Crate taladb 

TalaDB — an embedded, local-first document database with built-in vector search, for Rust apps. It runs inside your process and stores everything in one file; the same engine powers TalaDB in the browser, Node.js, React Native, Android and iOS.

Getting started

use serde::{Deserialize, Serialize};
use serde_json::json;

#[derive(Serialize, Deserialize)]
struct Book {
    title: String,
    year: i64,
}

let db = taladb::Database::open_in_memory()?; // or taladb::open("app.db")
let books = db.typed::<Book>("books")?;

books.insert(&Book { title: "Noli Me Tángere".into(), year: 1887 })?;
let found = books.find(json!({ "year": { "$gte": 1800 } }))?;
assert_eq!(found.len(), 1);

Database::typed gives a TypedCollection of your own serde types, queried with the JSON filter language every TalaDB platform shares. The untyped Collection underneath works with Values and typed Filters directly.

What’s in here

• TypedCollection — serde-typed documents, JSON filters and updates, and typed live queries (TypedWatch).
• json — the JSON query language: documents, filters and updates.
• Database — open a database (Database::open, Database::open_in_memory, or Database::open_with_backend for a custom storage backend such as OPFS in the browser) and hand out Collections.
• Collection — insert, query, update, and index documents.
• Filter and FindOptions — the query language: comparisons, logical combinators, sorting, pagination, and projection.
• vector — vector indexes and find_nearest, exact by default and persistent approximate (HNSW) on every platform.
• bm25 / fts — full-text search over the same documents.
• TalaDbError — one error type for everything, carrying the originating storage error as a source.

Scope

TalaDB is an embedded database: it owns on-device storage and retrieval, and deliberately does not replicate. Applications that need to notify a backend of local writes use the change webhook in the taladb TypeScript client, which works identically on every runtime.

Runtime targets

Native Rust, the browser (via wasm32-unknown-unknown), Node.js, React Native, Android and iOS all run this crate. Anything that would bloat a WASM bundle is feature-gated — see the config-yaml, encryption and legacy-migration features.

Re-exports

pub use aggregate::Accumulator;

pub use aggregate::GroupKey;

pub use aggregate::Pipeline;

pub use aggregate::Stage;

pub use audit::AuditEntry;

pub use audit::AuditOp;

pub use audit::read_audit_log;

pub use audit::read_audit_log_since;

pub use collection::Collection;

pub use collection::CollectionIndexInfo;

pub use collection::Update;

pub use config::DurabilityConfig;

pub use config::TalaDbConfig;

pub use config::load_auto;

pub use config::load_from_path;

pub use document::Document;

pub use document::Value;

pub use document::derive_doc_id;

pub use engine::RedbBackend;

pub use engine::StorageBackend;

pub use error::TalaDbError;

pub use migration::BUILTIN_MIGRATIONS;

pub use migration::CURRENT_SCHEMA_VERSION;

pub use migration::Migration;

pub use migration::run_migrations;

pub use query::Filter;

pub use query::options::FindOptions;

pub use query::options::SortDirection;

pub use query::options::SortSpec;

pub use time::now_ms;

pub use typed::Scored;

pub use typed::TypedCollection;

pub use typed::TypedWatch;

pub use vector::HnswOptions;

pub use vector::VectorMetric;

pub use vector::VectorSearchResult;

pub use collection::VectorBuildProgress;

pub use collection::VectorExecution;

pub use collection::VectorIndexStatus;

pub use collection::VectorQueryOptions;

pub use collection::VectorQueryResult;

pub use collection::VectorSearchMode;

pub use redb;

pub use redb2;

pub use serde_json;

Modules

aggregate

Aggregation pipeline for TalaDB.

audit

Append-only audit log for TalaDB.

bm25

BM25 relevance scoring for the full-text index.

collection

config

TalaDB config loader.

crypto

Encryption at rest for TalaDB.

document

engine

error

The crate-wide error type.

fts

Full-text search index for TalaDB.

index

Secondary index key encoding.

json

TalaDB’s JSON query language: documents, filters and updates as JSON.

json_depth

Depth guard for untrusted JSON, applied before any recursive parser walks it.

migrate

Bringing a pre-redb-4 database file forward.

migration

query

time

Wall-clock helpers.

typed

Serde-typed collections: store and load your own structs, query with JSON.

vector

Exact vector storage and scoring. Persistent HNSW lives in vector_graph.

watch

Live queries / reactive subscriptions for TalaDB.

Structs

Database

The main TalaDB database handle.

GraphOptions

Ulid

Re-exported so bindings can parse and construct document ids without taking a direct ulid dependency (and risking a version skew against the engine’s). A Ulid is a unique 128-bit lexicographically sortable identifier

VectorCacheStats

Estimated decoded search allocations. Reservations conservatively charge the full allowance of active graph loans (including large-filter bitmaps) and vector-block construction. Capped filter ID sets, page windows/group keys, result documents, storage and build scratch are outside these estimates.

Enums

MemoryPressure

Host-reported pressure, independent of the configured normal budget. Recovery is explicit: a timer or app foreground event does not prove that memory pressure has ended.

Quantization

VectorCachePolicy

Functions

open

Open (or create) the database file at path. Shorthand for Database::open.