HeliosDB Lite is a self-hosted, PostgreSQL-wire-compatible database for production SaaS workloads that bundles HeliosProxy for connection pooling and routing with the HeliosCore storage engine. It runs in-process or as a server, serves the PostgreSQL wire protocol and a REST API, runs SQL and PL/pgSQL routines, and, in builds with the ha-tier1 feature, streams its write-ahead log to a warm standby. Licensed SSPL-1.0 and available to licensed customers through sales; current source release v3.6.0.
Last updated
Edge database with HeliosProxy, the connection router with multi-tier caching, a GraphQL gateway and query routing. It shares the HeliosDB SQL core, vector search, branching and time travel, and adds HTAP storage, PL/pgSQL stored procedures, online backups and 94 system views for SaaS platforms, edge computing and AI agent infrastructure.
Lite is designed for teams that run production SaaS workloads without the complexity of a distributed system.
# /etc/heliosdb/heliosdb.toml
[server]
bind = "0.0.0.0:5432"
max_connections = 500
[proxy]
enabled = true
[proxy.cache]
l1_size = "256MB"
l2_size = "1GB"
[proxy.graphql]
enabled = true
port = 4000
[storage]
engine = "helioscore"
direct_io = trueShips an AGENTS.md aggregate plus a .claude/skills/ catalogue. Every CLI flag, REPL meta-command and public API indexed for LLM coding agents. Read the guide →
cargo install --path ./heliosdb-lite
heliosdb-lite skills install ~/.claude/skillsNot yet. Code-Graph (hdb_code), Graph-RAG (WITH CONTEXT) and the native MCP server ship in HeliosDB Nano as build-from-source features; equivalents for Lite are planned. The cards below describe the Nano features.
CREATE EXTENSION hdb_codeSQL-callable LSP via lsp_* table functions. ON BRANCH + AS OF on every call. Eight languages built-in plus a runtime grammar registry.
WITH CONTEXT SQL clauseVector pre-filter + BFS expand + centrality re-rank, expressed as a single SQL clause. Eight ingestion adapters incl. Docling-backed PDF / Office / audio / image.
In Nano: POST /mcp, /mcp/ws and /mcp/sse with 16 core tools, built with --features mcp-endpoint. A native MCP server for HeliosDB Lite is planned.
In Nano: build with --features code-embed for fastembed-rs + BGESmallENV15 (384-dim) in-process, with no embedding API to deploy. Planned for Lite.
Lite is the licensed production tier: it bundles HeliosProxy for connection pooling and routing, adds the HeliosCore storage engine and runs SQL and PL/pgSQL routines, while vector search, branching, time-travel queries and encryption are available in both editions. Nano is the Apache 2.0 embedded engine; Lite is licensed through sales.
HNSW + PQ compression with 1K-5K QPS (internal tests, HeliosDB Lite embedded SQL path, HNSW + PQ; workload-dependent).
Git-like branching at the storage layer.
AS OF syntax for point-in-time access.
TDE (enabled in the configuration with a key source), an opt-in FIPS 140-3 build mode and Row-Level Security. Zero-Knowledge Encryption is planned.
HeliosProxy adds connection pooling, read/write splitting, health checks and failover handling in front of Lite, and its optional platform features add query caching, a GraphQL gateway and WASM plugins. Lite bundles it; HeliosProxy is also a standalone Apache 2.0 proxy that works with every HeliosDB edition and with stock PostgreSQL.
# /etc/heliosdb/proxy.toml
[cache]
l1_size = "256MB"
l2_size = "1GB"
[graphql_gateway]
enabled = true
[routing]
mode = "workload_aware"
read_preference = "nearest"
# Load a WASM plugin
[[plugins]]
name = "rate_limiter"
path = "plugins/rate_limiter.wasm"HeliosDB Lite runs SQL and PL/pgSQL routines; PL/SQL, T-SQL and DB2 SQL PL are planned. Lite also accepts some Oracle-style functions today, such as NVL, SYSDATE and TO_CHAR, which helps when moving queries from Oracle. See the procedural language guide.
