Everything in one binary
Object, graph, time-series, geo, relational, vector and text — one durable store, one language, one process. Replace a rack of specialized databases and services with a single ~4.6 MB self-contained binary you host yourself.
One store, every shape of data
One transactional store, every data model side by side — replacing seven specialized databases.
Objects & graph
Typed objects on disk, traversed with a dot — the graph is your model. No JOINs, SQL or Cypher.
Native time-series
nodeTime with as-of resolveAt queries and
server-side sampling.
Indexes & lists
nodeIndex for ordered O(log n) keys; nodeList
for large sparse lists.
Native geospatial
nodeGeo spatial index and geo type — distance,
bearing, geohash and geometry.
Tabular & relational
Table and CSV with automatic schema inference.
Vector & text indexes
Built-in VectorIndex and full-text TextIndex —
detailed in AI-ready.
Tensors
N-dimensional Tensor type with FFT and SIMD.
Durable & evolvable store
LMDB-backed and transactional — online defrag, delta backups and ABI-aware schema migration.
Built-in graph explorer
Browse and visualize nodes and relationships in the browser, built in.
One language, easy to build & maintain
A strongly-typed language with the tooling of a modern stack.
One strongly-typed model
One type system from storage to API — no ORM, no mapping, errors caught at compile time.
Analyzer, LSP & formatter
Static analyzer, LSP (VS Code, JetBrains, Zed) and an opinionated formatter.
First-class testing
A built-in @test framework.
One-command typed SDKs
Typed client SDKs for TypeScript, Python, Java, Rust and C in one command.
Auto-generated OpenAPI v3
OpenAPI v3 generated from your function signatures.
Modules & version pinning
@library modules with reproducible per-project version
pinning.
AI-ready
Search, embeddings, RAG and ML in the same engine — no extra infrastructure, no data leaving the box.
Full-text & hybrid search
One TextIndex, 15 modes (BM25, fuzzy, phonetic, semantic…),
fused with vectors via RRF.
Built-in vector index
VectorIndex with HNSW search — cosine, L2 and squared-L2.
On-device embeddings
embed / embed_batch in-process on a
statically-linked llama.cpp.
33 languages & RAG chunking
33-language tokenization and RAG-style document chunking.
Knowledge-graph RAG
Combine graph, text and vector across tiered indexes for grounded, explainable retrieval.
In-engine ML & analytics
PCA, k-means, neural nets (Dense/LSTM/GRU/Conv2D), streaming stats and DTW/SAX patterns.
Built-in MCP server
Two annotations turn any function into a callable MCP tool.
Open skills marketplace
An open AI-agent skills marketplace at github.com/datathings/marketplace.
On-device LLM generation roadmap
On-device text generation and chat — defined in the API, on the roadmap.
One binary, easy to ship
One process to deploy, secure and back up — not a stack of six services.
One self-contained binary
A ~4.6 MB binary replacing DB, search, vector store and web/app server.
HTTP server & RPC
JSON-RPC 2.0, path-RPC, gzip/keep-alive and a /files API.
Serve your web app
Static web serving from a webroot — ship the UI from the same binary.
Scheduler & resumable jobs
Cron-like scheduler and parallel jobs that checkpoint and resume.
Telemetry & reflection
Introspect and observe the running application.
Connectors & domain libraries
PostgreSQL, Kafka, OPC-UA, MQTT, S3, SMTP, SFTP/FTP, OSM, IFC/BIM — plus weather, solar and power-grid.
Yours to run
Self-hosted and on-device — your data and AI never leave the box.
Self-hosted & EU-sovereign
Runs on your own infrastructure — cloud, datacenter or air-gapped; built in Luxembourg (EU).
On-device AI
Embeddings and inference in-process — no external API, no third parties.
Edge to terabytes
ARM/Raspberry Pi to terabytes and billions of nodes, same engine.
Secure by default
One identity and access model across data, APIs, files and AI tools.
Unified RBAC
@permission / @role and token auth govern data,
endpoints, files and MCP tools alike.
Enterprise OIDC SSO
OpenID Connect SSO with Authorization Code PKCE and JWKS.
Crypto toolbox
SHA-256, HMAC, RSA and UUID, backed by mbedTLS.
Fast where it counts
Written in C, accelerated where it matters, scaling from edge to terabytes.
C-fast & in-process
SIMD/C hot paths, ~1.7M rows/s ingest, no network hops between data and logic.
Billions of nodes, no cluster
Terabytes and billions of nodes on a single node — no sharding.
Many-worlds branching coming soon
What-if simulation via many-worlds branching.