Request a demo
Platform · Specifications

Built from scratch for temporality and scale.

Pometry is a temporal graph system written in Rust on Apache Arrow. The graph lives on disk, carries the full history of every node and edge, and runs from a 10MB binary wherever your data already sits.

<100ms

Query time on a multi-terabyte graph

3M rps

Ingest throughput, on a single laptop

~90%

Lower compute cost than holding the data in memory

10MB

Binary, deployable anywhere

Photographassets-v5/specs-hero.jpg · 2560 × 1040
Temporal model

Time is part of storage, not a property on top.

Most systems keep one current graph and hang timestamps off it, so every historical question becomes a filter across the whole structure. Pometry versions each node and edge in the storage layer instead.

Full history per element

Every node and edge carries its own chronological record, rather than a last-modified stamp on a current-state row.

Point-in-time at flat cost

A query about three years ago addresses the state that existed then. The cost of asking doesn't grow with history.

Temporal algorithms

Run any of them against a past snapshot, window them to the millisecond, or add your own in Rust or Python.

Storage engine

The graph stays on disk.

Systems that hold the working set in memory get sized against the total volume of data rather than the question being asked. Pometry memory-maps files and reads only what a query touches.

Milliseconds from cold to queryable

A multi-terabyte graph answers as soon as the process starts. There is no load phase to sit through first, and no point at which the graph has to fit in RAM.

200x the prior state of the art

Rust, zero-copy columnar execution, SIMD and lock-free parallelism. 200x faster than the US Department of Defense benchmark system.

< 5 min to full recovery

Restore from a copy of the on-disk data. There's no in-memory rebuild to sit through after an outage.

Deployment

Runs where your data already is.

Pometry installs alongside your existing stack and reads from it in place. There is no replatforming step and no data migration.

Environments

On-premise bare metal, private cloud, hybrid, Kubernetes, air-gapped, or embedded inside an existing application.

Footprint

A single 10MB binary. There is no cluster to provision, and no vendor-hosted service between your data and your answers.

Running cost

A multi-terabyte deployment runs on one 128GB instance at roughly $10k a month, against roughly $100k for the same data held in memory across a cluster.

Interfaces

Python SDK, native Rust bindings, GraphQL, and MCP for LLMs and agents.

Batch ingestion

Any Arrow-compatible tabular data: CSV, Parquet, Pandas, DuckDB. Records can arrive unordered and are merged chronologically.

Streaming ingestion

Webhook and event-stream connectors. Changes appear immediately and are kept in the temporal history.

Experience Pometry in action
with a live demo.

Find out how you can close your visibility gap with unique, institutional intelligence.