embedflow 0.1.1__tar.gz → 0.2.0__tar.gz

{embedflow-0.1.1 → embedflow-0.2.0}/CHANGELOG.md

@@ -1,5 +1,13 @@
 # Changelog
 
+## v0.2.0 — pgvector backend
+
+- Added a read-only pgvector backend for existing PostgreSQL vector tables.
+- Added cosine, Euclidean/L2, and inner-product retrieval with safe SQL
+  identifier composition.
+- Added pgvector table auditing, optional transaction-local HNSW/IVFFlat
+  settings, Docker example, and numerical/SQL-safety tests.
+
 ## v0.1.0 — initial public release
 
 - Candidate-compatibility analysis and Mode A evaluation workflows.

{embedflow-0.1.1 → embedflow-0.2.0}/CITATION.cff

@@ -2,7 +2,7 @@ cff-version: 1.2.0
 title: "EmbedFlow: Upgrading Legacy Embeddings Without Full Upfront Re-Embedding"
 message: "If EmbedFlow contributes to your work, please cite this software release."
 type: software
-version: 0.1.1
+version: 0.2.0
 date-released: 2026-09-06
 repository-code: "https://github.com/arnsri33/embedflow"
 url: "https://github.com/arnsri33/embedflow"

{embedflow-0.1.1 → embedflow-0.2.0}/MANIFEST.in

@@ -1,10 +1,11 @@
 include README.md README_PYPI.md LICENSE CHANGELOG.md CONTRIBUTING.md SECURITY.md CITATION.cff requirements.txt requirements-dev.txt
 recursive-include docs *
-recursive-include examples *.md *.yaml *.json *.jsonl *.py
+recursive-include examples *.md *.yaml *.json *.jsonl *.py *.sql *.sh
 recursive-include scripts *.sh *.py
 recursive-include frozen *.md *.sha256
 prune .github
 prune tests
+prune examples/*/runtime
 global-exclude __pycache__
 global-exclude *.py[cod]
 global-exclude *.index *.npy *.npz *.sqlite *.sqlite3 *.db

{embedflow-0.1.1 → embedflow-0.2.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: embedflow
-Version: 0.1.1
+Version: 0.2.0
 Summary: Progressive embedding-model migration over existing vector indexes.
 Author: Arnav Srivastav
 License-Expression: AGPL-3.0-only
@@ -8,7 +8,7 @@ Project-URL: Homepage, https://embedflow.org
 Project-URL: Repository, https://github.com/arnsri33/embedflow
 Project-URL: Documentation, https://github.com/arnsri33/embedflow#readme
 Project-URL: Issues, https://github.com/arnsri33/embedflow/issues
-Keywords: embeddings,vector-search,rag,information-retrieval,faiss,qdrant
+Keywords: embeddings,vector-search,rag,information-retrieval,faiss,qdrant,pgvector
 Classifier: Development Status :: 3 - Alpha
 Classifier: Intended Audience :: Developers
 Classifier: Intended Audience :: Science/Research
@@ -27,9 +27,11 @@ Provides-Extra: faiss
 Requires-Dist: faiss-cpu>=1.8.0; extra == "faiss"
 Provides-Extra: qdrant
 Requires-Dist: qdrant-client>=1.9; extra == "qdrant"
+Provides-Extra: pgvector
+Requires-Dist: psycopg[binary]>=3.2; extra == "pgvector"
 Provides-Extra: models
-Requires-Dist: huggingface-hub>=0.20; extra == "models"
-Requires-Dist: transformers>=4.45; extra == "models"
+Requires-Dist: huggingface-hub<1.0,>=0.34; extra == "models"
+Requires-Dist: transformers<5.0,>=4.45; extra == "models"
 Requires-Dist: sentence-transformers>=3.0; extra == "models"
 Requires-Dist: torch>=2.2; extra == "models"
 Provides-Extra: dashboard
@@ -39,8 +41,9 @@ Requires-Dist: pydantic>=2.0; extra == "dashboard"
 Provides-Extra: all
 Requires-Dist: faiss-cpu>=1.8.0; extra == "all"
 Requires-Dist: qdrant-client>=1.9; extra == "all"
-Requires-Dist: huggingface-hub>=0.20; extra == "all"
-Requires-Dist: transformers>=4.45; extra == "all"
+Requires-Dist: psycopg[binary]>=3.2; extra == "all"
+Requires-Dist: huggingface-hub<1.0,>=0.34; extra == "all"
+Requires-Dist: transformers<5.0,>=4.45; extra == "all"
 Requires-Dist: sentence-transformers>=3.0; extra == "all"
 Requires-Dist: torch>=2.2; extra == "all"
 Requires-Dist: fastapi>=0.100; extra == "all"
@@ -59,7 +62,7 @@ Dynamic: license-file
 EmbedFlow lets a new embedding model serve over candidates from an existing
 vector index while target document vectors are materialized progressively. It
 supports migration analysis, persistent caching, background work, FAISS,
-Qdrant, a CLI, and FastAPI.
+Qdrant, pgvector, a CLI, and FastAPI.
 
