AletheiaDB

A high-performance bi-temporal graph database in Rust, combining graph traversal, vector similarity search, and full temporal history in a single consistent query — with no external dependencies.

Most databases answer what is true now. AletheiaDB also answers what was true at any point in the past and what is semantically related, all from one query. Why that matters →

Install

[dependencies]

aletheiadb = "0.1"

Requires Rust 1.92+.

Quick Start

use aletheiadb::prelude::*;

fn main() -> Result<()> {
    let db = AletheiaDB::new()?;

    // Create nodes and a relationship
    let alice = db.create_node("Person", properties! { "name" => "Alice" })?;
    let bob   = db.create_node("Person", properties! { "name" => "Bob"   })?;
    db.create_edge(alice, bob, "KNOWS", properties! {})?;

    // Snapshot the time before an update
    let before = aletheiadb::time::now();

    // Update Alice
    db.write(|tx| tx.update_node(alice, properties! { "role" => "engineer" }))?;

    // Time-travel: what did Alice look like before the update?
    let past = db.get_node_at_time(alice, before, before)?;
    assert!(past.properties.get("role").is_none());

    Ok(())
}

See the Getting Started guide for a full walkthrough.

Hybrid Queries

Graph traversal, vector ranking, and temporal snapshots compose into a single query with a consistent view of the data:

use aletheiadb::prelude::*;
use aletheiadb::HnswConfig;

fn main() -> Result<()> {
    let db = AletheiaDB::new()?;
    db.vector_index("embedding").hnsw(HnswConfig { dimensions: 2, ..Default::default() }).enable()?;

    let alice = db.create_node("Person", properties! { "name" => "Alice" })?;
    let query_embedding = vec![0.1, 0.2];
    let valid_time = aletheiadb::time::now();
    let tx_time = aletheiadb::time::now();

    let _results = db.query()
        .as_of(valid_time, tx_time)              // temporal: point-in-time snapshot
        .start(alice)                            // graph: starting node
        .traverse("KNOWS")                       // graph: follow edges
        .rank_by_similarity(&query_embedding, 10) // vector: re-rank by similarity
        .execute(&db)?;

    Ok(())
}

See the Hybrid Query guide.

Performance

Benchmarks run on every push to trunk. 📊 Latest results

Averages across 30–212 datapoints of continuous CI runs:

Feature Flags

All features are off by default except config-toml.

nova does not include semantic-search — use both flags if you want everything:

aletheiadb = { version = "0.1", features = ["nova", "semantic-search"] }

Documentation

MCP Server (Claude / LLM Integration)

cargo run --bin aletheia-mcp --features mcp-server

Exposes AletheiaDB as a set of MCP tools over stdio: node/edge CRUD, multi-hop traversal, vector search, temporal queries, and hybrid queries. Compatible with Claude, Claude Code, and any MCP-capable host.

Contributing

git clone https://github.com/madmax983/AletheiaDB

cargo build

just check-all   # format + lint + test + coverage

See DEVELOPMENT_WORKFLOW.md. All PRs must maintain ≥ 85% line coverage and pass cargo clippy -- -D warnings.

Known Limitations

Orphaned edges: delete_node removes only the node; edges connecting to it remain. Use delete_node_cascade to atomically remove the node and all its edges.

License

Licensed under either of MIT or Apache 2.0 at your option.