Use sqlite-vec extension instead of chromadb for embeddings

docker/main/Dockerfile

@@ -30,6 +30,16 @@ RUN --mount=type=tmpfs,target=/tmp --mount=type=tmpfs,target=/var/cache/apt \
     --mount=type=cache,target=/root/.ccache \
     /deps/build_nginx.sh
 
+FROM wget AS sqlite-vec
+ARG DEBIAN_FRONTEND
+
+# Build sqlite_vec from source
+COPY docker/main/build_sqlite_vec.sh /deps/build_sqlite_vec.sh
+RUN --mount=type=tmpfs,target=/tmp --mount=type=tmpfs,target=/var/cache/apt \
+    --mount=type=bind,source=docker/main/build_sqlite_vec.sh,target=/deps/build_sqlite_vec.sh \
+    --mount=type=cache,target=/root/.ccache \
+    /deps/build_sqlite_vec.sh
+
 FROM scratch AS go2rtc
 ARG TARGETARCH
 WORKDIR /rootfs/usr/local/go2rtc/bin
@@ -163,7 +173,7 @@ RUN wget -q https://bootstrap.pypa.io/get-pip.py -O get-pip.py \
 COPY docker/main/requirements.txt /requirements.txt
 RUN pip3 install -r /requirements.txt
 
-# Build pysqlite3 from source to support ChromaDB
+# Build pysqlite3 from source
 COPY docker/main/build_pysqlite3.sh /build_pysqlite3.sh
 RUN /build_pysqlite3.sh
 
@@ -177,6 +187,7 @@ RUN pip3 wheel --no-deps --wheel-dir=/wheels-post -r /requirements-wheels-post.t
 # Collect deps in a single layer
 FROM scratch AS deps-rootfs
 COPY --from=nginx /usr/local/nginx/ /usr/local/nginx/
+COPY --from=sqlite-vec /usr/local/lib/ /usr/local/lib/
 COPY --from=go2rtc /rootfs/ /
 COPY --from=libusb-build /usr/local/lib /usr/local/lib
 COPY --from=tempio /rootfs/ /
@@ -197,12 +208,11 @@ ARG APT_KEY_DONT_WARN_ON_DANGEROUS_USAGE=DontWarn
 ENV NVIDIA_VISIBLE_DEVICES=all
 ENV NVIDIA_DRIVER_CAPABILITIES="compute,video,utility"
 
-# Turn off Chroma Telemetry: https://docs.trychroma.com/telemetry#opting-out
-ENV ANONYMIZED_TELEMETRY=False
-# Allow resetting the chroma database
-ENV ALLOW_RESET=True
 # Disable tokenizer parallelism warning
+# https://stackoverflow.com/questions/62691279/how-to-disable-tokenizers-parallelism-true-false-warning/72926996#72926996
 ENV TOKENIZERS_PARALLELISM=true
+# https://github.com/huggingface/transformers/issues/27214
+ENV TRANSFORMERS_NO_ADVISORY_WARNINGS=1
 
 ENV PATH="/usr/local/go2rtc/bin:/usr/local/tempio/bin:/usr/local/nginx/sbin:${PATH}"
 ENV LIBAVFORMAT_VERSION_MAJOR=60

docker/main/build_sqlite_vec.sh

@@ -0,0 +1,31 @@
+#!/bin/bash
+
+set -euxo pipefail
+
+SQLITE_VEC_VERSION="0.1.3"
+
+cp /etc/apt/sources.list /etc/apt/sources.list.d/sources-src.list
+sed -i 's|deb http|deb-src http|g' /etc/apt/sources.list.d/sources-src.list
+apt-get update
+apt-get -yqq build-dep sqlite3 gettext git
+
+mkdir /tmp/sqlite_vec
+# Grab the sqlite_vec source code.
+wget -nv https://github.com/asg017/sqlite-vec/archive/refs/tags/v${SQLITE_VEC_VERSION}.tar.gz
+tar -zxf v${SQLITE_VEC_VERSION}.tar.gz -C /tmp/sqlite_vec
+
+cd /tmp/sqlite_vec/sqlite-vec-${SQLITE_VEC_VERSION}
+
+mkdir -p vendor
+wget -O sqlite-amalgamation.zip https://www.sqlite.org/2024/sqlite-amalgamation-3450300.zip
+unzip sqlite-amalgamation.zip
+mv sqlite-amalgamation-3450300/* vendor/
+rmdir sqlite-amalgamation-3450300
+rm sqlite-amalgamation.zip
+
+# build loadable module
+make loadable
+
+# install it
+cp dist/vec0.* /usr/local/lib
+

