CLUE
NEW

CLUE

Every answer leaves a clue.

The knowledge database for AI. One database engine unifies vector, keyword and knowledge graph: four RAG modes behind a single API.

Knowledge DBVector · keyword · knowledge graphRAG · 4 modesPinecone-compatibleHA · sharded cluster
Join the MVP waitlist
clue.aiclude.com / playground
G0000003
admin@aiclude.complatform_admin

RAG Playground

Run Classic · Hybrid · Graph · Graph N-hop in parallel and compare quality, latency, and cost.

Query settings

Specify the query and hyperparameters, then run the selected modes in parallel.

Modes to run
4 modes selected
Core capabilities

Organize your knowledge so AI can answer with it.

Stop running a vector DB, a search engine and a graph DB side by side. CLUE is one engine.

Four RAG search modes

Basic (vector), Hybrid (vector + keyword), Knowledge Graph (entity links), Multi-hop reasoning: switch modes in one parameter.

30+ format auto-ingest

Drop in PDFs, DOCX, XLSX, images, audio or video and they are queryable in 30 seconds.

Vector · knowledge graph · keyword in one txn

Vector, knowledge graph and keyword indexes live in the same database transaction.

Tenant isolation (data + key split)

Per-tenant physical tables plus row-level access controls give SaaS workloads a hard boundary.

HA · sharded cluster

RTO 30s, RPO 0 synchronous replication; scale past a billion vectors with horizontal sharding when you need it.

10 migration adapters

Move from Pinecone, Weaviate, Qdrant, Milvus, Chroma, Elasticsearch, PostgreSQL vector extension and more in one click.

Architecture

File to answer, in one transaction.

Average 30 seconds from upload to answer: every step lives on the same ACID engine.

PipelineFile → Answer · single Postgres transaction
Ingest
Upload · 30+ formats
Chunk
Semantic splitting
Embed
1024-d text
Store
Vector · keyword · knowledge graph
Search
Route across 4 modes
Answer
LLM + grounded citations
4 ModesSame API · `mode` param swap
mode = "classic"
Classic
Vector index · vector similarity
mode = "hybrid"
Hybrid
Vector + keyword hybrid
mode = "graph"
Graph
Entity-graph traversal
mode = "multi-hop"
Multi-hop
Community detection + multi-hop reasoning
Engine
Vector SearchKeyword SearchKnowledge GraphDistributed ClusterBackup · RestoreKubernetes-nativeText Embedding

Stop syncing a separate vector DB, search engine and graph DB: consistency comes for free.

Self-hosted Stack

Self-hosted Docker stack

Start with a single-host profile. Scale to HA, sharded clusters, or Kubernetes.

Docker Containers
API · SDK clients
Monitoring
clue-fenginx
Admin console
Nginx + React. Collections, search and ops UI.
clue-be
API server
REST + gRPC. Routes search across the four modes.
conn-pool
Connection pool
Transaction pooling keeps connection counts down.
clue-serverengine
Vector · keyword · knowledge graph engine
Vector, keyword and knowledge graph in one transaction.
File input (PDF · DOCX · audio · video)
redisqueue
Redis
Ingest queue and search cache backend.
clue-ingestpipeline
Ingest pipeline
PDF, DOCX, image, audio and video, all auto-processed.
clue-embedding
Embedding
1024-d text embeddings.
clue-clip
CLIP
Image and caption embeddings.
db-exportermetrics
DB metrics exporter
prometheusscrape
Prometheus
Metric scraping + alert rules.
grafanadashboards
Grafana
Search latency and ingest throughput dashboards.
  • Minimal (single)
    vCPU
    4 vCPU
    Memory
    16 GB
    Disk
    100 GB SSD

    Single node, Community license or higher.

  • Recommended (HA)Recommended
    vCPU
    4 vCPU
    Memory
    16 GB
    Disk
    200 GB SSD

    3 nodes, Professional license, RTO 30s.

  • Large (sharding)
    vCPU
    8 vCPU
    Memory
    32 GB
    Disk
    500 GB SSD

    1 coordinator + 3 workers, Enterprise license.

  • K8s (CloudNativePG)
    vCPU
    K8s 1.27+
    Memory
    CNPG Operator
    Disk
    PVC

    Kubernetes-native, Enterprise license.

  • One database handles vector, keyword and knowledge graph, so no separate search engine is needed.
  • Reuse an external embedding container to share GPU nodes across services.
  • Lift-and-shift to Kubernetes with CloudNativePG without changing the container set.
Use cases

For every AI that needs internal knowledge.

AICLUDE itself runs on CLUE: battle-tested through dogfooding.

Internal wiki RAG

Ingest Confluence, Notion or SharePoint and serve answers scoped by department permissions.

10x faster lookup

Code & doc search

Tree-sitter code chunking and semantic doc chunking share one index across IDE and chatbot.

30% review time saved

Customer support KB

FAQs, tickets and manuals unify so chatbot and agent see the same source of truth.

+25% first-touch resolution

Multimodal KB

PDFs, images, audio and video embed into the same collection for true multimodal search.

Format-agnostic search
Why CLUE

Everything enterprise expects, on one database engine.

Vector · keyword · knowledge graph in one engine
Vector, keyword and knowledge graph indexes share one engine: no dual-index drift, no split-brain consistency.
HA · sharded cluster
RTO 30s · RPO 0 sync replication plus horizontal sharding: built for enterprise loads.
Audit log · disk encryption · column encryption
Audit logging, disk-level encryption and PII column encryption with search-while-encrypted indexes: standard.
Pinecone-compatible SDK
Pinecone-compatible SDK; existing client code runs as-is.
10 migration adapters
Pinecone, Weaviate, Qdrant, Milvus, Chroma, Elasticsearch, MongoDB Atlas, Redis VSS, PostgreSQL vector extension: one click.
Dogfooded by AICLUDE
AICLUDE runs on CLUE. Operational know-how, recovery and tuning come bundled.
Reserve your spot

Every answer leaves a clue.

MVP waitlist members get a 30-day trial and free migration consulting.

Join the MVP waitlist