patchvec 0.5.8__tar.gz → 0.5.9__tar.gz

{patchvec-0.5.8 → patchvec-0.5.9}/ABOUT.md

@@ -12,7 +12,7 @@ Upload → chunk → index (with metadata) → search via REST and CLI.
 - Metadata filters on search (`{"filters": {"docid": "DOC-1"}}`)
 - REST and CLI entry points
 - Health/metrics endpoints + Prometheus exporter
-- Pluggable embeddings and stores; default backend is local
+- Pluggable embeddings and stores; default stack is local FAISS + SBERT
 
 ## Requirements
 - Python 3.10–3.14
@@ -47,27 +47,30 @@ export PATCHVEC_AUTH__GLOBAL_KEY="your-secret"
 ```
 
 ## Minimal config (optional)
-By default PatchVec runs with sensible local defaults. To customize, create
-`config.yml`:
+By default PatchVec runs with sensible local defaults. For a user install,
+customize `~/patchvec/config.yml`:
 ```yaml
 vector_store:
-  type: default
+  type: faiss
 embedder:
-  type: default
+  type: sbert
 auth:
   mode: static
   global_key: ${PATCHVEC_GLOBAL_KEY}
 ```
 Then export:
 ```bash
-export PATCHVEC_CONFIG=./config.yml
 export PATCHVEC_GLOBAL_KEY="your-secret"
 ```
+If you keep the file elsewhere, point the runtime at it explicitly:
+```bash
+export PATCHVEC_CONFIG=/path/to/config.yml
+```
 
 ## CLI example
 ```bash
 pavecli create-collection demo books
-pavecli upload demo books demo/20k_leagues.txt --docid=verne-20k \
+pavecli ingest demo books demo/20k_leagues.txt --docid=verne-20k \
   --metadata='{"lang":"en"}'
 pavecli search demo books "captain nemo" -k 5
 ```

{patchvec-0.5.8/patchvec.egg-info → patchvec-0.5.9}/PKG-INFO

@@ -1,7 +1,7 @@
 Metadata-Version: 2.4
 Name: patchvec
-Version: 0.5.8
-Summary: Patchvec — A lightweight, pluggable vector search microservice.
+Version: 0.5.9
+Summary: PaveDB — A lightweight, pluggable vector search microservice.
 Author: Rodrigo Rodrigues da Silva
 Author-email: rodrigo@flowlexi.com
 License: AGPL-3.0-or-later
@@ -24,7 +24,6 @@ Description-Content-Type: text/markdown
 License-File: LICENSE
 Requires-Dist: fastapi>=0.115.0
 Requires-Dist: uvicorn[standard]>=0.30.6
-Requires-Dist: txtai>=6.3.0
 Requires-Dist: pydantic>=2.8.2
 Requires-Dist: python-multipart>=0.0.9
 Requires-Dist: pypdf>=5.0.0
@@ -32,14 +31,14 @@ Requires-Dist: pyyaml>=6.0.2
 Requires-Dist: python-dotenv>=1.0.1
 Requires-Dist: faiss-cpu>=1.7.1
 Requires-Dist: torch>=2.10.0
+Requires-Dist: sentence-transformers>=2.7.0
 Provides-Extra: cpu
-Provides-Extra: sbert
-Requires-Dist: sentence-transformers>=2.7.0; extra == "sbert"
 Provides-Extra: openai
 Requires-Dist: openai>=1.0.0; extra == "openai"
 Provides-Extra: test
 Requires-Dist: pytest; extra == "test"
 Requires-Dist: httpx; extra == "test"
+Requires-Dist: datasets>=3.5.0; extra == "test"
 Dynamic: author
 Dynamic: author-email
 Dynamic: classifier
@@ -67,7 +66,7 @@ Upload → chunk → index (with metadata) → search via REST and CLI.
 - Metadata filters on search (`{"filters": {"docid": "DOC-1"}}`)
 - REST and CLI entry points
 - Health/metrics endpoints + Prometheus exporter
-- Pluggable embeddings and stores; default backend is local
+- Pluggable embeddings and stores; default stack is local FAISS + SBERT
 
 ## Requirements
 - Python 3.10–3.14
@@ -102,27 +101,30 @@ export PATCHVEC_AUTH__GLOBAL_KEY="your-secret"
 ```
 