 The full project README and architecture diagram are on
 <https://github.com/arnsri33/embedflow>.
@@ -76,6 +79,12 @@ For FAISS and the dashboard:
 python -m pip install "embedflow[faiss,dashboard]"
 ```
 
+For an existing PostgreSQL/pgvector table:
+
+```bash
+python -m pip install "embedflow[pgvector]"
+```
+
 Qdrant and model-runtime extras are documented in the
 [installation guide](https://github.com/arnsri33/embedflow/blob/main/docs/installation.md).
 For model-backed analysis, install `embedflow[faiss,models,dashboard]`.
@@ -176,9 +185,11 @@ for definitions and reproduction details.
 | --- | --- |
 | FAISS | Supported |
 | Qdrant | Supported |
+| pgvector | Supported |
 
-See the [FAISS guide](https://github.com/arnsri33/embedflow/blob/main/docs/integrations/faiss.md)
-and [Qdrant guide](https://github.com/arnsri33/embedflow/blob/main/docs/integrations/qdrant.md).
+See the [FAISS guide](https://github.com/arnsri33/embedflow/blob/main/docs/integrations/faiss.md),
+[Qdrant guide](https://github.com/arnsri33/embedflow/blob/main/docs/integrations/qdrant.md),
+and [pgvector guide](https://github.com/arnsri33/embedflow/blob/main/docs/integrations/pgvector.md).
 
 ## CLI
 
@@ -188,7 +199,7 @@ embedflow analyze --help
 embedflow serve --config ./embedflow.yaml
 embedflow status --config ./embedflow.yaml
 embedflow registry list
-embedflow economics --corpus-size 1000000000 --docs-per-second 106.98 --gpu-price 3.29
+embedflow economics --corpus-size 1000000000 --docs-per-second 100 --gpu-price 3.29
 embedflow doctor --config ./embedflow.yaml
 ```
 
@@ -198,7 +209,7 @@ cover the remaining commands and endpoints.
 
 ## Status
 
-EmbedFlow v0.1.0 is an alpha release for research and early real-world
+EmbedFlow v0.2.0 is an alpha release for research and early real-world
 testing. T2-v1 is an empirical finite-tail diagnostic, partial rankings can
 differ from fully warm target reranking, and ANN fidelity needs a reference
 comparison to audit.

{embedflow-0.1.1 → embedflow-0.2.0}/README.md

@@ -7,7 +7,7 @@
 EmbedFlow lets a new embedding model serve over candidates from an existing
 vector index while target document vectors are materialized progressively. It
 supports migration analysis, persistent caching, background work, and serving
-through FAISS, Qdrant, a CLI, and FastAPI.
+through FAISS, Qdrant, pgvector, a CLI, and FastAPI.
 
