Your entire backend.
One process. Zero to wire.
Purple8 is the AI-native backend that replaces 20+ services — database, vector search, graph, RAG, document intelligence, agentic process automation, auth, and encryption — with one embedded engine. AI agents automate end-to-end workflows out of the box. You build the frontend; Purple8 is everything behind it.
Free forever · No credit card · Runs on your own machine · Upgrade in one click at launch
* Measured on real, reproducible benchmark runs. See methodology & conditions →
How Purple8 works
Your entire backend. One process.
You write 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.
Composite MCP
Agents connect once. Purple8's MCP server exposes its 74 native tools and re-exposes any external MCP servers you register — one endpoint, one auth, one RBAC boundary.
Native · 11 namespaces
Federated · Register MCP Server
→ 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.
Why Purple8
A category of one.
One process. No orchestration tax.
Storage, search, workflow, RAG, auth, and encryption — single binary, single port. Replaces 20+ services before you write a line of code.
AI agents operate it natively.
A first-class agent interface with 82 tools across 11 namespaces, RBAC-enforced. Claude, Cursor, and Copilot connect directly — no wrapper, no glue code.
Three RAG modes. Self-tuning.
Flat vector, graph-augmented, and hybrid retrieval built in. The system profiles your corpus and persists the optimal configuration automatically.
Agentic process automation. Built in.
Stateful workflows, SLA enforcement, human-approval gates, and an immutable audit trail — all in the same process. AI agents orchestrate end-to-end business processes through 82 MCP tools, with no DAG files, no separate orchestration service, no tracing backend. Replaces LangChain, LangGraph, LangSmith, and Airflow.
29 services become one.
The modern AI stack is a sprawl of databases, vector stores, orchestration, auth, and observability — each its own service, its own bill, its own failure mode. Two products, Purple8 and DocIntel, replace all of it.
You start 9–15 GB ahead.
Before a single record is stored, a 29-service stack has already reserved most of a machine — just to stand up. Purple8's entire runtime comes up in about 350 MiB. The rest of the box goes to your work, not to the plumbing between services.
reserved before any data is stored
Even at a charitable 50 MB floor per service, the runtimes alone are ~1.45 GB. The rest is the connective tissue: network hops between services, the same rows cached three to five times over, and the thread and connection pools every service holds to talk to every other.
one runtime, up and ready
One process, one thread pool, one cache. No wire between services, no duplicated caches, no cross-service pools. Data is streamed from disk on demand behind a bounded working window — so the footprint is set by the machine, not the corpus.
What that overhead is made of
29 separate runtimes — interpreter, heap, idle worker threads — each loaded before storing a single row.
One query crosses 4–8 service hops. Every hop carries socket buffers, a TLS session, and serialize-then-deserialize on both ends.
The same document sits in three to five caches at once — rows, nodes, index segments, vectors — plus the CDC fabric that keeps them in sync.
Every service runs its own pool, and every caller holds a pool to it — pre-opened connections that scale with the square of the service count.
This isn't about being slightly better at each service. The 9–15 GB and 28 services sitting between them are overhead engineering teams carry forever — on every node, at idle. We removed them by construction, not by optimisation. What's left is technology that's cleaner for the environment, simpler to use, faster to build with, and better on ROI — so teams keep their focus on their customers' value, while the whole stack comes out end-to-end cost-optimized as a consequence. And because the footprint is bounded by the machine, not the data, that value widens as the corpus grows and compounds as tenants multiply.
28 fewer idle runtimes, no cross-service chatter, no duplicated caches — less energy and water per unit of work, because the overhead is architecturally absent, not tuned away.
One process, one runbook, one thing to secure, one metrics endpoint — instead of a 29-service mesh to orchestrate and babysit.
The backend is already one thing — no stitching 20+ services together before you ship. Teams build the frontend and deliver customer value sooner.
One bounded node instead of twenty. Spend goes to outcomes, not idle infrastructure — and the saving compounds with every tenant you add.
We stay focused on our customers' value, so they can stay focused on theirs — and the whole stack comes out end-to-end cost-optimized as a result.
No API gateway. No sidecar. No Kubernetes operators. Just one process, one port, one binary.
