Vector — API Reference

The vector data engine exposes one control-plane route — the managed search dispatcher — and otherwise routes you to the tabled data plane, where vector<n> columns are read and written over the wire your client already speaks.

There is no VectorService. dodil.data.vector.v1 was removed and the whole /:bucket/vector/… route tree went with it. Collection creation is now a Pipelines concern (PipelineService, vector facet); vector data is dodil.tables.v1.TablesVector on the plane; only the managed search route stays here.

In this reference

Page	Surface	What it covers
Search	`POST /:bucket/search/vector` (HTTP-only)	The Scriptum-template search dispatcher for pipeline-created collections. Text + file queries; RRF across collection groups.

The data path — `TablesVector` on the plane

KNN and every vector write run on the tabled data plane, not the control plane:

SQL / Postgres — pgvector operators (<-> <#> <=>) on any vector(n) column over pg.uk-lon-1.dodil.io:5432 (sslmode=require — the pg wire terminates TLS; never prefer, which downgrades to plaintext silently).
Qdrant / Pinecone SDKs — KNN + writes at qdrant.uk-lon-1.dodil.io / pinecone.uk-lon-1.dodil.io. The api-key header carries the db id (the bucket name), not a credential.
Typed gRPC / REST — dodil.tables.v1.TablesVector (UpsertVectors, QueryVectors, FetchVectors, DeleteVectors, DescribeVectorStats), REST at /v1/vector/:table/{upsert,query,fetch,delete,stats} on the tables door (table.uk-lon-1.dodil.io).
GraphQL — {table}_similar(vector topK metric) fields at gql.uk-lon-1.dodil.io/graphql.

Full contract for the data path lives in Tables (SQL) → API Reference and Connect & wire adapters. The pre-embedded KNN examples are on the Search page.

Collection + template lifecycle live in Pipelines

Creating, listing and deleting an auto-embed collection — and the embedding template catalog — moved to the control plane’s Pipelines section:

Want	Now at
Create / list / get / delete a collection	Pipelines → Vector Collections
The embedding template catalog	Pipelines → Templates
The generic `PipelineService` surface	Pipelines → API Reference

What was retired

Gone	Replacement
`dodil.data.vector.v1.VectorService` (whole service)	`PipelineService` (vector facet) for lifecycle; `TablesVector` for data
`AddVectorPipeline` / `ListCollections` / `GetCollection` / `DeleteCollection`	`CreateVectorPipeline` / `ListPipelines?facet=vector` / `GetPipeline` / `DeletePipeline` — Pipelines
`AddVectorCollection` (manual / `EXTERNAL` mode)	Removed. BYO embeddings are a table with a `vector(n)` column on the plane.
The `Search` RPC	`POST /:bucket/search/vector` (HTTP-only; response type is `dodil.data.common.v1.SearchResponse`)
`InsertVectors`	`dodil.tables.v1.TablesVector.UpsertVectors` on the data plane
The Engine RPCs (`ConfigureEngine` / `GetEngine` / …)	Nothing — vector is implicit per bucket. Capacity is governed by reservations.

Endpoint roots

Endpoint	When to use
`https://api.data.dodil.io`	HTTP — control plane (the search route)
`https://table.uk-lon-1.dodil.io`	HTTP — tabled data plane (`/v1/vector/:table/…`)
`pg.uk-lon-1.dodil.io:5432`	Postgres wire — SQL KNN (`sslmode=require`)
`qdrant.uk-lon-1.dodil.io` / `pinecone.uk-lon-1.dodil.io`	Qdrant / Pinecone wire adapters

Auth: bearer JWT in Authorization on the control plane; the db id in api-key on the Qdrant/Pinecone wires. See Conventions.