LANGUAGE plpgsqlLANGUAGE sql-- PL/pgSQL function
CREATE FUNCTION get_similar_docs(
query_vec VECTOR(768),
limit_n INT
) RETURNS TABLE (
id INT, title TEXT, score FLOAT
) AS $$
BEGIN
RETURN QUERY
SELECT d.id, d.title,
1 - (d.embedding <=> query_vec)
FROM documents d
ORDER BY d.embedding <=> query_vec
LIMIT limit_n;
END;
$$ LANGUAGE plpgsql;HeliosDB Lite serves the PostgreSQL wire protocol and a REST API (JSON over HTTP, OpenAPI-documented, JWT or API-key auth). A gRPC client API, an MCP server, an Oracle (TNS) listener and MySQL wire-protocol support are planned.
psql/v1 and /api/v1 on the HTTP port (8080 by default)/openapi.json, with /docs and /redoc# PostgreSQL wire protocol
psql -h localhost -p 5432 -U app
# REST API - run SQL on the main branch
curl -X POST http://localhost:8080/v1/branches/main/query \
-H "Authorization: Bearer $TOKEN" \
-H "Content-Type: application/json" \
-d '{"sql": "SELECT * FROM users LIMIT 10"}'Over the PostgreSQL wire protocol or the REST API today. A native MCP server for HeliosDB Lite is planned; HeliosDB Nano ships one as a build-from-source feature. HeliosDB's AI add-ons, AI Support and the AI-DBA Agent, are available for Lite.
A release focused on the hot path: index-backed DELETE, batched TRUNCATE, and broad query-engine wins across JOINs, DISTINCT, subqueries, and CTEs. Measured release-over-release on the same host, release build, 35-category suite (internal tests, Lite 3.6.1 vs 3.6.0, see the April 2026 performance review).
Scorecard: 33 of 35 categories match or improve vs the previous release. No regressions over 5% beyond two flagged multi-row INSERT fixtures (see blog post for the open caveats). UPDATE with subquery dropped from 392 ms to 315 ms per query — down, but still the single slowest operation in the suite.
Not yet. Graph traversal, BM25 ranking and hybrid search with RRF, MMR or linear fusion as SQL functions are planned for HeliosDB Lite, together with PREPARE COMPILED; HeliosDB Nano ships BM25-ranked full-text search in SQL today, and graph traversal, BM25 indexes and hybrid search as MCP tools in its mcp-endpoint source build. In Lite, HNSW vector search already runs next to ordinary SQL filters in one query.
TF-IDF based keyword scoring as a SQL function, with configurable k1/b parameters.
Combine keyword BM25 scores with vector similarity via Reciprocal Rank Fusion or Maximal Marginal Relevance in one query.
Graph adjacency lists, path traversal and neighborhood queries on relational data as SQL functions.
Lite upgrades Nano's vector search with SIMD acceleration using AVX-512, AVX2, and NEON instruction sets. This delivers up to 4x faster similarity computations compared to scalar operations (internal micro-benchmark, AVX2/AVX-512/NEON vs scalar distance kernel), pushing throughput beyond 5K QPS (internal tests, HeliosDB Lite embedded SQL path, HNSW + PQ; workload-dependent).
