SynapCores
AI-native database
Last verified:
What is SynapCores?
SynapCores is an AI-native self-hosted database that combines SQL, vector search, graph traversal (Cypher), and embedded LLM inference in a single binary engine. It is designed to replace the fragmented stack of five separate services (Pinecone/Qdrant for vectors, PostgreSQL/MongoDB for relational data, Redis for caching, Neo4j/FalkorDB for graphs, and LangChain/LlamaIndex for orchestration) with one unified system. The platform provides persistent agent memory so AI agents don't forget users across sessions, offering durable, private, and relational memory with built-in vector recall, GraphRAG, semantic tool routing, and grounded generation.
Key features include native SQLv2 with built-in EMBED() and COSINE_SIMILARITY() operators for vector operations without extensions, embedded llama-cpp inference engine running Llama-3.2-1b GGUF locally, HNSW vector indexing, 161 ready-to-run recipes covering use cases from GraphRAG to fraud ring detection, a bundled React web UI on port 8080 with dashboard/query editor/AI chat/vector search/graph explorer, REST + WebSocket API, TypeScript and Python official typed SDKs, and native MCP support for Claude Code/Cursor/VS Code. The Community Edition runs locally with 100% private data, delivers <50ms unified query latency, and requires no signup or credit card.
SynapCores is for developers building AI agents, chatbots, copilots, or GraphRAG applications who need persistent memory without stitching together multiple databases and services. It targets technical teams with active AI agent prototypes in development or early production, support bot builders, internal copilot developers, fraud detection investigators needing graph+ML in one query, and teams building semantic cohorts over real data combining structured and unstructured search.
The platform offers two main tracks: a free Design Partner program for early-stage teams trading feedback for hands-on engineering support, and a Paid Pilot starting at $5,000 for fixed-fee engagements with production-grade private memory and formal SLA-lite support. Enterprise Edition adds Raft clustering, CDC inbound, fine-grained RBAC, SSO, and encryption-at-rest for multi-host production deployments.
SynapCores pricing
Pricing model: Freemium
Community Edition is completely free forever with no signup, no credit card, no waitlist. It includes graph + vector + LLM in one engine, 161 ready-to-run recipes, bundled local LLM (Llama-3.2-1b GGUF), multi-provider vision support, OCR + transcription bundled, and single-host scale. Paid options include: (1) Design Partner program - free track for early-stage teams trading active product feedback and case study for 2-4 weeks of direct integration support, dedicated private Slack/Discord channel, and roadmap influence; (2) Paid Pilot - starts at $5,000 for fixed-fee 2-4 week engagement with production-grade private memory, formal SLA-lite support, deployment review, written success report, complete memory schema design & audit, custom SDK wrapper, performance & latency benchmark report, and 30-day production support tail; (3) Enterprise Edition - includes Raft clustering, CDC inbound, fine-grained RBAC, SSO, and encryption-at-rest, with a free 12-month production license given to Design Partners after pilot transitions to live; (4) Enterprise Support available as paid SLA-backed contract for CE deployments without buying EE.
SynapCores pros
- Single binary replaces five separate services (vector DB, graph DB, relational DB, cache, LLM API)
- 100% local and private - data never leaves your machine in Community Edition
- <50ms unified query latency vs 400ms-1.2s for multi-service stacks
- No glue code needed - vector search, graph traversal, and LLM in one query
- 161 ready-to-run recipes from hello-world to advanced fraud detection
- Embedded LLM runs Llama-3.2-1b GGUF in-process with no external API calls
- Native SQLv2 with built-in EMBED() and COSINE_SIMILARITY() operators
- No extensions required - vector search built-in vs PostgreSQL's pgvector
- Official typed SDKs for TypeScript and Python
- Bundled React web UI on port 8080 with dashboard, query editor, AI chat, graph explorer
- Native MCP support for Claude Code, Cursor, and VS Code integration
- Inspectable retrieval traces showing exactly which memory agent used
- Cypher graph traversal + vector similarity in same SQL statement
- AutoML training and inference in one SQL statement without Python pipeline
- Free Community Edition with no signup, no credit card, no waitlist
- OCR (Tesseract) and transcription (Whisper) bundled for multimedia processing
- 3-5x faster than PostgreSQL + pgvector + external embeddings for vector workloads
SynapCores cons
- Early stage company with no customer logos yet
- Multi-host clustering only available in paid Enterprise Edition
- Fine-grained RBAC, SSO, and encryption-at-rest gated to Enterprise
- Designed for single-host scale (one VM/laptop) in Community Edition
- Source code is private - only public binary releases available
- Native package managers (Homebrew, .deb/.rpm, Windows MSI) still on roadmap
- Windows requires WSL2 (Ubuntu) terminal for installation
- GPU acceleration requires source build for CUDA-enabled hosts
- Limited to 3 Design Partner slots available
Frequently asked questions about SynapCores
What is SynapCores?
