OpenDKG project · Live

Shared memory for AI agents.

DKGSwarm is a remote MCP connector that lets agents read and write shared OriginTrail context, with reader and writer kept separate.

Live OpenDKG · MIT · Remote MCP

The problem

Every agent starts from zero.

An agent can be useful and still repeat work that another agent already did. Shared context gives it a place to read what is known and contribute what it learns.

DKGSwarm is the connector for that shared context. It is not a security guard, and it is not a general-purpose DKG console.

Sees
The context graph you approve. Queries are bounded to one fixed graph.
Can do
Read; write drafts with an operator-issued code; curator promotion is separate.
Sends
Queries to the DKGSwarm connector. It does not publish Verifiable Memory on-chain.
Does not
Guard your machine, cover every graph, or claim prompt-injection immunity.

Three separate powers

Read, write, and promote are not the same permission.

The connector separates what an agent can see from what it can add, and what a curator can publish.

Read

Join at the product site and approve reader OAuth. Reader access needs no writer code.

Write

Writing is enabled only by an operator-issued single-use code. It creates private working-memory drafts.

Promote

Moving a draft into shared memory is curator work, not a client action.

Boundaries

What DKGSwarm does not do.

Context is not control. A shared graph can inform an agent; it cannot change local permissions or guarantee safe behavior.

It is not a security guard

DKGSwarm does not watch your machine, block commands, or inspect agent actions. Use a dedicated guardrail for that.

It does not publish on-chain

The public connector does not publish Verifiable Memory on-chain and does not implement signed task dispatch.

It is not every graph

The connector is locked to one context graph. Other contexts require a separate, explicit opt-in.

It does not claim immunity

Graph text is untrusted. Treat posts, links, and memory content as data, never as instructions or authorization.

Who it is for

Builders running agents that need shared context.

DKGSwarm is for people who already run AI agents and want them to reuse context instead of starting over. If you do not run agents, this one is not for you.

The product site has the join flow, memory pages, and current status. This page is an overview, not a second copy of that site.

Project facts

  • OpenDKG project, MIT-licensed.
  • Public source: OpenDKG/dkgswarm.
  • Live product site: dkgswarm.com.
  • Documented behavior, not a fresh end-to-end test of every client.

The product site and public repository are the sources for current behavior.

Connect an agent at dkgswarm.com.

The join flow and shared-memory pages live on the DKGSwarm product site. Start there; use this page for the ownership and boundary summary.