-- SIMD auto-detected at startup
-- AVX2/AVX-512/NEON enabled by default
-- Vector search with SQL filters
SELECT title,
embedding <=> $1::vector(768) AS distance
FROM documents
WHERE category = 'support'
ORDER BY distance
LIMIT 20;HeliosDB Lite includes HeliosCore, a custom storage backend built from the ground up in Rust. It bypasses the filesystem cache with O_DIRECT I/O, providing predictable latency and maximum throughput for database workloads.
helios_filter_* system views are present but not yet populated# heliosdb.toml: WAL durability and cache
[storage]
wal_enabled = true
wal_sync_mode = "sync" # sync | async | group_commit
cache_size = 536870912 # bytes (512 MB)
# Lite refuses to start without a REST token-signing secret
$ export HELIOSDB_JWT_SECRET="$(openssl rand -hex 32)"
# Start Lite on a data directory
$ heliosdb-lite start --data-dir /data/helios \
--config ./heliosdb.toml --port 5432 \
--auth scram-sha-256 --password "$PGPASS"
-- Then plain SQL over the PostgreSQL wire
CREATE TABLE events (
id INT PRIMARY KEY,
kind TEXT STORAGE DICTIONARY,
amount FLOAT8 STORAGE COLUMNAR
);
INSERT INTO events VALUES (1, 'click', 0.5);HeliosDB Lite includes foreign key enforcement, MVCC garbage collection, online backups, and 94 system views, including an Oracle-compatible helios_dictionary for migrations from Oracle.
information_schema, pg_catalog equivalents and helios_dictionary (Oracle DICT compatible); the five helios_filter_* views return no rows until Filter Advisor ships.-- Oracle DICT-compatible view
SELECT * FROM helios_dictionary
WHERE table_name = 'ALL_TABLES';
-- System catalog introspection
SELECT table_name, column_name, data_type
FROM information_schema.columns
WHERE table_schema = 'public';
-- FK constraints enforced
CREATE TABLE orders (
id SERIAL PRIMARY KEY,
customer_id INT
REFERENCES customers(id)
ON DELETE CASCADE
);
-- Online backup
BACKUP DATABASE TO
'/backups/lite-20260419.snap';Let AI agents and developers generate DDL from JSON samples. The Schema Generation API analyzes your data, detects types (including vectors), infers constraints, suggests indexes, and returns confidence scores — all via a single REST call.
_agent_hints with readiness flags, suggested indexes, and potential issues for LLM consumption.# Infer schema from JSON samples
POST /api/v1/schema/infer
{
"samples": [
{"id": 1, "name": "Widget",
"price": 29.99,
"embedding": [0.1, 0.2, 0.3]}
],
"table_name": "products"
}
# Response: generated DDL
{
"ddl": "CREATE TABLE products (
id INTEGER NOT NULL,
name VARCHAR(12) NOT NULL,
price NUMERIC NOT NULL,
embedding VECTOR(3) NOT NULL,
PRIMARY KEY (id)
);",
"confidence": 0.95,
"_agent_hints": {
"ready_to_execute": true
}
}Not yet: Supabase/PostgREST-style /rest/v1 table routes are planned for HeliosDB Lite. Today its REST API runs SQL on a branch and manages branches under /v1 (also /api/v1), with JWT or API-key auth. An agent-memory, vector-store and document-ingest REST API for Lite is planned.
POST /v1/branches/:name/query for reads and /execute for writes, with optional parameters and time travel./v1/branches./openapi.json, with /docs and /redoc.# Write on the main branch
POST /v1/branches/main/execute
{"sql": "INSERT INTO users (name, email) VALUES ('Alice', 'alice@example.com')"}
# Read it back
POST /v1/branches/main/query
{"sql": "SELECT id, name FROM users ORDER BY id LIMIT 10"}Reduce infrastructure complexity. Fewer services to run, a simpler security posture and easier compliance (SOC 2, GDPR).
Faster time-to-market. No infrastructure setup meetings. Developers productive in minutes. Built-in features eliminate build-or-buy decisions for vector search, caching, and auth.
Focus on product, not plumbing. One connection string, one query language, one mental model. PostgreSQL compatibility means your existing skills and tools work immediately.
First-class vector support. HNSW with tunable parameters, Product Quantization cuts resident RAM about 16× at recall-safe defaults (internal tests, Nano 3.32.1 vs an unquantized index, see benchmarks.html), hybrid queries combine vectors + SQL filters, and Schema Inference generates DDL from JSON.
