Add MSSQL support with vector collection and filtering

.github/workflows/ci.yml

@@ -0,0 +1,24 @@
+name: CI
+
+on:
+  push:
+    branches:
+      - master
+      - main
+  pull_request:
+
+jobs:
+  test:
+    runs-on: ubuntu-latest
+
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v4
+
+      - name: Setup Go
+        uses: actions/setup-go@v5
+        with:
+          go-version-file: go.mod
+
+      - name: Run unit tests
+        run: go test ./...

.github/workflows/release.yml

@@ -0,0 +1,79 @@
+name: Release
+
+on:
+  push:
+    tags:
+      - "v*"
+  workflow_dispatch:
+    inputs:
+      version:
+        description: "Semver version (e.g. 0.2.0 or v0.2.0)"
+        required: true
+        type: string
+
+permissions:
+  contents: write
+
+jobs:
+  release:
+    runs-on: ubuntu-latest
+
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+
+      - name: Setup Go
+        uses: actions/setup-go@v5
+        with:
+          go-version-file: go.mod
+
+      - name: Run unit tests
+        run: go test ./...
+
+      - name: Resolve release tag
+        id: tag
+        shell: bash
+        run: |
+          set -euo pipefail
+
+          if [[ "${GITHUB_EVENT_NAME}" == "workflow_dispatch" ]]; then
+            raw="${{ inputs.version }}"
+            raw="${raw#v}"
+            if [[ ! "$raw" =~ ^[0-9]+\.[0-9]+\.[0-9]+(-[0-9A-Za-z.-]+)?(\+[0-9A-Za-z.-]+)?$ ]]; then
+              echo "Invalid version: $raw"
+              exit 1
+            fi
+            tag="v${raw}"
+          else
+            tag="${GITHUB_REF_NAME}"
+          fi
+
+          echo "tag_name=$tag" >> "$GITHUB_OUTPUT"
+
+      - name: Create git tag (manual release)
+        if: github.event_name == 'workflow_dispatch'
+        shell: bash
+        run: |
+          set -euo pipefail
+
+          tag="${{ steps.tag.outputs.tag_name }}"
+
+          if git rev-parse "$tag" >/dev/null 2>&1; then
+            echo "Tag $tag already exists locally"
+          else
+            git tag "$tag"
+          fi
+
+          if git ls-remote --tags origin "refs/tags/$tag" | grep -q "$tag"; then
+            echo "Tag $tag already exists on origin"
+          else
+            git push origin "$tag"
+          fi
+
+      - name: Publish GitHub release
+        uses: softprops/action-gh-release@v2
+        with:
+          tag_name: ${{ steps.tag.outputs.tag_name }}
+          generate_release_notes: true

README.md

@@ -9,7 +9,8 @@ A lightweight Go vector-store library inspired by `Microsoft.Extensions.VectorDa
 
 MVP scope:
 - Go 1.24.x
-- Postgres + `pgvector` only
+- Postgres + `pgvector`
+- MSSQL connector
 - Record-based core API with optional typed codec wrapper
 
 This library can be used to build retrieval systems such as:
@@ -21,12 +22,14 @@ This library can be used to build retrieval systems such as:
 
 - Go 1.24.x
 - PostgreSQL 16+ with `pgvector` extension
+- SQL Server 2022+ (for `stores/mssql`)
 - Docker (optional, for integration tests and sample compose flows)
 
 ## Project layout
 
 - `vectordata`: backend-agnostic core interfaces, record model, filters, typed wrapper
 - `stores/postgres`: Postgres implementation with `pgxpool`
+- `stores/mssql`: SQL Server implementation with `database/sql`
 - `samples`: runnable demos (see `samples/README.md`)
 - `docs`: architecture and implementation notes
 