 [Quickstart](#try-it) · [Documentation](#documentation) · [Research](#research)
 
@@ -189,11 +189,13 @@ is measured separately and is `UNKNOWN` until an exact reference is supplied.
 | --- | --- |
 | FAISS | Supported |
 | Qdrant | Supported |
+| pgvector | Supported |
 
 Backend-specific setup and examples:
 
 - [FAISS](https://github.com/arnsri33/embedflow/blob/main/docs/integrations/faiss.md)
 - [Qdrant](https://github.com/arnsri33/embedflow/blob/main/docs/integrations/qdrant.md)
+- [pgvector](https://github.com/arnsri33/embedflow/blob/main/docs/integrations/pgvector.md)
 - [Adding a backend](https://github.com/arnsri33/embedflow/blob/main/CONTRIBUTING.md)
 
 ## CLI
@@ -204,7 +206,7 @@ embedflow analyze --help
 embedflow serve --config ./embedflow.yaml
 embedflow status --config ./embedflow.yaml
 embedflow registry list
-embedflow economics --corpus-size 1000000000 --docs-per-second 106.98 --gpu-price 3.29
+embedflow economics --corpus-size 1000000000 --docs-per-second 100 --gpu-price 3.29
 embedflow doctor --config ./embedflow.yaml
 ```
 
@@ -228,7 +230,7 @@ OpenAPI documentation; see
 
 ## Status
 
-EmbedFlow v0.1.0 is an alpha release for research and early real-world
+EmbedFlow v0.2.0 is an alpha release for research and early real-world
 testing.
 
 - T2-v1 reports an empirical finite-tail diagnostic.

{embedflow-0.1.1 → embedflow-0.2.0}/README_PYPI.md

@@ -5,7 +5,7 @@
 EmbedFlow lets a new embedding model serve over candidates from an existing
 vector index while target document vectors are materialized progressively. It
 supports migration analysis, persistent caching, background work, FAISS,
-Qdrant, a CLI, and FastAPI.
+Qdrant, pgvector, a CLI, and FastAPI.
 
 The full project README and architecture diagram are on
 <https://github.com/arnsri33/embedflow>.
@@ -22,6 +22,12 @@ For FAISS and the dashboard:
 python -m pip install "embedflow[faiss,dashboard]"
 ```
 
+For an existing PostgreSQL/pgvector table:
+
+```bash
+python -m pip install "embedflow[pgvector]"
+```
+
 Qdrant and model-runtime extras are documented in the
 [installation guide](https://github.com/arnsri33/embedflow/blob/main/docs/installation.md).
 For model-backed analysis, install `embedflow[faiss,models,dashboard]`.
@@ -122,9 +128,11 @@ for definitions and reproduction details.
 | --- | --- |
 | FAISS | Supported |
 | Qdrant | Supported |
+| pgvector | Supported |
 
-See the [FAISS guide](https://github.com/arnsri33/embedflow/blob/main/docs/integrations/faiss.md)
-and [Qdrant guide](https://github.com/arnsri33/embedflow/blob/main/docs/integrations/qdrant.md).
+See the [FAISS guide](https://github.com/arnsri33/embedflow/blob/main/docs/integrations/faiss.md),
+[Qdrant guide](https://github.com/arnsri33/embedflow/blob/main/docs/integrations/qdrant.md),
+and [pgvector guide](https://github.com/arnsri33/embedflow/blob/main/docs/integrations/pgvector.md).
 
