If Frigate and Ollama run in separate Docker containers on the same ZimaOS host, localhost:11434 inside Frigate points back to the Frigate container—not to Ollama. Put both services on a shared Docker network and use the Ollama service name, or publish Ollama to a host address that Frigate can reach.
The source thread never reached a verified final configuration, but it did establish the correct diagnostic path: first prove network reachability from the Frigate container, then troubleshoot Frigate's genai: configuration. Current Frigate documentation also requires a vision-capable model for image/event descriptions.
Why localhost Fails Between Containers
Every normal Docker container has its own network namespace. Inside Frigate:
http://localhost:11434means “port 11434 inside the Frigate container.” It does not mean the ZimaOS host and does not automatically resolve to the Ollama container.
Best Setup: Shared Docker Network
Put Frigate and Ollama on the same user-defined Docker network and use a stable service/container name such as:
genai:
provider: ollama
base_url: http://ollama:11434
model: qwen3-vl:4bThe current Frigate Generative AI guide documents the Ollama provider and requires a suitable vision model for GenAI descriptions.
Step 1: Check the Containers and Networks from the ZimaOS Host
Run this from the ZimaOS host terminal, not from inside Frigate:
docker ps --format 'table {.Names} {.Image} {.Networks}' | grep -Ei 'NAMES|frigate|ollama'The source user accidentally ran a Docker command inside the Frigate container and received docker: not found. That is expected: containers normally do not contain or control the host Docker CLI.
Step 2: Test Ollama from Inside Frigate
Once you know the container names/networks, test the actual HTTP path. From the host:
docker exec -it frigate sh
wget -qO- http://ollama:11434/api/tagsIf the response lists Ollama models, networking works. If it cannot resolve ollama, the containers probably do not share the same network. If it resolves but connection is refused, check Ollama listening/binding and port state.
Alternative: Use the ZimaOS Host IP
If Ollama publishes port 11434 to the host, Frigate can use a LAN-reachable host address such as:
base_url: http://192.168.1.20:11434This is simple but routes traffic out through the host instead of using Docker service discovery. Make sure Ollama is actually listening on an interface reachable from the Frigate container.
Use a Vision-Capable Ollama Model
Network connectivity alone is not enough. Frigate's current GenAI feature sends visual context, so the configured model must support vision. A text-only Ollama model can respond to API tests yet still be unsuitable for camera-event descriptions.
Do Not Put host.docker.internal Everywhere
host.docker.internal is convenient on Docker Desktop but is not guaranteed to behave identically on every Linux Docker host. A shared user-defined network plus service name is usually more portable on ZimaOS.
Verify the Frigate Configuration After Networking Works
Only after /api/tags succeeds should you focus on YAML indentation, provider name, model name, and GenAI feature settings. Restart Frigate and inspect logs for provider initialization.
The Docker networking guide provides the broader container-networking model.
FAQ
Why doesn't localhost:11434 work from Frigate?
Because localhost inside Frigate is the Frigate container itself, not the Ollama container.
What is the best Ollama URL?
On a shared Docker network, use the Ollama service/container name such as http://ollama:11434.
Why does docker exec say docker is not found?
You likely ran Docker CLI inside a container. Run Docker management commands from the ZimaOS host terminal.
Can any Ollama model work with Frigate GenAI?
No. Current Frigate documentation says the GenAI workflow needs a vision-capable model.