 ## Minimal config (optional)
-By default PatchVec runs with sensible local defaults. To customize, create
-`config.yml`:
+By default PatchVec runs with sensible local defaults. For a user install,
+customize `~/patchvec/config.yml`:
 ```yaml
 vector_store:
-  type: default
+  type: faiss
 embedder:
-  type: default
+  type: sbert
 auth:
   mode: static
   global_key: ${PATCHVEC_GLOBAL_KEY}
 ```
 Then export:
 ```bash
-export PATCHVEC_CONFIG=./config.yml
 export PATCHVEC_GLOBAL_KEY="your-secret"
 ```
+If you keep the file elsewhere, point the runtime at it explicitly:
+```bash
+export PATCHVEC_CONFIG=/path/to/config.yml
+```
 
 ## CLI example
 ```bash
 pavecli create-collection demo books
-pavecli upload demo books demo/20k_leagues.txt --docid=verne-20k \
+pavecli ingest demo books demo/20k_leagues.txt --docid=verne-20k \
   --metadata='{"lang":"en"}'
 pavecli search demo books "captain nemo" -k 5
 ```

patchvec-0.5.9/README.md

@@ -0,0 +1,258 @@
+<!-- (C) 2025, 2026 Rodrigo Rodrigues da Silva <rodrigo@flowlexi.com> --> 
+<!-- SPDX-License-Identifier: AGPL-3.0-or-later -->
+
+# 🛣️ PatchVec — Vector search you can understand & deploy within minutes
+
+PatchVec is a single-process vector search engine for AI applications.
+
+It ingests your documents, chunks and embeds them, and gives you semantic search with
+full provenance — document id, page, character offset, and the exact snippet that
+matched.
+
+Built for developers shipping **RAG (Retrieval-Augmented Generation)** systems, PatchVec
+provides a straightforward service for **vector search, embeddings pipelines, and
+semantic retrieval**. It runs as a **REST service, a CLI tool, or an embedded library**,
+so you can ship your first version quickly and keep growing with the same codebase as
+your application scales. No cluster. No opaque pipelines.
+
+Drop a file in. Search it. See exactly what came back — and why. Minutes after your
+first commit.
+
+## ⚙️ Why PatchVec
+
+- **Ingest files, not embeddings** — hand it a PDF, CSV, or TXT (more formats to come)
+  and PatchVec chunks, embeds, and indexes it. No preprocessing pipeline to build.
+- **Full provenance on every hit** — every search result traces back to a document,
+  page, and character offset. Latency and request traceability are built into every
+  response.
+- **Multi-tenant by default** — tenant/collection namespacing is built in, not bolted
+    on (and transparent when you just don't need it).
+- **REST, CLI, or embed it** — run as an HTTP service, script via the CLI, or import the
+  library directly in your Python app.
+- **Pluggable embeddings** (soon) — swap models per collection; wire in local or hosted
+  embedding backends.
+
+## 🧭 How to
+
+### 🐳 Docker workflow (prebuilt images)
+
+Pull the image that fits your hardware from the
+[Flowlexi Container Registry](https://gitlab.com/flowlexi/patchvec/container_registry)
+on GitLab (CUDA builds publish as `latest-gpu`, CPU-only as
+`latest-cpu`).
+
+```bash
+docker pull registry.gitlab.com/flowlexi/patchvec/patchvec:latest-gpu
+docker pull registry.gitlab.com/flowlexi/patchvec/patchvec:latest-cpu
+```
+
+Run the service by choosing the tag you need and mapping the API port locally:
+
+```bash
+docker run -d --name patchvec \
+  -p 8086:8086 \
+  registry.gitlab.com/flowlexi/patchvec/patchvec:latest-cpu
+```
+
+Use the bundled CLI inside the container to create a tenant/collection, ingest a demo
+document, and query it:
+
+```bash
+docker exec patchvec pavecli create-collection demo books
+docker exec patchvec pavecli ingest demo books /app/demo/20k_leagues.txt \
+  --docid=verne-20k --metadata='{"lang":"en"}'
+docker exec patchvec pavecli search demo books "captain nemo" -k 3
+```
+
+See below for REST and UI.
+
+Stop the container when you are done:
+
+```bash
+docker rm -f patchvec
+```
+
+### 🐍 PyPI workflow
+
+Install Patchvec from PyPI inside an isolated virtual environment. You can run it
+purely from env vars, or later point it at an explicit config file.
+
+**Requires Python 3.10–3.14.**
+
+```bash
+mkdir -p ~/pv && cd ~/pv  # or wherever
+python -m venv .venv-pv
+source .venv-pv/bin/activate
+python -m pip install --upgrade pip
+pip install "patchvec[cpu]"
+
+# create the default instance under ~/patchvec
+pavecli init
+
+# sample demo corpus
+curl -LO https://raw.githubusercontent.com/rodrigopitanga/patchvec/main/demo/20k_leagues.txt
+
+# set an admin key for the generated config
+export PATCHVEC_GLOBAL_KEY=super-sekret
+
+# option A: run the service (stays up until you stop it)
+pavesrv
+
+# option B: operate entirely via the CLI (no server needed)
+pavecli create-collection demo books
+pavecli ingest demo books 20k_leagues.txt --docid=verne-20k --metadata='{"lang":"en"}'
+pavecli search demo books "captain nemo" -k 3
+```
+
+> **CPU-only deployments:** The command above pulls the default PyTorch wheel
+> from PyPI, which includes CUDA support (~2 GB). For a leaner, CPU-only torch
+> install, point pip at the PyTorch CPU index:
+> ```bash
+> pip install "patchvec[cpu]" \
+> --index-url https://download.pytorch.org/whl/cpu \