 ## CLI
 
@@ -134,7 +142,7 @@ embedflow analyze --help
 embedflow serve --config ./embedflow.yaml
 embedflow status --config ./embedflow.yaml
 embedflow registry list
-embedflow economics --corpus-size 1000000000 --docs-per-second 106.98 --gpu-price 3.29
+embedflow economics --corpus-size 1000000000 --docs-per-second 100 --gpu-price 3.29
 embedflow doctor --config ./embedflow.yaml
 ```
 
@@ -144,7 +152,7 @@ cover the remaining commands and endpoints.
 
 ## Status
 
-EmbedFlow v0.1.0 is an alpha release for research and early real-world
+EmbedFlow v0.2.0 is an alpha release for research and early real-world
 testing. T2-v1 is an empirical finite-tail diagnostic, partial rankings can
 differ from fully warm target reranking, and ANN fidelity needs a reference
 comparison to audit.

{embedflow-0.1.1 → embedflow-0.2.0}/docs/api.md

@@ -13,8 +13,8 @@ Interactive OpenAPI documentation is available at
 
 | Method | Path | Purpose |
 | --- | --- | --- |
-| GET | `/health` | Process and backend health |
-| GET | `/status` | Cache, queue, model, and migration state |
+| GET | `/health` | Lightweight process liveness |
+| GET | `/status` | Cache, queue, model, migration, and backend metadata |
 | POST | `/search` | Source retrieval and target reranking |
 | POST | `/analyze` | Run or retrieve migration analysis |
 | POST | `/prewarm` | Queue document materialization |
@@ -48,6 +48,10 @@ The response includes the result list and migration fields such as:
 request. A partial response scores the available target vectors; it can differ
 from the fully warm ranking.
 
+`/status` includes safe backend metadata. For pgvector this names the
+schema/table and vector contract without returning the DSN or credentials;
+`/health` remains a compact liveness response for probes and load balancers.
+
 ## Errors
 
 Validation errors use FastAPI/Pydantic's normal JSON response. Backend,

{embedflow-0.1.1 → embedflow-0.2.0}/docs/cli.md

@@ -52,7 +52,7 @@ Registry matching uses model contracts and corpus identity. See
 ```bash
 embedflow economics \
   --corpus-size 1000000000 \
-  --docs-per-second 106.98 \
+  --docs-per-second 100 \
   --gpu-price 3.29
 embedflow doctor --config ./embedflow.yaml
 embedflow demo

{embedflow-0.1.1 → embedflow-0.2.0}/docs/configuration.md

@@ -14,12 +14,14 @@ target:
   device: cuda
 
 index:
-  backend: faiss                 # faiss or qdrant
+  backend: faiss                 # faiss, qdrant, or pgvector
   path: ./legacy.index
   ids: ./legacy.index.ids.json   # FAISS sidecar
   metric: cosine
   nprobe: 64
   # Qdrant fields: url, collection, vector_name, api_key_env
+  # pgvector fields: dsn_env, schema, table, id_column, vector_column, text_column
+  #                 hnsw_ef_search, ivfflat_probes
 
 documents:
   path: ./documents.jsonl
@@ -78,6 +80,14 @@ EMBEDFLOW_INDEX_URL
 EMBEDFLOW_INDEX_COLLECTION
 EMBEDFLOW_INDEX_VECTOR_NAME
 EMBEDFLOW_QDRANT_API_KEY_ENV
+EMBEDFLOW_PGVECTOR_DSN_ENV
+EMBEDFLOW_PGVECTOR_SCHEMA
+EMBEDFLOW_PGVECTOR_TABLE
+EMBEDFLOW_PGVECTOR_ID_COLUMN
+EMBEDFLOW_PGVECTOR_VECTOR_COLUMN
+EMBEDFLOW_PGVECTOR_TEXT_COLUMN
+EMBEDFLOW_PGVECTOR_HNSW_EF_SEARCH
+EMBEDFLOW_PGVECTOR_IVFFLAT_PROBES
 EMBEDFLOW_INDEX_NPROBE
 EMBEDFLOW_DOCUMENTS_PATH
 EMBEDFLOW_CACHE_PATH

{embedflow-0.1.1 → embedflow-0.2.0}/docs/economics.md

@@ -6,7 +6,7 @@ throughput and GPU price supplied by the user:
 ```bash
 embedflow economics \
   --corpus-size 1000000000 \
-  --docs-per-second 106.98 \
+  --docs-per-second 100 \
   --gpu-price 3.29
 ```
 

