---
title: Weave (vector store)
description: Use fabriq as the vector backend for weave's RAG pipeline via the fabriqvec adapter — weave's document chunks persist into the fabric.
---

[Weave](https://github.com/xraph/weave) is a composable RAG pipeline (load → chunk → embed → store → retrieve → assemble) with a pluggable `vectorstore.VectorStore` port. The **`fabriqvec`** adapter (`github.com/xraph/weave/vectorstore/fabriq`, lives in the weave repo) implements that port over fabriq's [Vector](/docs/fabriq/(data-planes)/vector) plane, so weave's chunk embeddings live in the fabric instead of a parallel store.

This is why fabriq's `VectorQuerier` gained metadata [filtering](/docs/fabriq/(data-planes)/vector#metadata-filtering) and `DeleteByMeta` — to back weave's `Search` filters and `DeleteByMetadata` faithfully.

## Wiring

`fabriqvec.New` takes a fabriq `query.VectorQuerier` (from `f.Vector()`) and namespaces all weave vectors under one fabriq entity (default `weave_vector`).

```go
import (
	"github.com/xraph/weave/engine"
	fabriqvec "github.com/xraph/weave/vectorstore/fabriq"
)

eng, err := engine.New(
	engine.WithStore(metaStore),
	engine.WithVectorStore(fabriqvec.New(f.Vector(), fabriqvec.WithEntity("weave_vector"))),
	// ... or fabriqvec.EngineOption(container) to auto-discover the fabriq facade from DI.
)
```

The host must ensure the `fabriq_embeddings` table exists (fabriq [migrations](/docs/fabriq/(operations)/migrations)) and operate within a fabriq tenant scope. The namespacing entity need not be registered in fabriq's [registry](/docs/fabriq/(concepts)/registry) — `fabriq_embeddings` is keyed by an arbitrary `(tenant, entity, id)` triple, so the vector path works without it.

## How it maps

| weave `VectorStore` | fabriq `VectorQuerier` |
|---|---|
| `Upsert(entries)` | `Upsert(entity, id, vector, meta)` — weave `Content` is folded into `meta["content"]` (fabriq has no content column). |
| `Search(vec, opts)` | `Similar(VectorQuery{Filter: opts.Filter})` — `opts.TenantKey` maps to fabriq's tenant scope; `MinScore` post-filters; `Content`/`Metadata` are reconstructed from `meta`. |
| `Delete(ids)` | `Delete(entity, id)` per id. |
| `DeleteByMetadata(filter)` | `DeleteByMeta(entity, filter)`. |

<Callout type="info">
Two deliberate, documented differences from weave's `pgvector` store: the fabriq-backed `Search` returns a **nil** `Vector` on each hit (fabriq does not return embeddings on search — consumers use content + score + metadata), and `TenantKey` maps to fabriq's **native** per-tenant isolation rather than a `metadata.tenant_id` convention — stronger isolation.
</Callout>

<Cards>
  <Card title="Vector" href="/docs/fabriq/(data-planes)/vector">The fabriq embedding port the adapter is built on — including metadata filtering and DeleteByMeta.</Card>
  <Card title="Tenancy" href="/docs/fabriq/(concepts)/tenancy">The structural tenant isolation TenantKey maps onto.</Card>
  <Card title="Cortex brain" href="/docs/fabriq/(ecosystem)/cortex">The other ecosystem integration — fabriq as a brain for cortex agents.</Card>
</Cards>