+> --extra-index-url https://pypi.org/simple
+> ```
+
+Deactivate the virtual environment with `deactivate` when finished.
+
+By default, a non-dev runtime reads `~/patchvec/config.yml` if present, keeps
+tenant sidecar loading disabled unless `auth.tenants_file` is configured, and
+stores data in `~/patchvec/data`. You can override any of that with the
+`PATCHVEC_*` environment scheme or by pointing `PATCHVEC_CONFIG` at an explicit
+config file. For alternate instances, use `pavecli init /path/to/instance` and
+then point commands at that root with `pavesrv --home=/path/to/instance` or
+`pavecli <command> ... --home /path/to/instance`.
+
+### 🌐 REST API and Web UI usage
+
+When the server is running (either via Docker or `pavesrv`), the API listens on
+`http://localhost:8086`. The following `curl` commands mirror the CLI sequence
+above—adjust the file path to wherever you stored the corpus
+(`/app/demo/20k_leagues.txt` in Docker, `~/pv/20k_leagues.txt` for PyPI installs) and
+reuse the bearer token exported earlier:
+
+```bash
+# create collection
+curl -H "Authorization: Bearer $PATCHVEC_GLOBAL_KEY" \
+  -X POST http://localhost:8086/collections/demo/books
+
+# ingest document
+curl -H "Authorization: Bearer $PATCHVEC_GLOBAL_KEY" \
+  -X POST http://localhost:8086/collections/demo/books/documents \
+  -F "file=@20k_leagues.txt" \
+  -F 'metadata={"lang":"en"}'
+
+# run search
+curl -H "Authorization: Bearer $PATCHVEC_GLOBAL_KEY" \
+  "http://localhost:8086/collections/demo/books/search?q=captain+nemo&k=3"
+```
+
+Every hit comes back with provenance you can trace, plus latency
+and request id for observability:
+
+```json
+{
+  "matches": [
+    {
+      "id": "verne-20k::chunk_42",
+      "score": 0.82,
+      "text": "Captain Nemo conducted me to the central staircase ...",
+      "tenant": "demo",
+      "collection": "books",
+      "match_reason": "semantic",
+      "meta": {
+        "docid": "verne-20k",
+        "filename": "20k_leagues.txt",
+        "offset": 21000,
+        "lang": "en",
+        "ingested_at": "2026-03-07T12:00:00Z"
+      }
+    }
+  ],
+  "latency_ms": 12.4,
+  "request_id": "req-5f3a-b812"
+}
+```
+
+The Swagger UI is available at `http://localhost:8086/`.
+
+Health and metrics endpoints are available at `/health` and `/metrics`.
+
+Runtime options are also accepted via the `PATCHVEC_*` environment variable scheme
+(`PATCHVEC_SERVER__PORT`, `PATCHVEC_AUTH__MODE`, etc.), which precedes config files.
+
+### 🔁 Live data updates
+
+Patchvec supports live data refresh without restarting the server. Re-ingest the same
+`docid` to *replace* vector content (filename doesn't matter - metadata will change
+though), or explicitly delete the document and then ingest it again.
+
+Re-ingest to replace (CLI path example):
+
+```bash
+# initial ingest
+pavecli ingest demo books 20k_leagues.txt --docid=verne-20k
+
+# modify the content (filename can change — docid is what matters)
+cp 20k_leagues.txt 20k_leagues_v2.txt
+echo "THE END" >> 20k_leagues_v2.txt
+
+# re-ingest with the *same docid* to replace the indexed content
+pavecli ingest demo books 20k_leagues_v2.txt --docid=verne-20k
+```
+
+Delete by ID then ingest (REST path example):
+
+```bash
+curl -H "Authorization: Bearer $PATCHVEC_GLOBAL_KEY" \
+  -X DELETE http://localhost:8086/collections/demo/books/documents/verne-20k
+
+# make changes
+
+curl -H "Authorization: Bearer $PATCHVEC_GLOBAL_KEY" \
+  -X POST http://localhost:8086/collections/demo/books/documents \
+  -F "file=@demo/20k_leagues.txt" \
+  -F 'docid=verne-20k'
+```
+
+### 🛠️ Developer workflow
+
+Building from source relies on `Makefile` shortcuts (`make install-dev`, `make serve`,
+`make test`, `make check`, etc.).
+
+The full contributor workflow, target reference, and coding style live in
+[CONTRIBUTING.md](CONTRIBUTING.md). Performance benchmarks are documented in
+[README-benchmarks.md](README-benchmarks.md).
+
+## Logging
+
+PatchVec writes human-readable logs to stderr and optionally emits
+structured JSON lines (one per search/ingest/delete) for production
+observability. Enable the ops stream in `config.yml`:
+
+```yaml
+log:
+  ops_log: stdout   # null (off) | stdout | /path/to/ops.jsonl
+```
+
+See `config.yml.example` for the full logging configuration.
+
+## 🗺️ Roadmap
+
+Short/mid-term tasks and long-term plans are all tracked in
+[`ROADMAP.md`](ROADMAP.md). Pick one, open an issue titled `claim: <task ID>`, and
+ship a patch. If you find a bug, file it under the *Issues* tab.
+
+## 📜 License
+
+PatchVec is free software: you can use it, copy it, redistribute it and/or modify it
+free of charge under the terms of the GNU Affero General Public License as published by
+the Free Software Foundation, either version 3 of the License, or (at your option) any
+later version.
+
+PatchVec is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY;
+without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
+PURPOSE.  See the GNU Affero General Public License for more details.
+
+SPDX-License-Identifier: AGPL-3.0-or-later
+Copyright (C) 2025, 2026 Rodrigo Rodrigues da Silva <rodrigo@flowlexi.com>

