Crate motedb 

MoteDB

AI-native embedded multimodal database for embodied intelligence (robots, AR glasses, industrial arms). A single embedded Rust library providing columnar storage with ACID transactions, vector search, full-text search, and spatial indexing.

Status

Pre-1.0. The Database embedding API and the storage/transaction engine are stable and heavily tested. The SQL surface and the FFI bindings (ffi, tokenizers) are still evolving — see the crate README for the supported SQL subset. The internal modules (storage, index, txn, database) are exposed for advanced/embedded use but their exact types are not part of the stable API yet.

Quick start

use motedb::{Database, QueryResult};

let db = Database::create("my_data")?;
db.execute("CREATE TABLE t (id INT PRIMARY KEY, v INT)")?;
db.execute("INSERT INTO t VALUES (1, 100)")?;
let r = db.execute("SELECT * FROM t")?;
if let QueryResult::Select { rows, .. } = r.materialize()? {
    println!("{:?}", rows);
}

Architecture

• Storage: append-only columnar segments (source of truth) + WAL for durability, with Snappy/Zstd compression and mmap zero-copy reads.
• Indexes: DiskANN/Vamana (vector) + i-Octree (spatial) + inverted index (text) + B+Tree (column/timestamp).
• Transactions: MVCC version store with snapshot isolation and write-ahead logging for crash recovery.

Performance (indicative, Apple Silicon)

On a 300K-row × 4-column workload vs SQLite WAL: COUNT/SUM under WHERE ~5×, ORDER BY + LIMIT ~2.5×, PK point lookup sub-microsecond. See BENCHMARK.md and the docs for methodology and full numbers.

Re-exports

pub use config::AutoCheckpointConfig;

pub use config::DBConfig;

pub use config::DurabilityLevel;

pub use config::LSMConfig;

pub use config::WALConfig;

pub use catalog::TableRegistry;

pub use database::MoteDB;

pub use database::QueryProfile;

pub use database::TransactionStats;

pub use sql::ForEachResult;

pub use sql::QueryResult;

pub use sql::StreamingControl;

pub use sql::StreamingQueryResult;

Modules

cache

Cache module - LRU caches for performance optimization

catalog

config

Database configuration and durability levels

database

Database Module - Modularized Architecture

distance

Distance metrics for vector similarity computation

ffi

FFI (Foreign Function Interface) for C/Python/Node.js

index

Index layer implementation

sql

storage

Storage layer implementation

tokenizers

txn

Transaction layer implementation

types

Multi-modal data types for MoteDB

Macros

debug_log

Debug-level log. Compiled in but a no-op unless a logger is installed and debug level is enabled for the motedb target.

error_log

Error-level log (operation failed; the calling path returns an error too).

info_log

Info-level log (e.g. lifecycle events: open/close/checkpoint).

warn_log

Warn-level log (degraded-but-functional conditions, retries, fallbacks).

Structs

Database

MoteDB 数据库实例

Enums

StorageError

Functions

fsync_dir

Explicitly purge jemalloc’s dirty pages back to the OS. Fsync the parent directory of a file path. This ensures that a file rename or creation is durable across crashes on POSIX systems (Linux ext4/xfs, macOS APFS). Without this, a rename is not guaranteed to survive a crash even if the file itself was fsync’d.

purge_memory_to_os

Call after bulk operations (CREATE INDEX, compaction) that create large transient allocations, to keep RSS low on edge devices. No-op when jemalloc is not enabled.

Type Aliases

MoteDBError

Result