@@ -193,19 +196,46 @@ go test ./...
 ```
 
 Notes:
-- Integration tests start Postgres + pgvector automatically via Testcontainers
+- Integration tests start Postgres/pgvector and SQL Server automatically via Testcontainers
 - Docker daemon must be available when running integration tests
 - Optional override: set `PGVECTOR_TEST_DSN` to use an existing Postgres instance instead of starting a container
+- Optional override: set `MSSQL_TEST_DSN` to use an existing SQL Server instance instead of starting a container
+
+Run MSSQL integration tests against root compose service:
+
+```bash
+docker compose up -d mssql
+MSSQL_TEST_DSN="sqlserver://sa:YourStrong%21Passw0rd@localhost:14339?database=master&encrypt=disable" \
+  go test -tags=integration ./stores/mssql
+```
 
 ## Docker Compose (optional)
 
 `docker-compose.yml` at the repository root is kept for manual local runs.
 
 - Use root `docker-compose.yml` when you want a persistent local Postgres+pgvector instance outside tests
+- Root compose also includes SQL Server (`mssql`) for local MSSQL connector validation
 - Use Testcontainers (`go test -tags=integration ./...`) for integration tests
 - Sample apps have their own compose files at `samples/semantic-search/docker-compose.yml` and `samples/ragrimosa/docker-compose.yml`
 - Sample Dockerfiles use `golang:1.24-alpine` to match the repo Go version (`1.24.x`)
 
+## Release Automation
+
+GitHub Actions workflows are configured for:
+
+- CI on pushes/PRs (`.github/workflows/ci.yml`)
+- release publishing (`.github/workflows/release.yml`)
+
+Release options:
+
+1. Manual (recommended): run `Release` workflow via GitHub UI with `version` input (`0.2.0` or `v0.2.0`).
+2. Tag-driven: push a semver tag and the workflow publishes release notes automatically:
+
+```bash
+git tag v0.2.0
+git push origin v0.2.0
+```
+
 ## Samples
 
 - Overview and entry points: [`samples/README.md`](samples/README.md)

docker-compose.yml

@@ -13,3 +13,13 @@ services:
       interval: 2s
       timeout: 2s
       retries: 30
+
+  mssql:
+    image: mcr.microsoft.com/mssql/server:2022-latest
+    container_name: go-vectorstore-mssql
+    environment:
+      ACCEPT_EULA: "Y"
+      MSSQL_PID: "Developer"
+      MSSQL_SA_PASSWORD: "YourStrong!Passw0rd"
+    ports:
+      - "14339:1433"

docs/architecture.md

@@ -8,6 +8,7 @@ The project is split into two layers:
 
 - `vectordata`: backend-agnostic contracts and primitives
 - `stores/postgres`: PostgreSQL + pgvector implementation
+- `stores/mssql`: SQL Server implementation
 
 This keeps the public API stable while allowing additional storage engines later.
 
@@ -23,7 +24,7 @@ The base record type is `vectordata.Record`:
 Search returns `vectordata.SearchResult`:
 
 - `Record`: matched item
-- `Distance`: raw pgvector distance
+- `Distance`: backend-computed distance value
 - `Score`: normalized score derived from distance
 
 Score normalization:
@@ -57,6 +58,11 @@ All methods require `context.Context`.
 - `EnsureExtension`: `true`
 - `StrictByDefault`: `true`
 
+`mssql.StoreOptions` defaults (`mssql.DefaultStoreOptions()`):
+
+- `Schema`: `dbo`
+- `StrictByDefault`: `true`
+
 Collection defaults:
 
 - Metric defaults to cosine when omitted
@@ -71,7 +77,7 @@ Other operational defaults:
 
 ## 5) Request Flow
 
-### Ensure collection
+### Postgres: Ensure collection
 
 `PostgresVectorStore.EnsureCollection`:
 
@@ -90,7 +96,7 @@ metadata jsonb not null default '{}'::jsonb,
 content text
 ```
 
-### Insert / Upsert
+### Postgres: Insert / Upsert
 
 Bulk writes are chunked and executed as parameterized SQL:
 
@@ -103,7 +109,7 @@ Each record is validated before sending:
 - vector dimension match
 - metadata JSON serialization
 
-### Search
+### Postgres: Search
 
 `SearchByVector` builds a query plan, then executes it:
 
@@ -121,6 +127,22 @@ Each record is validated before sending:
 
 The pgvector operators are documented in the [pgvector README](https://github.com/pgvector/pgvector#querying).
 
+### MSSQL: Ensure / Search
+
+`MSSQLVectorStore.EnsureCollection`:
+
+1. Ensures target schema exists
+2. Ensures internal collection metadata table exists
+3. Creates or validates the collection table
+4. Persists and validates dimension/metric metadata
+
+`MSSQLCollection.SearchByVector`:
+
+1. Loads records from SQL Server
+2. Evaluates filters against records in-process
+3. Computes distance in-process (cosine/l2/inner product)
+4. Applies threshold, sorts by distance, returns top-k
+
 ## 6) Filter System (AST -> SQL)
 
 Filters are represented as an AST in `vectordata`:
@@ -147,6 +169,8 @@ Behavior details:
 
 JSON path extraction behavior comes from PostgreSQL [JSON/JSONB functions and operators](https://www.postgresql.org/docs/current/functions-json.html).
 
+For MSSQL in this MVP, the same AST is evaluated in-process against loaded records.
+
 ## 7) Schema Safety Modes
 
 `CollectionSpec.Mode` controls ensure behavior:
@@ -238,7 +262,8 @@ Errors are wrapped with context so callers can use `errors.Is(...)` against base
 
 Current MVP scope:
 
-- Postgres + pgvector only
+- Postgres + pgvector backend
+- MSSQL backend (vectors stored as JSON payloads)
 - single-vector column per collection
 - metadata filtering through a focused AST
 
@@ -247,8 +272,10 @@ Future extension points:
 - additional backends under `stores/`
 - richer filter operators
 - optional reranking strategies
+- native MSSQL vector indexing/query pushdown
 
