code.
Other methods run as skills

What stays yours

Your macros

The output is plain SQL or dbt models that sit in your project beside the macros and conventions your team already maintains.

Your CI

The generated code goes into your CI/CD process and runs on your platform. VaultSpeed sits outside the runtime, so nothing is needed to keep it executing.

Your Git

DDL, transformation code, workflows, tests, lineage and documentation land in your repository as one reviewable change.

Bronze, silver, gold

We build the whole medallion

Deterministic where the pattern allows, agentic where it does not. Rule based generators or decision models such as JEV produce the same code from the same data model on every run. Where no rule set exists, an LLM writes the code, grounded in the context store and reviewed on every run.

Land every source

Source objects and staging, generated by rule from the metadata harvested from your systems.
Land every source

Integrate once

Hubs, links and satellites on the business keys, generated by rule. Where keys do not line up across sources, agents propose and reviewers decide.
Integrate once

Deliver the products

Star schemas, marts and semantic views, designed with agents from the data product design and generated from the same graph.
Deliver the products
The medallion VaultSpeed builds: bronze landed from every source, silver integrated once in a Data Vault, gold shaped into data products, and a semantic layer on top

Review before release

Nothing ships without review

Every agent proposal is reviewed and versioned before it ships. An agent works in its own branch, commits the change and opens a pull request. The diff shows the data model change, the regenerated code and the lineage it touches. Your engineers and reviewing agents approve or send it back, and only merged changes reach the store.

What a reviewer can check

Changes arrive in Git

An agent proposes a change as a versioned commit. Engineers and reviewing agents inspect the diff through the pull request process your team already uses.
Changes arrive in Git

Sessions run in sandboxes

Every agent session runs in an ephemeral sandbox on Kubernetes, on your infrastructure in a hybrid setup, against your LLM gateway and approved LLMs.
Sessions run in sandboxes

Spend is capped

Token spend is capped at the platform level. Operators manage connections, permissions, schedules and the audit log of everything agents do.
Spend is capped

Language models

Your platform's LLMs, your token budget

Every agent reaches its LLM through the gateway. Point the gateway at the inference your organization already pays for and governs, and the tokens, the data boundary and the audit trail stay where your data platform already is.

LLM gatewayone endpoint for every agent, pointed at one of three places
SnowflakeCortex REST API

Inference inside your Snowflake account

The gateway calls Cortex in the account and region you already run. Consumption lands on your Snowflake bill, under the roles and budgets you already govern.

  • No data or prompt leaves the Snowflake boundary
  • LLMs Snowflake has approved for your account
  • Spend on the contract you already have
DatabricksModel Serving

Inference inside your Databricks workspace

The gateway calls Model Serving endpoints in your workspace, governed by Unity Catalog. Genie works from the same semantic views the generator produces.

  • Inference stays inside the workspace
  • Unity Catalog permissions apply
  • Spend on your Databricks commit
Amazon Bedrockplatform neutral

When you run more than one platform

When your data is not on one platform, the gateway points at Bedrock. Two ways to run it.

Your BedrockYour AWS account, your region, your IAM policies and logging
Our BedrockProvisioned by VaultSpeed in your subscription's region, ready from the first session
  • Prompts and responses are never used to train models

Which LLM you use is a configuration, not an architecture decision. In every case the LLM works from metadata in the context store; your data stays where it lives. Details on how the platform is secured are on the trust center.

Skills marketplace

Skills are what agents run

A skill packages one task: a conversion, a generator, a data product builder. An installed skill lands in your Git workspace next to the skills your own team wrote, so you can read what it does before you run it.

Buying through the Azure and Snowflake marketplaces

Platform skills

Connectivity, security, generation and scheduling, built and maintained by VaultSpeed as part of the platform.
Platform skills

Local skills

Written by your team for what is unique to your organization. They stay in your workspace and remain yours.
Local skills

Marketplace skills

From VaultSpeed, partners and the community, tested and categorized before publication, with the verification status visible before you install.
Marketplace skills

Customer story · Retail

Thanks to VaultSpeed, we industrialized our data layer without compromising agility or compliance. Our teams build and deploy data products faster, smarter, and at scale.
Colruyt GroupColruyt Group: warehouse rebuild and modernization at very high volume, data products on SnowflakeRead the story

Talk to us

Start with your data model, not a demo

Bring a source and the methodology you already use, and we walk through how each part of the platform handles it on your own metadata.