SynapCores is an AI-native self-hosted database that combines SQL, vector search, ML, and an embedded LLM inference engine in one single binary. It replaces five separate tools (vector DB like Pinecone/Qdrant, relational DB like PostgreSQL, cache like Redis, graph DB like Neo4j, and orchestration like LangChain) with one unified engine that provides persistent agent memory, GraphRAG, fraud ring detection, and semantic cohorts.
How do I install SynapCores?
Install in one line: curl -fsSL https://get.synapcores.com/install.sh | sh. This auto-detects your OS and architecture (Linux x86_64/aarch64, Ubuntu 24.04 variants, macOS Apple Silicon) and GPU (CUDA-enabled hosts get CUDA build automatically). On Windows, run inside WSL2 (Ubuntu) terminal. Alternatively, run locally in 1 second with docker run -d -p 8080:8080 ghcr.io/synapcores/community. The installer prints first-boot admin credentials and the URL of the bundled web UI (typically http://localhost:8080).
What's the difference between Community Edition and Enterprise Edition?
Community Edition is free forever and includes graph + vector + LLM in one engine, 161 ready-to-run recipes, bundled local LLM, multi-provider vision, OCR + transcription, and single-host scale. Enterprise Edition adds Raft clustering for multi-host deployments, CDC inbound, fine-grained RBAC, SSO, and encryption-at-rest. CE is designed for one VM/laptop while EE supports production clustering.
How does SynapCores handle agent memory?
SynapCores provides durable, private, relational memory through a Memory Loop that combines four steps: (1) Recall Memory via vector search for past user context, (2) Retrieve Knowledge via GraphRAG traversing knowledge graph relationships, (3) Semantic Tool Routing deciding which tool to trigger based on semantic similarity, (4) Generate Reply bundling all context for grounded generation, and (5) Persist Turn writing the transaction back to graph memory. Agents hold context across sessions without stitching together vector DB, graph DB, cache, and LLM API.
What recipes are available?
161 ready-to-run recipes ship inside the binary, searchable in the bundled web UI. Examples include: 020_agentic_memory for persistent agent memory, 016_graphrag_qna for GraphRAG without three databases, graph/fraud-ring patterns for fraud ring detection, recipes/multimedia for semantic + structured search, Hello GraphRAG (5 min, advanced), Fraud Ring Detection (7 min, intermediate), Customer Churn ML (5 min, beginner), Knowledge Graph from Earnings Call (6 min, beginner), Drug Repurposing (10 min, advanced), and Semantic Patient Cohorts (6 min, beginner).
How does SynapCores compare to PostgreSQL + pgvector?
PostgreSQL needs pgvector extension, separate embedding service (OpenAI API or local models), TimescaleDB for time series, custom ML pipeline, and complex orchestration. SynapCores has native vector operations with built-in EMBED() and COSINE_SIMILARITY() operators, integrated ML with PREDICT() as window function, true semantic understanding (knows 'make faster' = 'performance'), native time series, AutoML with CREATE EXPERIMENT, multimodal data support (IMAGE, AUDIO, VIDEO), and native OCR/transcription. Performance is 3-5x faster for vector workloads due to no network round-trips to external services, native HNSW vector indexes, and better cache utilization from single query plan.
What programming languages are supported?
Official typed SDKs are available for TypeScript and Python. The platform also provides REST + WebSocket API for integration with any language. The embedded React web UI allows direct SQL/Cypher query execution without needing a specific programming language.
Can I use external LLM providers?
Yes. While Community Edition includes a bundled local LLM (Llama-3.2-1b GGUF) that runs in-process with no external API calls, you can configure OpenAI, Anthropic Claude, or Ollama providers from the Settings UI. Multi-provider vision allows captioning images using OpenAI, Anthropic Claude, or Ollama LLaVA. External providers are optional until you bring your own API key.
What use cases does SynapCores support?
Primary use cases include: (1) AI agent that doesn't forget - support bot, internal copilot, or developer agent holding context across sessions with native MCP support and inspectable retrieval traces; (2) GraphRAG without three databases - Cypher graph traversal + vector similarity in same SQL statement with citeable paths; (3) Fraud ring detection - graph + ML in one query finding account clusters sharing devices/IPs/cards with AutoML or LLM scoring; (4) Semantic cohorts over real data - finding accounts/tickets/documents/contracts that 'look like' known examples across structured columns AND unstructured text using EMBED() and COSINE_SIMILARITY() as first-class SQL operators.
How do I get help or support?
Support options include: (1) Design Partner program with private Slack/Discord channel for sub-hour responses to bugs, performance questions, or deployment blockers; (2) Paid Pilot with 30-day production support tail and formal SLA-lite support; (3) Enterprise Support as paid SLA-backed contract for CE deployments; (4) Issue tracker and security disclosure on the documentation site; (5) 5-email course with one runnable SQL recipe per email plus new recipe every Monday; (6) Direct access to SynapCores' core database engineering team for Design Partners. The documentation includes Quickstart, Hardware requirements, First-run AI setup, Web UI guide, and Community vs Enterprise comparison.