 ## 13) References
 
 - pgvector project docs: https://github.com/pgvector/pgvector
 - PostgreSQL docs: https://www.postgresql.org/docs/current/index.html
+- SQL Server docs: https://learn.microsoft.com/sql/sql-server/

go.mod

@@ -4,7 +4,11 @@ go 1.24.0
 
 toolchain go1.24.9
 
-require github.com/jackc/pgx/v5 v5.7.2
+require (
+	github.com/jackc/pgx/v5 v5.7.2
+	github.com/microsoft/go-mssqldb v1.8.0
+	github.com/testcontainers/testcontainers-go v0.33.0
+)
 
 require (
 	dario.cat/mergo v1.0.0 // indirect
@@ -45,7 +49,6 @@ require (
 	github.com/shirou/gopsutil/v3 v3.23.12 // indirect
 	github.com/shoenig/go-m1cpu v0.1.6 // indirect
 	github.com/sirupsen/logrus v1.9.3 // indirect
-	github.com/testcontainers/testcontainers-go v0.33.0 // indirect
 	github.com/tklauser/go-sysconf v0.3.12 // indirect
 	github.com/tklauser/numcpus v0.6.1 // indirect
 	github.com/yusufpapurcu/wmi v1.2.3 // indirect

go.sum

@@ -36,6 +36,10 @@ github.com/go-ole/go-ole v1.2.6 h1:/Fpf6oFPoeFik9ty7siob0G6Ke8QvQEuVcuChpwXzpY=
 github.com/go-ole/go-ole v1.2.6/go.mod h1:pprOEPIfldk/42T2oK7lQ4v4JSDwmV0As9GaiUsvbm0=
 github.com/gogo/protobuf v1.3.2 h1:Ov1cvc58UF3b5XjBnZv7+opcTcQFZebYjWzi34vdm4Q=
 github.com/gogo/protobuf v1.3.2/go.mod h1:P1XiOD3dCwIKUDQYPy72D8LYyHL2YPYrpS2s69NZV8Q=
+github.com/golang-sql/civil v0.0.0-20220223132316-b832511892a9 h1:C/Fq8+Vm3x8+3xB3SO2/3A8Q+7WmfYI8MViMbum/v8A=
+github.com/golang-sql/civil v0.0.0-20220223132316-b832511892a9/go.mod h1:9x5EikdFyA8Kj4m2Vj93QShv2N6Yf4H6J5D9Q1QzM4Y=
+github.com/golang-sql/sqlexp v0.1.0 h1:DKM0nqnvWFfB6K6x2zI2Qf8F9Pwc7P0W8o2JQ7QY9MM=
+github.com/golang-sql/sqlexp v0.1.0/go.mod h1:hxXw1K3Zk7K8Y6nG5N7f6d4rV1E8j9a7q6x5y4z3w2Q=
 github.com/google/go-cmp v0.5.6/go.mod h1:v8dTdLbMG2kIc/vJvl+f65V22dbkXbowE6jgT/gNBxE=
 github.com/google/go-cmp v0.5.9/go.mod h1:17dUlkBOakJ0+DkrSSNjCkIjxS6bF9zb3elmeNGIjoY=
 github.com/google/go-cmp v0.6.0/go.mod h1:17dUlkBOakJ0+DkrSSNjCkIjxS6bF9zb3elmeNGIjoY=
@@ -57,6 +61,8 @@ github.com/lufia/plan9stats v0.0.0-20211012122336-39d0f177ccd0 h1:6E+4a0GO5zZEnZ
 github.com/lufia/plan9stats v0.0.0-20211012122336-39d0f177ccd0/go.mod h1:zJYVVT2jmtg6P3p1VtQj7WsuWi/y4VnjVBn7F8KPB3I=
 github.com/magiconair/properties v1.8.7 h1:IeQXZAiQcpL9mgcAe1Nu6cX9LLw6ExEHKjN0VQdvPDY=
 github.com/magiconair/properties v1.8.7/go.mod h1:Dhd985XPs7jluiymwWYZ0G4Z61jb3vdS329zhj2hYo0=
+github.com/microsoft/go-mssqldb v1.8.0 h1:7cyZ/AT7ycDsEoWPIXibd+aVKFtteUNhDGf3aobP+tw=
+github.com/microsoft/go-mssqldb v1.8.0/go.mod h1:6znkekS3T2vp0waiMhen4GPU1BiAsrP+iXHcE7a7rFo=
 github.com/moby/docker-image-spec v1.3.1 h1:jMKff3w6PgbfSa69GfNg+zN/XLhfXJGnEx3Nl2EsFP0=
 github.com/moby/docker-image-spec v1.3.1/go.mod h1:eKmb5VW8vQEh/BAr2yvVNvuiJuY6UIocYsFu/DxxRpo=
 github.com/moby/patternmatcher v0.6.0 h1:GmP9lR19aU5GqSSFko+5pRqHi+Ohk1O69aFiKkVGiPk=