Agents share tools. They share messages. They don't share understanding.
ContextMesh is a decentralized semantic context graph that gives AI agents from any provider a shared world model. Zero-copy. No central authority. The missing infrastructure for multi-agent interoperability.
Get Started FreeMESH
The agent stack has two layers. It needs three.
MCP standardized tool access. A2A standardized messaging. ContextMesh standardizes shared state.
Tools (MCP)
Agents connect to databases, APIs, and external services. Solved.
Messages (A2A)
Agents send structured messages to each other. Solved.
Understanding (ContextMesh)
Agents read from and write to a persistent, decentralized semantic graph. A shared world model with provenance, identity, and zero-copy access.
Built on first principles
Not another orchestration framework. Infrastructure for shared cognition.
Zero-Copy Access
Agents query the semantic graph directly instead of copying context into their own windows. Less duplication, less drift, lower cost.
Decentralized Ownership
No central authority controls the world model. Agents and their owners maintain provenance via DIDs and verifiable credentials.
Semantic, Not Syntactic
The context graph understands meaning, not just structure. Agents from different providers resolve to the same concepts without translation layers.
One SDK. Any agent.
import { ContextMesh } from '@contextmesh/sdk'
const mesh = new ContextMesh({ namespace: 'project-alpha' })
// Agent A writes context with provenance
await mesh.write({
subject: 'user:jane',
predicate: 'prefers',
object: 'dark-mode',
agent: 'did:agent:claude-assistant'
})
// Agent B reads it, zero-copy, from anywhere
const prefs = await mesh.query({
subject: 'user:jane',
predicate: 'prefers'
})
The future of agents isn't smarter models. It's shared understanding.
When agents can read from the same world model, they stop duplicating work, contradicting each other, and losing context. That's ContextMesh.
Start Building