{patchvec-0.5.8 → patchvec-0.5.9}/config.yml.example

@@ -2,31 +2,50 @@
 # Copy to config.yml and adjust.
 # Secrets (API keys) should live in a separate untracked file — see tenants.yml.example.
 #
-# Config file location (default: ~/patchvec/config.yml):
-#   PATCHVEC_CONFIG=/etc/patchvec/config.yml
+# Default user-install paths:
+#   config:  ~/pavedb/config.yml
+#   tenants: ~/pavedb/tenants.yml
+#   data:    ~/pavedb/data
 #
-# In Docker/systemd deployments always set PATCHVEC_CONFIG explicitly — the
+# Distro-like install example:
+#   config:  /etc/pavedb/config.yml
+#   tenants: /var/pavedb/tenants.yml
+#   data:    /var/pavedb/data
+#
+# Config file location override:
+#   PAVEDB_CONFIG=/etc/pavedb/config.yml
+#
+# In Docker/systemd deployments always set PAVEDB_CONFIG explicitly — the
 # default path expands ~ relative to the process user, which may not be what
 # you expect inside a container. Example compose snippet:
 #
 #   environment:
-#     PATCHVEC_CONFIG: /etc/patchvec/config.yml
+#     PAVEDB_CONFIG: /etc/pavedb/config.yml
 #   volumes:
-#     - ./config.yml:/etc/patchvec/config.yml:ro
+#     - ./config.yml:/etc/pavedb/config.yml:ro
 #
 # All keys can also be overridden inline via environment variables:
-#   PATCHVEC_<KEY>=value          (top-level, e.g. PATCHVEC_DATA_DIR)
-#   PATCHVEC_<SECTION>__<KEY>=val (nested,    e.g. PATCHVEC_LOG__LEVEL=debug)
+#   PAVEDB_<KEY>=value          (top-level, e.g. PAVEDB_DATA_DIR)
+#   PAVEDB_<SECTION>__<KEY>=val (nested,    e.g. PAVEDB_LOG__LEVEL=debug)
+# Legacy PATCHVEC_* vars still work in v0.5.9 but will be removed in v0.6.
+#
+# `auth.tenants_file` is optional. If unset, its default is `None` and no
+# tenants sidecar file is loaded.
+# If set, PaveDB loads that sidecar first. Then inline tenant config is
+# applied with precedence:
+#   env vars > config.yml > tenants.yml > defaults
+# Example: define tenant "acme" entirely from env:
+#   PAVEDB_AUTH__API_KEYS__acme=change-me
+#   PAVEDB_TENANTS__acme__MAX_CONCURRENT=5
 
