ChronosVector (CVX)

A high-performance temporal vector database that treats time as a geometric dimension of embedding space. ChronosVector answers not just "what is similar?" but "what was similar, what changed, and what will be similar?"

Documentation

Tutorials (with interactive Plotly visualizations)

Key Features

• Spatiotemporal kNN — composite distance d_ST = α·d_semantic + (1-α)·d_temporal
• 19 analytical functions — velocity, drift, Hurst, changepoints, path signatures, topology, anchor projection
• Anchor projection — project trajectories from ℝᴰ to interpretable ℝᴷ coordinates (RFC-006)
• SIMD-accelerated — auto-dispatched AVX2/NEON via pulp (cosine, L2, dot product)
• Index persistence — save/load HNSW graph via postcard binary serialization
• Tiered storage — hot (in-memory/RocksDB) → warm (file-based) with automatic compaction
• Crash safety — WAL with CRC32 validation and segment rotation
• Python bindings — pip install chronos-vector (PyO3/maturin)

Python Quick Start

import chronos_vector as cvx
import numpy as np

# Create and populate index
index = cvx.TemporalIndex(m=16, ef_construction=200)
index.bulk_insert(entity_ids, timestamps, vectors)
index.save("index.cvx")  # persist for fast reload

# Trajectory analysis
traj = index.trajectory(entity_id=1)
vel = cvx.velocity(traj, timestamp=t)
h = cvx.hurst_exponent(traj)
cps = cvx.detect_changepoints(1, traj)

# Anchor projection — measure relative to reference points
projected = cvx.project_to_anchors(traj, anchors, metric='cosine')
summary = cvx.anchor_summary(projected)  # {mean, min, trend, last}

# All analytics work on projected trajectory too
vel_anchor = cvx.velocity(projected, timestamp=t)
sig = cvx.path_signature(projected, depth=2)

Performance

Architecture

                    ┌──────────┐
                    │ cvx-api  │  REST + gRPC endpoints
                    └────┬─────┘
                         │
              ┌──────────┼──────────┐
              │          │          │
        ┌─────┴──┐ ┌─────┴──┐ ┌────┴──────┐
        │ ingest │ │ query  │ │ analytics │  19 functions
        └────┬───┘ └────┬───┘ └────┬──────┘  + anchor projection
             │          │          │
        ┌────┴──────────┴──────────┴────┐
        │           cvx-index           │
        │  ST-HNSW + SIMD + Roaring BM  │  save/load persistence
        └──────────────┬────────────────┘
                       │
        ┌──────────────┴────────────────┐
        │          cvx-storage          │
        │  Hot → Warm + WAL + RocksDB   │
        └──────────────┬────────────────┘
                       │
        ┌──────────────┴────────────────┐
        │           cvx-core            │
        │  Types, Traits, Config, Error │
        └───────────────────────────────┘

8 crates + Python bindings, 280+ tests.

Rust Quick Start

# Build
cargo build --workspace

# Run tests
cargo test --workspace

# Generate API docs
cargo doc --workspace --no-deps --open

# Start server
cargo run --release -p cvx-server

Python Installation

# From source (requires Rust toolchain)
cd crates/cvx-python
maturin develop --release

# Or with pip (when published)
pip install chronos-vector

Cross-Domain Research

CVX has been validated across 7 investigations in 6 domains. See the White Paper for details.

License

Licensed under the Elastic License 2.0.

Free to use for research, education, and internal purposes. Commercial use as a managed service requires a separate license. See LICENSE for details.