Hi HN — I'm Luis, I built SynapCores. It's an AI-native database in a single self-hosted binary: vector search (HNSW), a property-graph engine (Cypher), SQL, in-database AutoML, and a bundled local LLM.
Community Edition is free, no signup, no telemetry — it installs in about 30 seconds: curl -fsSL https://get.synapcores.com | sh
(or: docker run synapcores/community)
Why I built it: most "AI apps" become a 5-service stack — Postgres + a vector DB + a graph DB + a model-serving layer + glue code to keep them in sync. I wanted one engine where you embed, search, train, and
query across all of it without the pipeline. So you can write SELECT ... COSINE_SIMILARITY(embedding, EMBED('a quiet pair of headphones')), or CREATE EXPERIMENT ... WITH (task_type='binary_classification')
then PREDICT ... USING model — all server-side, in SQL.
Honest tradeoffs: CE is single-host (clustering and RBAC are in EE); the bundled model is a small Llama for convenience — bring your own keys for bigger ones; and it's young, so I'd genuinely value hearing
where it breaks for you. There are ~150 runnable recipes and the docs at synapcores.com if you want to kick the tires.
Happy to answer anything.