HeliosDB Lite is a self-hosted database for production SaaS workloads that PostgreSQL clients reach over the wire. It bundles HeliosProxy for connection pooling and routing with the HeliosCore storage engine, and runs in-process or as a server (deployment modes). It is licensed SSPL-1.0 and distributed to licensed customers.
Contact sales@heliosdb.com. Lite is source-available under SSPL-1.0 and available to licensed customers, who receive the source distribution. It is not published on crates.io, PyPI, npm, Docker Hub or Homebrew, so you build the library, server binary or container image from that distribution (installation guide).
Lite is the licensed production tier; Nano is the Apache 2.0 embedded engine. Lite bundles HeliosProxy for connection pooling and routing, adds the HeliosCore storage engine and runs SQL and PL/pgSQL routines, with PL/SQL, T-SQL and DB2 SQL PL planned. Lite is licensed through sales; Nano is free to self-host.
Lite runs SQL and PL/pgSQL routines; PL/SQL, T-SQL and DB2 SQL PL are planned. Lite also accepts some Oracle-style functions, such as NVL, SYSDATE and TO_CHAR (procedural language guide).
Partly. Lite stores embeddings in HNSW vector indexes that you query with SQL next to relational filters. Graph traversal, BM25 ranking and hybrid search as SQL functions, PREPARE COMPILED, Code-Graph and Graph-RAG (WITH CONTEXT) are planned for Lite; HeliosDB Nano ships Code-Graph and Graph-RAG as build-from-source features and graph, BM25 and hybrid search as MCP tools.
Not yet. A native MCP server for HeliosDB Lite is planned. Today agents reach Lite over the PostgreSQL wire protocol or its REST API, and HeliosDB Nano ships an MCP server as a build-from-source feature.
Lite combines row-level security policies, written with CREATE POLICY, with tenant management at three isolation levels: shared tables with a tenant column, a schema per tenant, or a database per tenant. Per-tenant quotas cap storage, connections and query rate. Policies are enforced for PostgreSQL-wire and REST clients according to the authenticated role; as in PostgreSQL, superusers bypass them, so use password or SCRAM authentication and non-superuser roles.
Lite's high-availability tiers start with a warm standby: Lite builds with the ha-tier1 feature stream the write-ahead log to a warm standby; automatic failover and split-brain protection are planned (switch over manually today). The tiers are enabled at build time. Back up with Lite's dump and restore commands and verify a restore before relying on it.
The benchmarks page publishes a 35-category comparison of Lite against PostgreSQL 16, measured on representative fixtures with Lite embedded and a small dataset that fits in memory (internal tests, a March 2026 HeliosDB Lite build in embedded mode vs PostgreSQL 16 over the network), and it reports the categories PostgreSQL wins. Reproduce the suite on your own workload and hardware before relying on any ratio.
Use HeliosDB Lite for production SaaS workloads that need PostgreSQL clients, connection pooling and per-tenant isolation on infrastructure you run yourself, under a commercial license.
Multi-tenant SaaS with Row-Level Security, connection pooling via HeliosProxy, and SQL and PL/pgSQL routines.
Deploy at the edge with a single compact binary (about 60 MB in internal builds). HeliosCore's O_DIRECT I/O and SIMD vector search deliver consistent performance on edge hardware.
PostgreSQL-compatible SQL, wire protocol support, PL/pgSQL routines and Oracle-style functions such as NVL and TO_CHAR ease migration from PostgreSQL; PL/SQL and T-SQL support is planned.
PostgreSQL wire and REST APIs and SIMD-optimized vector search in SQL; an agent-memory, vector-store and document-ingest REST API for Lite and a native MCP server are planned.
License Lite for SaaS workloads — or pair it with the AI-DBA Agent / AI Support add-ons to keep operations hands-off.