 # ---------------------------------------------------------------------------
 # Storage
 # ---------------------------------------------------------------------------
 
 # Data directory — ~ is expanded at startup.
-# Default (library/dev): ~/patchvec/data
-# For Docker/systemd use an absolute path:
-#   mkdir -p /var/lib/patchvec/data && chown patchvec:patchvec /var/lib/patchvec/data
-data_dir: ~/patchvec/data
+# Default (library/dev): ~/pavedb/data
+# For distro-like installs use an absolute path, e.g. /var/pavedb/data.
+data_dir: ~/pavedb/data
 
 # ---------------------------------------------------------------------------
 # Common collection
@@ -55,11 +74,13 @@ auth:
 
   # Global admin key — grants access to all tenants and admin routes.
   # Always read from the environment; never hardcode in committed files.
-  global_key: ${PATCHVEC_GLOBAL_KEY}
+  global_key: ${PAVEDB_GLOBAL_KEY}
 
-  # External tenant→key mapping file (untracked, merged at startup).
-  # Keys in this file override any inline api_keys entries with the same name.
-  tenants_file: ./tenants.yml
+  # External tenant→key mapping file.
+  # Keep the same key paths there (`auth.api_keys`, `tenants.*`).
+  # User install default:  ~/pavedb/tenants.yml
+  # Distro-like install:   /var/pavedb/tenants.yml
+  # tenants_file: ~/pavedb/tenants.yml
 
   # Inline tenant→key mapping (fallback; keep empty in the repo).
   api_keys: {}
@@ -69,51 +90,36 @@ auth:
 # ---------------------------------------------------------------------------
 
 vector_store:
-  # default — txtai + FAISS (built-in, no extra services required).
-  # qdrant  — Qdrant server (requires a running Qdrant instance).
-  type: default
+  # faiss   — local FAISS index (built-in, no extra services required).
+  type: faiss
 
-  # Options for type=default (txtai-backed FAISS store).
-  txtai:
+  # Options for type=faiss.
+  faiss:
     embed_model: sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2
-    backend: faiss
-
-  # Options for type=qdrant.
-  # NOTE: QdrantStore is currently a stub — all methods raise NotImplementedError.
-  # Contributions welcome! See pave/stores/qdrant_store.py.
-  qdrant:
-    url: http://localhost:6333
-    api_key: ${PATCHVEC_QDRANT_API_KEY}
-    prefer_payload_filters: true
-    collection_prefix: patchvec_
+    max_query_chars: 4000
 
 # ---------------------------------------------------------------------------
 # Embedder
 # ---------------------------------------------------------------------------
 # Controls how text is converted to vectors before indexing and querying.
 # The embedder is shared across all collections (per-collection config: v0.6).
-#
-# NOTE: the sbert and openai embedder types are currently stubs — the factory
-# exists but is not wired into the default store. Only type=default (txtai) is
-# fully functional today. Per-collection embedder selection lands in v0.6.
-# Contributions welcome! See pave/embedders/ and pave/stores/factory.py.
 
 embedder:
-  # default — txtai-managed sentence-transformers model (recommended).
-  # sbert   — direct sentence-transformers, more control over batching/device.
-  # openai  — OpenAI text-embedding-ada-002 (requires API key, adds latency).
-  type: default
-
-  txtai:
-    path: sentence-transformers/paraphrase-MiniLM-L3-v2
+  # sbert  — direct sentence-transformers (recommended local default).
+  # openai — OpenAI embeddings API (requires API key, adds latency).
+  type: sbert
 
+  # Used by type=sbert.
   sbert:
     model: sentence-transformers/all-MiniLM-L6-v2
     batch_size: 64
-    device: auto   # cpu | cuda | auto (auto detects CUDA at runtime)
+    device: auto   # cpu | cuda | mps | auto
 
