Common issues with Ollama Chat Model in Duaer

Expressions resolve to the first item

Ollama Chat Model is a sub-node. Most nodes resolve an expression once per input item; Duaer calls a sub-node with the first item only. Given five names, {{ $json.name }} is always the first name.

To use a different value per item, put the expression in the root node (such as the AI Agent prompt), or run Loop Over Items first so each run gets one item.

Cannot connect to a remote Ollama instance

Ollama Chat Model connects to the Base URL in the Ollama credential (default http://localhost:11434). For a remote Ollama behind an authenticating proxy, such as Open WebUI, set Base URL to the remote address and API Key to the proxy key. Duaer sends it as a Bearer token.

Cannot connect to a local Ollama instance from Docker

Each Docker container has its own localhost. If Duaer or Ollama runs in a container, they cannot reach each other on localhost. Set Base URL in the Ollama credential by setup:

• Only Ollama in Docker: publish the port with -p 11434:11434 and use http://localhost:11434.
• Only Duaer in Docker: have the host Ollama listen on 0.0.0.0 and use http://host.docker.internal:11434. On Linux, start the Duaer container with --add-host host.docker.internal:host-gateway; Docker Desktop does this for you.
• Separate containers: put both on one Docker network and use the Ollama container name, such as http://my-ollama:11434.
• Same container: use http://localhost:11434.

Error: connect ECONNREFUSED ::1:11434

localhost resolved to IPv6 while Ollama listens on IPv4 only. Set Base URL to http://127.0.0.1:11434.

Ollama and HTTP/HTTPS proxies

Ollama Chat Model does not support a custom HTTP proxy, so HTTP_PROXY or HTTPS_PROXY may have no effect. If you need a proxy, refer to Ollama’s proxy guide, or let Duaer reach Ollama directly.