{embedflow-0.1.1 → embedflow-0.2.0}/docs/installation.md

@@ -30,6 +30,7 @@ python -m pip install -e .
 | --- | --- |
 | `faiss` | FAISS source-index adapter |
 | `qdrant` | Qdrant client and adapter |
+| `pgvector` | Psycopg 3 binary driver and pgvector adapter |
 | `models` | PyTorch, Transformers, Sentence Transformers, and Hub client |
 | `dashboard` | FastAPI, Uvicorn, and Pydantic |
 | `dev` | Pytest, Ruff, and build tooling |
@@ -52,6 +53,12 @@ with `index.url`; a user-owned cloud or remote server can use an API key named
 by `index.api_key_env`. Keep the key in the environment. See
 [`integrations/qdrant.md`](integrations/qdrant.md).
 
+## PostgreSQL / pgvector
+
+Install the optional adapter with `python -m pip install "embedflow[pgvector]"`.
+The adapter connects to an existing table and reads the DSN from the
+environment; see [`integrations/pgvector.md`](integrations/pgvector.md).
+
 ## CPU and GPU
 
 The deterministic demo runs on CPU. Real model serving accepts `--device cpu`

embedflow-0.2.0/docs/integrations/pgvector.md

@@ -0,0 +1,140 @@
+# PostgreSQL / pgvector
+
+EmbedFlow can read candidates from an existing PostgreSQL table that uses the
+[pgvector](https://github.com/pgvector/pgvector) extension. The adapter uses
+the shared `VectorIndex` interface, so analysis, serving, cache warming, and
+the API follow the same path as FAISS and Qdrant.
+
+## Install
+
+```bash
+python -m pip install "embedflow[pgvector]"
+```
+
+The extra installs Psycopg 3 with its binary distribution. The base
+`embedflow` install does not import or require PostgreSQL dependencies.
+
+## Connection and table layout
+
+Keep the DSN in an environment variable:
+
+```bash
+export EMBEDFLOW_PGVECTOR_DSN='postgresql://user:password@localhost:5432/app'
+```
+
+Example configuration:
+
+```yaml
+source:
+  model: sentence-transformers/all-MiniLM-L6-v2
+  device: cpu
+
+target:
+  model: Qwen/Qwen3-Embedding-0.6B
+  device: cpu
+
+index:
+  backend: pgvector
+  dsn_env: EMBEDFLOW_PGVECTOR_DSN
+  schema: public
+  table: documents
+  id_column: id
+  vector_column: embedding
+  text_column: content
+  metric: cosine
+  # hnsw_ef_search: 100
+  # ivfflat_probes: 20
+
+documents:
+  # Omit this file when text_column is present in the table. A separate
+  # JSONL document store can be supplied when the table stores IDs only.
+  path: ./documents.jsonl
+  id_field: id
+  text_field: text
+```
+
+When `documents.path` does not exist, EmbedFlow resolves candidate text from
+`text_column` in the configured table. If a JSONL file exists, it remains the
+document store and the pgvector table is used for vectors and IDs. The adapter
+only reads the table during ordinary `analyze`, `serve`, `search`, `status`,
+and `audit-index` operations; it does not create extensions, indexes, tables,
+or modify rows.
+
+The ID column may be an integer, bigint, UUID, or text type. IDs are normalized
+to EmbedFlow's canonical string representation without numeric coercion.
+The configured vector dimension is checked against `vector(n)` where available
+and against a bounded non-null row for unbounded `vector` columns.
+
+## Metrics and score direction
+
+Supported metrics are `cosine`, `l2`/`euclidean`, and
+`inner_product`/`dot`. pgvector's distance operators are converted to the
+EmbedFlow convention where larger scores rank first:
+
+| EmbedFlow metric | pgvector operator | Internal score |
+| --- | --- | --- |
