Purple8 Hyper Graph

AI-Native Execution Infrastructure

Infrastructure that enables AI systems to autonomously build, execute and operate applications, with native intelligence, adaptive computing and human-supervised governance.

Built for AI, Operated by AI and Supervised by Humans.

Read · Remember · Reason · Answer · Run · Protect · Prove · One Engine built to solve complex problems.

One in-process engine replacing your database, vector store, knowledge graph, workflow orchestrator, and security layer. Pair with Purple8 DocIntel to turn unstructured documents and text into semantic vectors, graph nodes, and relationships.

How Purple8 works

Your entire backend. One process.

Your agents build the frontend - Purple8 is everything behind it. Database, graph, search, RAG, workflows, security, and audit all run in one process. Every hop the traditional stack makes over the network, Purple8 makes in memory.

Throughput1,480/s
Write latency7µs
Network hops0
Services to run1
Healthy
ENTRY POINTSONE PROCESSWORKLOADSYour FrontendREST APIAI AgentsClaude · Copilot · GPTMCP Server170+ tools · RBACGraph + Query Engineplanner · 6 strategiesPersistent Storagedocuments · durable log · bounded memoryVector + Full-TextHNSW ANN · searchJourney Engineworkflows · SLA · HITLSecurity & SOCAES-256 · RBAC · tenancyData & Graph APICRUD · traversal · CDCSearch & RAGvector + graph + textWorkflows & SLAorchestration · HITLSecurity & Auditencryption · immutable trailAnalytics & Opsmetrics · dashboards

Composite MCP

Agents connect once. Purple8's MCP server exposes its 170+ native tools and re-exposes any external MCP servers you register - one endpoint, one auth, one RBAC boundary.

19 Native namespaces

graphschemajourneyragdatamodelsaecsocsupergraphquantumevolutionsentinelguardrailsregistryegressfeedbackschedulermemoryadmin
+

Federated · Register MCP Server

your internal MCPvendor toolsSaaS connectorscustom servers

→ presented to the agent as a single, unified tool surface. Register, remove, or test federated tools at runtime via registry.*.

Structural comparison: a typical production AI stack stitches together ~29 separate services with 4–8 network hops per query. Purple8 runs one. Live figures illustrate steady-state behavior on commodity hardware.

How it is different

The only inverted, fluid-state backend built for AI agents

Inverted

Agents build the backend too

Agents already write the frontend. The backend stayed human-built because its architecture was rigid and opinionated: fixed schemas, separate stores, hand-wired services. Purple8 inverts that. It is built MCP-first, so an agent can ingest data, shape the graph, define workflows, run RAG and route decisions through 170+ tools. The whole application, front to back, is agent-built.

Fluid state

One living state, not eight copies

Documents, vectors, relationships, workflow stages and the audit trail are the same graph. A workflow step is a node, a transition is an edge, a decision is a record. Nothing to sync between stores, nothing to drift, and agents reason over the whole state at once.

One process, zero hops

Database, graph, vector, full-text, workflows and security share memory. No network calls between services, because there are no services.

Memory bounded by hardware

Writes seal and evict in fixed batches, so peak memory tracks the machine, not the corpus. A bigger knowledge base does not mean a bigger server.

Graph-guided retrieval

The graph picks candidates before the vector index runs, so search skips the parts of the index that cannot matter.

Audit is the data

Every transition, AI action and human decision is an immutable edge. Tracing is a query, not a separate product.

Secure from the first line

Encryption at rest, RBAC, tenant isolation and a built-in SOC agent are on by default. There is no harden-later step.

Humans stay in the loop

Workflow stages can require human approval with SLAs. Agents propose; people decide where it matters.

By the numbers

Fast because nothing leaves the process

0.33 ms

query P99

Vector query at 100K nodes, in-process.

40K/s

batch ingest

40,551 documents per second, batches of 1,000.

916 q/s

at 1M vectors

768 dimensions, recall@10 of 0.946 on VectorDBBench.

8.8M

docs on a laptop

Full MS MARCO on a 24 GB machine, ~9 GB peak memory.

23 ms

P95 over HTTP

1024-d search with auth, WAL, encryption and RBAC all on.

0%

errors under soak

1.18M requests, 500 users, 30 minutes.

See the benchmarks and method →

Licensing & Deployment

Evaluate in the Cloud Sandbox. License for VPC, On-Premise & Air-Gapped Production.

Start immediately in the free Cloud Sandbox with all 170+ MCP tools, hybrid GraphRAG pipelines, and Journey workflows. For commercial, enterprise, multi-cluster HA, or OEM deployments across your own VPC, on-premise, Kubernetes, or air-gapped infrastructure - contact our team.

How do I start?

$0 Free · Unlimited MCP Connections

Free Playground ($0)

Evaluate immediately in the Cloud Sandbox (10K nodes), connect your AI agent over MCP in 0 seconds, or run locally via Docker (50K nodes).

Try Hyper Graph

$39 · $149 · $499/mo (Cloud or Self-Hosted)

Builder, Production & Growth

Start with Builder ($39/mo · 250K nodes · BYO LLM), Production ($149/mo · 1M nodes · Daily Backups & HA), or Growth ($499/mo · 5M nodes · Multi-Tenant RBAC & SSO).

View Plans & Subscribe →

Custom Annual License · Sovereign & Enterprise

Enterprise / BYOC (VPC & Air-Gapped)

Deploy inside your own AWS/Azure/GCP VPC, Red Hat OpenShift (UBI9), on-premise, or air-gapped network with unlimited nodes, BYO KMS, SuperGraph, Autonomous SOC, and 24×7 SLAs.

Contact Us

What is inside

Store and search

  • LPG (Property), RDF (Semantic/N-Quads) & Hypergraph store
  • openCypher & semantic triple queries + sub-ms multi-hop traversal
  • Vector, full-text, hybrid search & self-tuning GraphRAG

Intelligence

  • Ranking, communities, centrality
  • Paths and link prediction
  • Optimisation engines

Workflow

  • Journeys with deadlines and SLAs
  • Human approval gates
  • Immutable audit trail

Security

  • Encryption at rest (AES-256-GCM)
  • Roles, API keys, record-level access
  • Multi-tenancy and PII masking

Operations

  • One process, one port
  • Write-ahead log and crash recovery
  • Cloud, on-premise or air-gapped

Interfaces

  • MCP server for AI agents
  • REST and GraphQL APIs
  • Python & TypeScript SDKs