New in v1.9 — CREATE AGENT · durable in-database agents

CREATE AGENT —
live in minutes.

Durable AI agents that fire on a schedule and on data changes, remember across restarts, and log every action to a tamper-evident audit trail — all declared with CREATE AGENT.One binary. No GPU. Running in 5 minutes.

Reference agent on GitHub6 on GitHubv1.8 ships native LLM inference — no daemon187 production-ready recipes

Fork the reference agent on GitHub·github.com/SynapCores/synapcores-agent

incident_triage.sql
-- A durable agent, declared in SQL.
CREATE AGENT incident_triage
  PERSONA 'aidb-assistant'
  TASK 'Triage it: classify, name the attack
        pattern, recommend an escalation tier.'
  ON INSERT INTO incidents WHERE severity = 'critical'
  WITH (max_iterations = 3, allow_writes = FALSE,
        budget_tokens_per_day = 200000);

-- INSERT a critical row → the agent wakes
-- and acts, off the write path. Every run
-- lands in the immutable audit log.

The agent lives in the database. No orchestrator, no queue, no worker fleet to run.

★ 5 LIVE DEMOS187 READY-TO-RUN RECIPES★ NATIVE MCP★ OPENCLAW MEMORY★ MACOS + LINUX + DOCKER★ FREE COMMUNITY EDITION

Reference agent on GitHub·Self-hosted — your data never leaves your box·No lock-in — single binary·Become a design partner and shape the roadmap →

The problem

Agents are easy to demo. Hard to actually run.

To put one real agent into production you assemble a stack — a framework, a queue, a worker fleet, a separate vector store, and an audit tool bolted on after. Five systems to run, secure, and keep in sync. Your data leaves the box. And when the agent acts, nothing is provable.

The stack you operate

Agent framework

LangGraph / CrewAI

Queue

Temporal / Celery

Worker fleet

always-on, scaled, paged

Vector store

a second database

Audit tooling

bolted on after

Five things to run, secure, and page someone about at 3am.

What SynapCores runs

The database.

One self-hosted binary. The agent, its memory, its schedule, and its audit trail all live inside it.

The shift

The database is the agent runtime.

The one sentence a bolt-on stack can't repeat

SynapCores is the only self-hosted, single-binary, no-GPU database where durable AI agents live inside the engine — declared with CREATE AGENT, fired by schedule and data changes, with memory, governance, and an immutable audit trail built in.

PROOF 01

Native schema-object agents

Agents are DDL objects that fire on INSERT / UPDATE / DELETE and on a schedule. No orchestrator, queue, or worker fleet to run or secure.

PROOF 02

Immutable audit, built in

Every agent action lands in a tamper-evident ledger — because the runtime is the system of record. Provable compliance, not a bolted-on log.

PROOF 03

One binary. No GPU. Your box.

Native in-process inference means the whole agentic stack runs on a cheap VPS with no model server — and your data never leaves.

See it fire

INSERT a row. Watch the agent act.

Three statements, one engine. No service to deploy, no queue to drain, no worker to page.

Watch a real agent run end-to-end →

The same loop, recorded live against the engine — no mockups.

One query. Three systems other databases need.

SynapCores unifies graph traversal, vector similarity, and LLM inference into a single execution engine.

Graph traversal, HNSW vector similarity, and in-database ML aren't three services here — they're one execution engine. Each answers in a single statement, from microseconds for graph to a couple of milliseconds with embedding or model inference in the loop.

On Postgres that's pgvector + Apache AGE + a model server + your application code stitching them together. On SynapCores it's one query — and you can compose all three in a single MATCH.

Stack Comparison

StackSystemsRound-trips
Pinecone + Postgres + OpenAI rerank3 + your service4–5
Neo4j + external embedding + LLM3 + your service3–4
SynapCores11

Build agent memory in 5 emails.

One runnable SQL recipe per email. No LangChain. Free.

5-email course · free

Ship an agent with persistent memory in 5 emails

A 5-email course where you build a real agent with persistent memory — one runnable SQL recipe per email, no LangChain, no Pinecone, no Redis. Lesson 1 hits your inbox the second you sign up. After that: a new recipe every Monday. No spam.

Free forever. We'll never share your email.

What's in Community Edition

Community Edition is a complete AI-native engine. Sales is an upgrade path, not a front door.

FeatureCommunityEnterprise
Core SQL Engine
Vector Search (HNSW)
Graph Database (Cypher)
AI / LLM Integration
MCP Server (Model Context Protocol)
Multimedia (PDF/AV)
AutoML
Multi-node Clustering
Raft Replication
Fine-grained RBAC
Audit Logging (Scale)
SSO / SAML / LDAP
Immutable Tables

Where we're going: v1.6 (Q3 2026) targets binary wire protocol, shared buffer pool, and B-tree indexes for OLTP at PG scale.

Building production agents? Partner with us early.

We're taking on a small number of design partners shaping the SynapCores roadmap. You get direct access to the engineering team, priority on the features you need, and SLA-backed support. We get a real workload to build against.

Become a design partner →

Need clustering, RBAC, or production support?

The Enterprise Edition (EE) ships everything in CE plus mission-critical scale and security. Paid Enterprise Support is also available for CE deployments.

Scale

Multi-node clustering + Raft replication + CDC inbound from MySQL/Postgres binlog.

Security & Compliance

Fine-grained RBAC, SSO/SAML, LDAP, encryption-at-rest, audit logging, and immutable tables.

Performance

Binary wire protocol, shared buffer pool, row-store fast path (v1.6+).

Enterprise Support (also available for CE)

SLA-backed support for production CE deployments: prioritized bug fixes, custom feature work, roadmap input, and direct access to the engineering team. Buy support without buying EE.