SEMANTIC MEMORY FOR AI AGENTS

Build knowledge graphs
your agents can trust

Validated property graphs with vector embeddings. Give your agents structured semantic context instead of raw document chunks.

Deploy a Graph Read the Docs

The problem with vector-only RAG

Retrieved chunks have no guaranteed relationships to each other
No schema means agents can write contradictory information
Cannot traverse dependencies or hierarchies
Context quality degrades as the knowledge base grows

Graph + Vector = Structured Memory

Semantic search combined with relationship traversal
Schema validation prevents invalid data at write time
Navigate connections between entities for richer context
Hybrid queries: "Find similar modules that import auth-service"

How it works

Every graph instance runs PostgreSQL with Apache AGE and pgvector. You or your agents define schemas, then write validated entities.

Define your ontology

Create DTDL models that define what entities exist and how they relate. You or your agents can define schemas, and data writes are validated against them.

module.dtdl.jsonJSON

{
  "@id": "dtmi:code:Module;1",
  "@type": "Interface",
  "contents": [
    {
      "@type": "Property",
      "name": "path",
      "schema": "string"
    },
    {
      "@type": "Property",
      "name": "embedding",
      "schema": { "@type": "Array", "elementSchema": "double" }
    },
    {
      "@type": "Relationship",
      "name": "imports",
      "target": "dtmi:code:Module;1"
    }
  ]
}

Ingest via MCP, SDK or REST API

Use the SDK or REST API directly or connect AI agents via the Model Context Protocol. Every write is validated against your models.

claude_desktop_config.jsonJSON

{
  "mcpServers": {
    "konnektr-memory": {
      "httpUrl": "https://mcp.graph.konnektr.io/mcp?resource_id={resource_id}"
    }
  }
}

Query with Cypher + vectors

Combine graph traversal with semantic similarity in a single query. Navigate relationships and rank by embedding distance.

query.cypherCypher

MATCH (file:Twin)-[:imports]->(dep:Twin)
WHERE file.`$metadata`.`$model` = 'dtmi:code:Module;1'
  AND dep.name = 'auth-service'
RETURN file.path, file.lastModified
ORDER BY l2_distance(file.embedding, $query_vector)
LIMIT 10

MODEL CONTEXT PROTOCOL

Connect to Claude, Cursor, or any MCP client

The MCP server exposes your graph as tools. Agents can create entities, build relationships, and query the graph through natural conversation.

search_digital_twins

Semantic search across all entities

query_digital_twins

Execute Cypher for graph traversal

create_or_replace_digital_twin

Write validated entities to the graph

Agent building codebase knowledge

I'll search for modules related to authentication...

search_digital_twins

{ "search_text": "authentication middleware" }

Found 3 modules: auth-middleware.ts, jwt-validator.ts, session-handler.ts

Let me see what depends on auth-middleware...

query_digital_twins

{ "query": "MATCH (m:Twin)-[:imports]->(auth:Twin {name: 'auth-middleware'}) RETURN m.path" }

12 modules import auth-middleware: api/routes/users.ts, api/routes/admin.ts, ...

I'll add a note about this dependency pattern...

create_or_replace_digital_twin

{ "twin_id": "note-auth-pattern", "model_id": "dtmi:code:ArchitecturalNote;1", "properties": { "content": "auth-middleware is a critical dependency..." } }

Created architectural note linked to auth-middleware

Built for structured domains

When your agents need to understand relationships, not just content.

Codebase Knowledge

Map modules, dependencies, and architectural decisions. Ask "what breaks if we change this?" and get graph-backed answers.

"Which services depend on the payment module?"

Real-Time Context

Allow agents to read real-time device states. Sync sensor networks to the graph for live situational awareness.

"Get current temperature of Pump-A"

Compliance & Audit

Track decision lineage and regulatory requirements. Every node has provenance, every relationship is typed.

"Show the regulation that requires this control"

Built on proven foundations

PostgreSQL

ACID compliant, battle-tested. Your data lives in the most reliable database in the world.

Apache AGE

Graph extension for Postgres. Native Cypher queries without leaving your relational database.

pgvector

Vector similarity search alongside graph nodes. True hybrid queries in a single database.

Start free, scale as you grow

The Developer tier is free forever with 500 entities and full API access. Upgrade when you need more capacity or enterprise features.

Deploy Free Graph

Build knowledge graphsyour agents can trust