Products
Two products. Your entire backend.
Purple8 is the backend your app runs on — storage, search, RAG, workflows, auth, and encryption in one process. DocIntel turns any document into queryable knowledge inside it. Together they replace 20+ services — you build only the frontend.
Purple8
A category of one
A unified AI-native backend — storage, vector search, document store, full-text, workflow orchestration, RAG, auth, and encryption in a single process. No JVM. No external services. Docker image and pip wheel available.
- Write only the frontend — storage, search, RAG, workflows, auth, and encryption are all Purple8
- Storage, vector search, document store, and full-text — one embedded engine, one process, one port
- Native hybrid search with a cost-based query planner — no separate search service
- Three built-in RAG modes: flat vector, graph-augmented, and hybrid retrieval
Purple8 DocIntel
Document intelligence microservice
A stateless document intelligence microservice that parses 70+ enterprise document formats — including CAD, BIM, sketches, and audio — through a proprietary extraction model and LLM-assisted relationship mapping. Emits structured knowledge directly into Purple8.
- PDF, DOCX, XLSX, PPTX, HTML, EML, EPUB, TXT/CSV — all standard enterprise formats
- IFC/IFCZIP, DXF, DWG, STEP, STL, OBJ, GLTF/GLB, G-code — full CAD & BIM native
- SAP IDocs, financial XML, structured config formats
- Proprietary NER model — two-pass hybrid: named entity extraction + LLM relationship mapping
We run our own stack on Purple8.
Zero external databases.
Purple8's backend operations and customer portal — the entire commercial backend — runs on a single Purple8 instance. No Postgres. No Redis. No message queue. One process, one port.
If we trust it for our own operations, you can trust it for yours.
# Purple8 Command Center
Customer ──PURCHASED──► LicenseKey
LicenseKey ──ISSUED_FOR──► Product
StripeEvent ──TRIGGERED──► LicenseKey
Install ──SENT──► TelemetryEvent
Find your tier in 10 seconds
What are you building?
No node counting, no capacity math. Tell us the app and how many people it serves — we'll point you to the right tier. You still start free either way.
Knowledge base + conversation history over a RAG pipeline.
Total people over the app's life — not concurrent users. (500 concurrent usually means millions of total customers.)
An AI chatbot / assistant for 10,000 users — we recommend:
~80K graph objects at your scale fits Micro.
≈ 80K graph objects · 500K capacity
Upgrade only when you outgrow the free tier. No card to start.
Pricing
Start free. Pay for capacity, never for features.
Self-hosted — runs in your own infrastructure, nothing leaves your environment. Every production tier from Micro to Starter ships the exact same feature set. Moving up the ladder buys more capacity, not more capability.
Developer
50K nodes · 1 MCP agent
The full engine on your machine — every RAG pipeline, every MCP tool, every algorithm, and the full Journey Engine (workflows, HITL, SLA). No credit card, no expiry.
Download freeMicro
500K nodes · 3 MCP agents
Production scale starts here — 10× the capacity, 3 agents & seats, plus agentic/late-chunking RAG, local KMS encryption, data lineage and local backup.
Mini
1M nodes · 3 MCP agents
Same full production feature set as Micro — twice the capacity for growing workloads.
Growth
3M nodes · 3 MCP agents
The comfortable middle. Full production feature set at 3M capacity — before you need multi-project or compliance.
Starter
5M nodes · 3 MCP agents
Top of the production family — maximum single-project capacity with the same full feature set.
Pro
50M nodes · 10 MCP agents
Multi-project & compliance-ready: SSO, RBAC, immutable audit, PII, CDC, durable writes, Applied Graphs, AEC Core, cloud backup.
Enterprise
Unlimited nodes · Unlimited agents
Mission-critical scale: Raft HA, sharding, federation, SOC vertical, managed KMS, Supergraph, AEC Advanced.
Contact salesThe full production engine is $119/mo. Micro, Mini, Growth, and Starter are the same product — you pick the capacity you need and upgrade only when you outgrow it.
Not sure which one? Use the tier finder ↑ · All paid tiers will be available on AWS, GCP & Azure Marketplace soon.
Questions about which plan fits your team?
Talk to us →