Postgres with vectors, maps, and graphs

pgvector ships enabled: build HNSW indexes over embedding columns and run approximate nearest-neighbor queries in SQL, next to the rest of your data.

CREATE INDEX ON docs
  USING hnsw (embedding vector_cosine_ops);

SELECT id, title
FROM docs
ORDER BY embedding <=> $1
LIMIT 10;

PostGIS ships enabled: geometry and geography types, spatial indexes, and distance/containment queries.

SELECT name
FROM places
WHERE ST_DWithin(
  geog,
  ST_MakePoint($lng, $lat)::geography,
  5000  -- within 5 km
)
ORDER BY geog <-> ST_MakePoint($lng, $lat)::geography;

Model edges as rows and traverse with recursive CTEs — multi-hop reachability, shortest paths, and dependency walks without a separate graph database. openCypher via Apache AGE is on the roadmap.

WITH RECURSIVE reach AS (
  SELECT dst, 1 AS depth FROM edges WHERE src = $start
  UNION
  SELECT e.dst, r.depth + 1
  FROM edges e JOIN reach r ON e.src = r.dst
  WHERE r.depth < 4
)
SELECT DISTINCT dst, min(depth) FROM reach GROUP BY dst;

Combine vector similarity, full-text rank, and recency in one query — or fan out to a gobed addon for GPU-scale candidate generation and rerank in SQL.

SELECT id, title,
  0.6 * (1 - (embedding <=> $emb))
+ 0.4 * ts_rank(fts, plainto_tsquery($q)) AS score
FROM docs
WHERE fts @@ plainto_tsquery($q)
   OR embedding <=> $emb < 0.35
ORDER BY score DESC LIMIT 20;