+| cosine | `<=>` | `1 - distance` |
+| l2 / euclidean | `<->` | `-distance` |
+| inner_product / dot | `<#>` | `-distance` |
+
+The last conversion accounts for pgvector's negative inner-product operator.
+
+## ANN settings
+
+EmbedFlow never creates or changes an ANN index. If configured, `hnsw_ef_search`
+and `ivfflat_probes` are applied with transaction-local `set_config` calls for
+the retrieval query. Omit them to use the database/index defaults. The
+`metadata()` and `audit-index` output reports configured settings and detected
+valid HNSW/IVFFlat indexes when PostgreSQL exposes them.
+
+## Commands
+
+```bash
+embedflow analyze --config embedflow.yaml
+embedflow serve --config embedflow.yaml
+embedflow search --config embedflow.yaml "what causes auroras?"
+embedflow status --config embedflow.yaml
+embedflow audit-index --config embedflow.yaml
+```
+
+`audit-index` checks table and column presence, pgvector type and dimension,
+NULL vectors, duplicate canonical IDs, a retrieval probe, and available ANN
+metadata. A successful connection is not an ANN recall audit; use an exact
+reference comparison when fidelity matters.
+
+## Local example
+
+The repository includes a small Docker example that creates a disposable
+pgvector table and index:
+
+```bash
+cd examples/pgvector
+docker compose up -d
+export EMBEDFLOW_PGVECTOR_DSN='postgresql://embedflow:embedflow@localhost:5432/embedflow'
+./run_demo.sh
+```
+
+The example is for local testing only. It contains no credentials for a real
+service and does not represent a production database configuration.
+
+## Troubleshooting
+
+- `pgvector support requires psycopg`: install `embedflow[pgvector]` in the
+  active environment.
+- `DSN not found`: export the variable named by `index.dsn_env`.
+- `table ... was not found`: check the schema/table spelling and database.
+- `is not a pgvector column`: install/enable the extension in the database and
+  point `vector_column` at a `vector` column.
+- Dimension mismatch: set `source.dimension` to the encoder's dimension and
+  verify the table's vector type.
+- Authentication errors are redacted in EmbedFlow output; inspect PostgreSQL
+  server logs without pasting passwords into issue reports.
+
+## Limitations
+
+The adapter currently exposes the common table layout and no arbitrary SQL
+filters. Metadata filtering will follow the shared backend interface when that
+interface gains a portable filter contract. Connection operations on one
+adapter are serialized; applications needing a larger pool can create several
+sessions behind their own pool. ANN health is reported as `UNKNOWN` unless a
+reference comparison is run.

{embedflow-0.1.1 → embedflow-0.2.0}/docs/limitations.md

@@ -1,6 +1,6 @@
 # Limitations and release scope
 
-EmbedFlow v0.1.0 is an alpha release for research and early real-world
+EmbedFlow v0.2.0 is an alpha release for research and early real-world
 testing. The serving path is designed to make migration experiments concrete;
 production rollout still requires application-specific validation.
 

{embedflow-0.1.1 → embedflow-0.2.0}/docs/releasing.md

@@ -18,8 +18,8 @@ python -m twine check dist/*
 Inspect both archives before uploading:
 
 ```bash
-unzip -l dist/embedflow-0.1.0-py3-none-any.whl
-tar -tzf dist/embedflow-0.1.0.tar.gz
+unzip -l dist/embedflow-0.2.0-py3-none-any.whl
+tar -tzf dist/embedflow-0.2.0.tar.gz
 sha256sum dist/*
 ```
 