+  # Used by type=openai.
   openai:
-    api_key: ${PATCHVEC_OPENAI_API_KEY}
+    model: text-embedding-3-small
+    batch_size: 256
+    api_key: ${PAVEDB_OPENAI_API_KEY}
     dim: 1536
 
 # ---------------------------------------------------------------------------
@@ -194,7 +200,7 @@ server:
 
 log:
   # Dev log level (stderr). DEBUG | INFO | WARNING | ERROR — default INFO.
-  # Overridden by PATCHVEC_LOG__LEVEL env var (e.g. in Makefile: debug).
+  # Overridden by PAVEDB_LOG__LEVEL env var (e.g. in Makefile: debug).
   # Per-namespace overrides: log.debug / log.watch / log.quiet (list of loggers).
   level: INFO
 

{patchvec-0.5.8 → patchvec-0.5.9/patchvec.egg-info}/PKG-INFO

@@ -1,7 +1,7 @@
 Metadata-Version: 2.4
 Name: patchvec
-Version: 0.5.8
-Summary: Patchvec — A lightweight, pluggable vector search microservice.
+Version: 0.5.9
+Summary: PaveDB — A lightweight, pluggable vector search microservice.
 Author: Rodrigo Rodrigues da Silva
 Author-email: rodrigo@flowlexi.com
 License: AGPL-3.0-or-later
@@ -24,7 +24,6 @@ Description-Content-Type: text/markdown
 License-File: LICENSE
 Requires-Dist: fastapi>=0.115.0
 Requires-Dist: uvicorn[standard]>=0.30.6
-Requires-Dist: txtai>=6.3.0
 Requires-Dist: pydantic>=2.8.2
 Requires-Dist: python-multipart>=0.0.9
 Requires-Dist: pypdf>=5.0.0
@@ -32,14 +31,14 @@ Requires-Dist: pyyaml>=6.0.2
 Requires-Dist: python-dotenv>=1.0.1
 Requires-Dist: faiss-cpu>=1.7.1
 Requires-Dist: torch>=2.10.0
+Requires-Dist: sentence-transformers>=2.7.0
 Provides-Extra: cpu
-Provides-Extra: sbert
-Requires-Dist: sentence-transformers>=2.7.0; extra == "sbert"
 Provides-Extra: openai
 Requires-Dist: openai>=1.0.0; extra == "openai"
 Provides-Extra: test
 Requires-Dist: pytest; extra == "test"
 Requires-Dist: httpx; extra == "test"
+Requires-Dist: datasets>=3.5.0; extra == "test"
 Dynamic: author
 Dynamic: author-email
 Dynamic: classifier
@@ -67,7 +66,7 @@ Upload → chunk → index (with metadata) → search via REST and CLI.
 - Metadata filters on search (`{"filters": {"docid": "DOC-1"}}`)
 - REST and CLI entry points
 - Health/metrics endpoints + Prometheus exporter
-- Pluggable embeddings and stores; default backend is local
+- Pluggable embeddings and stores; default stack is local FAISS + SBERT
 
 ## Requirements
 - Python 3.10–3.14
@@ -102,27 +101,30 @@ export PATCHVEC_AUTH__GLOBAL_KEY="your-secret"
 ```
 
 ## Minimal config (optional)
-By default PatchVec runs with sensible local defaults. To customize, create
-`config.yml`:
+By default PatchVec runs with sensible local defaults. For a user install,
+customize `~/patchvec/config.yml`:
 ```yaml
 vector_store:
-  type: default
+  type: faiss
 embedder:
-  type: default
+  type: sbert
 auth:
   mode: static
   global_key: ${PATCHVEC_GLOBAL_KEY}
 ```
 Then export:
 ```bash
-export PATCHVEC_CONFIG=./config.yml
 export PATCHVEC_GLOBAL_KEY="your-secret"
 ```
+If you keep the file elsewhere, point the runtime at it explicitly:
+```bash
+export PATCHVEC_CONFIG=/path/to/config.yml
+```
 
 ## CLI example
 ```bash
 pavecli create-collection demo books
-pavecli upload demo books demo/20k_leagues.txt --docid=verne-20k \
+pavecli ingest demo books demo/20k_leagues.txt --docid=verne-20k \
   --metadata='{"lang":"en"}'
 pavecli search demo books "captain nemo" -k 5
 ```