How it worksPublish, subscribe, and keep shared data connected.CapabilitiesDefine how datasets collect, validate, and present data.AssuranceUnderstand integrity, access, and verifiable history.
All use casesRegulatory registersUtility ecosystemsSupply chainsAI and agentsOpen networks
DevelopersBlogAbout
Join the waitlist
ProductHow it worksCapabilitiesAssuranceUse casesAll use casesRegulatory registersUtility ecosystemsSupply chainsAI and agentsOpen networksMoreDevelopersBlogAboutJoin the waitlist
AI and agents
Join the waitlist
Use cases/AI and agents

Give AI what’s current.
And how it got there.

Bring published updates and record-level history into your AI workflows. Commons keeps datasets locally available and connected to their sources, with MCP capabilities that let assistants list, read, and search the records they need.

Join the waitlist →Explore MCP capabilities
AssistantsAgentsRetrieval and searchCompliance checksAnalytics copilotsCustomer support
What an assistant can draw on
Current recordThe latest published state, synced locally
Now
Earlier versionsWhat it said before, and what changed
History
Version usedThe input behind an earlier answer
Reference
Dataset toolsList, read, and search through MCP
commons mcp

The latest value isn’t the story.
Context is.

Current data answers what the source says now. Record-level history adds what it said before, what changed, and which version your workflow used.

Commons keeps all three locally available and connected to the source, so an assistant can reason with them rather than with a one-time export.

NowCurrent inputsFollow published changes instead of a one-time copy. Received changes feed your refresh and review steps; a new version doesn’t update an index, a cached prompt, or a pending decision on its own.
BeforeHistory to reason withCompare earlier and later versions. Give a workflow evidence of how the information evolved.
AccessTools to use the dataAttach MCP capabilities so an assistant can list, read, and search a dataset. Add custom tools for more specific operations.

Every participant.
A reason to use Commons.

Whoever publishes the data, builds on it, or builds tools for it, Commons gives AI the same connected, versioned source.

Data providers
Publish once, reach every partner

Publish registers, documents, and reference data once. Authorized AI workflows can receive new versions and use them to refresh their inputs.

AgenciesRegistersDocument owners
AI builders
Ground answers in current data

Subscribe to the datasets your assistants and agents need, and record the version behind every answer.

Assistant teamsAgent buildersAnalytics teams
Tool and service providers
Build tools every assistant can use

Prepare and enrich datasets, package MCP tools around them, and publish them for others to attach.

Capability authorsData platformsDomain experts

Capture the data version
with the decision.

When an assistant answers or a workflow recommends, it can store the record and dataset version it used as part of that decision. The record of what was decided carries the record of what was known.

Revisit the decision later and you can open that exact version and see what the record stated at the time, then set it beside what the record says now.

That turns history into something teams can act on: confirm a decision still holds, explain it with confidence, or spot the ones worth looking at again.

The Q3 report captured v41 when it was signed off. Reopening it shows what record/118 stated then, alongside the v42 change that followed.

Make the dataset
usable through MCP.

The built-in MCP capability lets a compatible assistant list, read, and search a dataset through commons mcp. MCP is the Model Context Protocol: a standard interface between AI applications and the tools they use.

Attach it to a dataset and any MCP-aware assistant can work with the records in your environment, within the permissions you set.

listreadsearch

Build tools around the questions your AI needs to ask.

A dataset can attach a signed MCP capability release, or bundle MCP tools with other dataset behavior. Teams create tools suited to their records and workflows.

A custom capability or application could expose retained versions to answer the questions on the right. Those are examples of what you can build; the built-in tools are list, read, and search.

Keep each tool focused on the records and versions its task needs. A rich history doesn’t mean putting the whole archive in every prompt.

Example questions a custom tool could answer
“What changed since our last review?”Custom toolCompares the version recorded at review time with the latest received version.
“How has this record evolved?”Custom toolReturns the retained versions of one record, with their publication times.
“Find licenses withdrawn this quarter.”Application logicFilters a search result by status change and period.
“Show me this record.”Built inList, read, and search through commons mcp.

Prepare the data once.
Use it everywhere.

Authorized assistants, agents, and applications can reuse the same local dataset. Publish data once it’s cleaned, combined, or enriched, rather than leaving every AI workflow to repeat the preparation.

Commons stores and synchronizes bytes. Ingestors, materializers, and applications interpret the formats; storing a format doesn’t make it searchable or understandable to a model.

The tools run next to the data, inside the App that holds your local copy. Nothing leaves that environment unless your application chooses to send it: it decides which results go to a model, and which stay local.

What Commons gives AI, and what stays with you.

Commons keeps the data current, versioned, and reachable. Judgment, reasoning, and action stay with your application.

FreshnessCurrent means current to the last sync.

Source publication, ingestion, sync, and application refresh all matter. A local copy can be available while behind the source.

HistoryHistory starts when capture starts.

Captured, retained versions only. A publication or receipt time isn’t the date a fact became effective; keep the source’s time fields.

ToolsTools expose what you configure.

MCP exposes the configured operations, not unrestricted access. Your application decides what reaches a model.

EvidenceEvidence informs; it doesn’t decide.

Versioned inputs make comparison possible. They don’t guarantee correct reasoning, identical outputs, or permission to act.

Give AI the current record and its history.

Local, versioned data with tools an assistant can use, and the version behind every answer.

Join the waitlist→Review record history→See all use cases→