@@ -32,7 +32,7 @@ Test the wheel outside the source tree:
 ```bash
 python -m venv /tmp/embedflow-wheel-test
 /tmp/embedflow-wheel-test/bin/python -m pip install --upgrade pip
-/tmp/embedflow-wheel-test/bin/python -m pip install dist/embedflow-0.1.0-py3-none-any.whl
+/tmp/embedflow-wheel-test/bin/python -m pip install dist/embedflow-0.2.0-py3-none-any.whl
 cd /tmp
 /tmp/embedflow-wheel-test/bin/python -c "import embedflow; print(embedflow.__version__)"
 /tmp/embedflow-wheel-test/bin/embedflow --help
@@ -65,7 +65,7 @@ python -m venv /tmp/embedflow-testpypi
 /tmp/embedflow-testpypi/bin/python -m pip install \
   --index-url https://test.pypi.org/simple/ \
   --extra-index-url https://pypi.org/simple/ \
-  embedflow==0.1.0
+  embedflow==0.2.0
 cd /tmp
 /tmp/embedflow-testpypi/bin/python -c "import embedflow; print(embedflow.__version__)"
 /tmp/embedflow-testpypi/bin/embedflow --help
@@ -79,11 +79,11 @@ Test optional integrations in a second clean environment:
 /tmp/embedflow-testpypi/bin/python -m pip install \
   --index-url https://test.pypi.org/simple/ \
   --extra-index-url https://pypi.org/simple/ \
-  "embedflow[faiss,dashboard]==0.1.0"
+  "embedflow[faiss,dashboard]==0.2.0"
 ```
 
 If the same filename already exists on TestPyPI, use a pre-release such as
-`0.1.0rc1` for the TestPyPI-only trial. Keep production `0.1.0` unchanged.
+`0.2.0rc1` for the TestPyPI-only trial. Keep production `0.2.0` unchanged.
 
 ## Trusted Publishing configuration
 
@@ -115,7 +115,7 @@ above keeps the test step explicit.
 3. Run the final release gate and review the generated report.
 4. Configure the PyPI pending publisher and protected `pypi` environment.
 5. Create a Git tag and GitHub Release for the exact package version, for
-   example `v0.1.0`.
+   example `v0.2.0`.
 6. Approve the `pypi` environment when the release workflow is ready.
 7. Verify the files and metadata on PyPI.
 8. Install from production PyPI in a directory outside this checkout.
@@ -123,7 +123,7 @@ above keeps the test step explicit.
 The GitHub workflow builds once, validates the metadata and wheel, transfers
 those exact files as an artifact, and then publishes them. PyPI filenames are
 immutable, so a correction after publishing requires a new version such as
-`0.1.1`.
+`0.2.1`.
 
 ## After production publication
 

{embedflow-0.1.1 → embedflow-0.2.0}/embedflow/__init__.py

@@ -1,12 +1,12 @@
 """EmbedFlow: progressive embedding-model migration for existing indexes."""
 
-__version__ = "0.1.1"
+__version__ = "0.2.0"
 
 from .config import EmbedFlowConfig, load_config
 
 
 def migrate(*args, **kwargs):
-    """Start progressive migration over an existing FAISS/Qdrant index.
+    """Start progressive migration over an existing FAISS, Qdrant, or pgvector index.
 
     Imported lazily to keep the lightweight configuration package free of
     model-serving dependencies at import time.  See ``embedflow.migration``

{embedflow-0.1.1 → embedflow-0.2.0}/embedflow/analysis.py

@@ -129,7 +129,7 @@ def analyze_migration(
         )
         if use_registry and registry.level != "EXACT REGISTRY MATCH":
             raise ValueError("--use-registry requires an EXACT REGISTRY MATCH")
-        engine = open_engine(config_path, device=device, demo=demo, start_worker=False)
+        engine = open_engine(config_path, device=device, demo=demo, start_worker=False, documents=docs)
         result = run_probe(
             engine.source_model,
             engine.target_model,
@@ -185,6 +185,10 @@ def analyze_migration(
     finally:
         if engine is not None:
             engine.close()
+        elif 'docs' in locals():
+            close_documents = getattr(docs, "close", None)
+            if callable(close_documents):
+                close_documents()
         if temporary_config is not None:
             temporary_config.unlink(missing_ok=True)