docker/main/requirements-wheels.txt

@@ -2,6 +2,7 @@ click == 8.1.*
 # FastAPI
 starlette-context == 0.3.6
 fastapi == 0.115.0
+uvicorn == 0.30.*
 slowapi == 0.1.9
 imutils == 0.5.*
 joserfc == 1.0.*
@@ -32,12 +33,12 @@ unidecode == 1.3.*
 # OpenVino (ONNX installed in wheels-post)
 openvino == 2024.3.*
 # Embeddings
-chromadb == 0.5.7
+transformers == 4.45.*
 onnx_clip == 4.0.*
 # Generative AI
-google-generativeai == 0.6.*
-ollama == 0.2.*
-openai == 1.30.*
+google-generativeai == 0.8.*
+ollama == 0.3.*
+openai == 1.51.*
 # push notifications
 py-vapid == 1.9.*
 pywebpush == 2.0.*

docker/main/rootfs/etc/s6-overlay/s6-rc.d/log-prepare/run

@@ -4,7 +4,7 @@
 
 set -o errexit -o nounset -o pipefail
 
-dirs=(/dev/shm/logs/frigate /dev/shm/logs/go2rtc /dev/shm/logs/nginx /dev/shm/logs/certsync /dev/shm/logs/chroma)
+dirs=(/dev/shm/logs/frigate /dev/shm/logs/go2rtc /dev/shm/logs/nginx /dev/shm/logs/certsync)
 
 mkdir -p "${dirs[@]}"
 chown nobody:nogroup "${dirs[@]}"

docs/docs/configuration/semantic_search.md

@@ -5,7 +5,7 @@ title: Using Semantic Search
 
 Semantic Search in Frigate allows you to find tracked objects within your review items using either the image itself, a user-defined text description, or an automatically generated one. This feature works by creating _embeddings_ — numerical vector representations — for both the images and text descriptions of your tracked objects. By comparing these embeddings, Frigate assesses their similarities to deliver relevant search results.
 
-Frigate has support for two models to create embeddings, both of which run locally: [OpenAI CLIP](https://openai.com/research/clip) and [all-MiniLM-L6-v2](https://huggingface.co/sentence-transformers/all-MiniLM-L6-v2). Embeddings are then saved to a local instance of [ChromaDB](https://trychroma.com).
+Frigate has support for two models to create embeddings, both of which run locally: [OpenAI CLIP](https://openai.com/research/clip) and [all-MiniLM-L6-v2](https://huggingface.co/sentence-transformers/all-MiniLM-L6-v2). Embeddings are then saved to Frigate's database.
 
 Semantic Search is accessed via the _Explore_ view in the Frigate UI.
 
@@ -29,7 +29,7 @@ If you are enabling the Search feature for the first time, be advised that Friga
 
 ### OpenAI CLIP
 
-This model is able to embed both images and text into the same vector space, which allows `image -> image` and `text -> image` similarity searches. Frigate uses this model on tracked objects to encode the thumbnail image and store it in Chroma. When searching for tracked objects via text in the search box, Frigate will perform a `text -> image` similarity search against this embedding. When clicking "Find Similar" in the tracked object detail pane, Frigate will perform an `image -> image` similarity search to retrieve the closest matching thumbnails.
+This model is able to embed both images and text into the same vector space, which allows `image -> image` and `text -> image` similarity searches. Frigate uses this model on tracked objects to encode the thumbnail image and store it in the database. When searching for tracked objects via text in the search box, Frigate will perform a `text -> image` similarity search against this embedding. When clicking "Find Similar" in the tracked object detail pane, Frigate will perform an `image -> image` similarity search to retrieve the closest matching thumbnails.
 
 ### all-MiniLM-L6-v2
 

frigate/api/app.py

@@ -384,12 +384,12 @@ def vainfo():
 
 @router.get("/logs/{service}", tags=[Tags.logs])
 def logs(
-    service: str = Path(enum=["frigate", "nginx", "go2rtc", "chroma"]),
+    service: str = Path(enum=["frigate", "nginx", "go2rtc"]),
     download: Optional[str] = None,
     start: Optional[int] = 0,
     end: Optional[int] = None,
 ):
-    """Get logs for the requested service (frigate/nginx/go2rtc/chroma)"""
+    """Get logs for the requested service (frigate/nginx/go2rtc)"""
 
     def download_logs(service_location: str):
         try:
@@ -408,7 +408,6 @@ def download_logs(service_location: str):
         "frigate": "/dev/shm/logs/frigate/current",
         "go2rtc": "/dev/shm/logs/go2rtc/current",
         "nginx": "/dev/shm/logs/nginx/current",
-        "chroma": "/dev/shm/logs/chroma/current",
     }
     service_location = log_locations.get(service)