Unified data infrastructure
GitHub Let's talk

Future data models — VECTOR and GRAPH

Explicitly not in v0.1.0. What exists today, what does not, and how to prepare.

Version
Latest
v0.1.0 · latest 2 min read
On this page
  1. VECTOR (not available)
  2. GRAPH (not available)
  3. BLOBS / multi-region / distributed
Warning

NOT AVAILABLE in v0.1.0. Roadmap firewall: a mention here is not a promise of syntax, timeline, or compatibility. This page exists so CTOs/architects are never misled. Do not build on it.

diagram
flowchart TB
    subgraph In["v0.1.0 — documented, supported"]
        R["relational"] --- J["JSON documents"] --- T["time-series"]
    end
    subgraph Out["future — not in v0.1.0"]
        V["vectors"] --- G["graphs"] --- B["blobs / distributed"]
    end
    In -.->|"may grow in v0.2.0+ as versioned sections"| Out
Diagram source · mermaidcopy included
mermaidsource
flowchart TB
    subgraph In["v0.1.0 — documented, supported"]
        R["relational"] --- J["JSON documents"] --- T["time-series"]
    end
    subgraph Out["future — not in v0.1.0"]
        V["vectors"] --- G["graphs"] --- B["blobs / distributed"]
    end
    In -.->|"may grow in v0.2.0+ as versioned sections"| Out

VECTOR (not available)#

Question Answer
Vector column type? No. No vector(n), no halfvec, no sparse type.
Distance operators (<->, <#>, <=>)? No. query/vector.rs is columnar aggregate planning, not pgvector.
ANN/HNSW/IVF indexes? No.
What you CAN do today Store embeddings as JSONB arrays or FLOAT8[]; fetch candidates with relational filters; compute distance client-side or with scalar math per row (no index acceleration).
sqlsource
-- TODAY (no vector index; full scan + client-side ranking):
CREATE TABLE docs (id BIGINT PRIMARY KEY, emb JSONB);
SELECT id FROM docs WHERE (emb->>'dim')::int = 1536 LIMIT 100;

GRAPH (not available)#

Question Answer
Node/edge tables, MATCH, traversals? No. No graph DDL, no path queries, no traversal executor.
What you CAN do today Model adjacency relationally (edges(src, dst) + recursive CTEs for bounded hops).
sqlsource
-- TODAY: bounded hops with RECURSIVE (relational, unindexed-traversal):
WITH RECURSIVE hops(n, d) AS (
  SELECT 1, 0 UNION SELECT e.dst, h.d+1 FROM edges e JOIN hops h ON e.src = h.n WHERE h.d < 3
) SELECT * FROM hops;

BLOBS / multi-region / distributed#

Also not in v0.1.0: no blob type, no multi-region, no distributed deployment. Single-node server + WAL + checkpoints only.

When these land: they will appear as versioned sections (v0.2.0/...) with ADDED/CHANGED tables. v0.1.0 pages will not be rewritten.

Was this page helpful?
Next

Continue

Next: Executor — what it can execute Complete dispatch map — every statement shape, which engine path, what it returns. Playground Run the documented statements against the sample dataset. Architecture The layers underneath the documented surface.