Add option FOUNDRY_HOST in order to allow exposing port 5272 to the network

[elsewhat]

Foundry currently binds to port 5272 only to 127.0.0.1

>netstat -an | find "5272"
  TCP    127.0.0.1:5272         0.0.0.0:0              LISTENING
  TCP    [::1]:5272             [::]:0                 LISTENING

This makes it challenging to connect to the REST API from WSL2 and from local devcontainers docker as the port is not exposed to them.
For example the following curl call from WSL2 gets blocked

curl --verbose http://$(hostname).local:11434/v1/chat/completions \
-H "Content-Type: application/json" \
-d '{
  "model": "phi3:latest",
  "messages": [{"role": "user", "content": "What is the capital of France?"}],
  "temperature": 0.7,
  "max_tokens": 50
}'

When using Ollama, I and many others typically develop in docker devcontainers and connect to Ollama running on windows.

Ollama provides option OLLAMA_HOST. When this is set to 0.0.0.0 through set OLLAMA_HOST="0.0.0.0" it exposes the port to the network. Firewalls on the machine will prevent it to be exposed further

>netstat -an | find "11434"
  TCP    0.0.0.0:11434          0.0.0.0:0              LISTENING
  TCP    [::]:11434             [::]:0                 LISTENING

Suggestion is to add FOUNDRY_HOST which does the same or add some otherway of configuring the ip-address foundry is bound to.

[skottmckay]

Does service set --host 0.0.0.0 achieve what you need?

[SorraTheOrc]

There is no such parameter

[filipw]

I ran into the same problem - also trying to access the model from WSL.

Explicitly setting host as 0.0.0.0 seems promising

PS C:\Users\filip> foundry service set --host 0.0.0.0
Saving new settings
Restarting service...
🔴 Service is stopped.
🟢 Service is Started on http://0.0.0.0:5272, PID 10428!
PS C:\Users\filip> netstat -an | findstr :5272
  TCP    0.0.0.0:5272           0.0.0.0:0              LISTENING

however then the model loading fails:

PS C:\Users\filip> foundry model load deepseek-r1-7b-npu
🕛 Loading model... Exception: Request to local service failed. Uri:http://0.0.0.0:5272/openai/load/qnn-deepseek-r1-distill-qwen-7b?ttl=600
IPv4 address 0.0.0.0 and IPv6 address ::0 are unspecified addresses that cannot be used as a target address. (Parameter 'hostName') (0.0.0.0:5272)
Please check service status with 'foundry service status'.

[HaoHoo]

Yes I got the 503...

[HaoHoo]

The host setting of the service in 0.2.9261 is limited to 127.0.0.1, etc...
But we need the service face to 0.0.0.0 😄

[ioanc]

It is working for me to connect to the service from wls2, with the default configuration

[jr-MS]

Hi team, any update on this issue? This functionality is really important for us to expose the foundry local capabilities to the computers within private network. And this command doesn't work for me.
foundry service set --host 0.0.0.0

[SorraTheOrc]

This should still be an active issue, not a discussion. At present it is not possible to use Foundry Local within WSL, without an intermediary service. Coupled with the fact NPU drivers are not available in WSL we remain stuck in an environment that is lacking much of the AI ecosystem.