Talon
Now Available

The Digital Backbone
for AI Agents

Multi-model fusion data engine for AI applications. SQL·KV·TimeSeries·MQ·Vector·FTS·GEO·Graph + AI abstraction, 8+1 engines in one.

0.01ms PK Lookup
625K KV Read
8+1 Unified
1362+ Test Cases

8+1 Engines, 1 Shared Storage.
Zero Latency.

A unified architecture replaces a fragmented data stack. Connect your agents to everything with zero overhead.

database
ACID

SQL Engine

ACID transactions, PK lookup <0.02ms, batch insert 165K rows/s. JOIN/CTE/window functions.

SELECT * FROM users

WHERE vector_similarity(embedding, ?) > 0.9

AND last_active > NOW() - INTERVAL '5m';

Throughput 165K rows/s
hub

Vector Search

Custom HNSW index, KNN search P95=0.2ms, SQ8 quantization & metadata filtering.

HNSW SQ8
V
E
C
T
O
R
bolt ACTIVE

Key-Value Store

625K ops/s

GET Throughput

timeline INGEST

Time Series

353K pts/s

Write Throughput

travel_explore

Geospatial

Geohash-based spatial indexing, Redis GEO command compatible.

psychology
smart_toy Core Engine

AI Engine

Native Session/Context/Memory/RAG/Agent/Trace + Hybrid Recall Pipeline (BM25+Vec+Temporal+Rerank+Graph), Auto-Embedding, Auto-Summarize.

Abstractions 6 Primitives
Vector KNN 0.2 ms P95
queue

Message Queue

1.6M msg/s

share

Graph Engine

935K reads/s

search

Full-Text Search

BM25 scoring + inverted index with Chinese tokenizer & hybrid search.

BM25 ES_COMPAT

Simplify Through Unification

Stop managing complex infrastructure glue code.

warning Traditional Stack (Fragmented)

Database Postgres
Cache Redis
Search Elastic
Vector Pinecone
High Latency • Data Silos • Complex ETL
Recommended

check_circle Talon Stack (Unified)

TALON
./talon-server start
Zero-copy access • <1ms latency • Single binary

Built for AI-Native

Core primitives for the modern AI stack.

Explore All Features arrow_forward
memory

Long-Term Memory

Persist agent state effortlessly. Auto-embed memories via configured LLM Provider. Semantic search with automatic query embedding.

manage_accounts

Session Management

Handle millions of concurrent agent sessions with sub-millisecond latency. Smart context window with auto-summarize for long conversations.

psychology

Hybrid RAG

Combine keyword search and vector semantic search for the most accurate retrieval-augmented generation results.