Continue

The VS Code and JetBrains assistant, with Tium as an openai-provider model block.

Setup

Point Continue at Tium

You need the base URL https://api.tium.ai/v1 and an API key from your dashboard. Keys are shown once at creation.

  1. 1.Add the model block to config.yaml

    provider: openai selects the wire format, not the vendor. It is what routes the request through Continue's OpenAI-compatible path. name, version and schema are the three required top-level fields.

    config.yaml
    name: Tium
    version: 0.0.1
    schema: v1
    
    models:
      - name: GLM-5.3 Flash
        provider: openai
        model: glm-5.3-flash
        apiBase: https://api.tium.ai/v1
        apiKey: ${TIUM_API_KEY}
        capabilities:
          - tool_use
  2. 2.Add one entry per model you want offered

    Any slug from the catalog goes in the model field. Each needs its own entry in the models list; there is no wildcard.

Worth knowing

What Continue gets wrong

Tool-specific behaviour that surprises people, and what to do about it.

apiBase must include /v1

Continue's own documentation is explicit that apiBase points at an endpoint with the /v1 path included. Dropping it is the same 404 that catches people in Jan and LibreChat.

Declare tool_use or Agent mode will not offer the model

Continue autodetects capabilities from the model name, and it has never heard of ours, so it detects nothing. capabilities is ADDITIVE, and their docs are clear you cannot override autodetection, only add to it, so listing tool_use is what makes the model available in Agent mode. Add image_input as well for glm-5.3-flash or kimi-k3, which take images; the other three do not.

The default roles do not include autocomplete

Omitting roles gives you chat, edit, apply and summarize, which is what this block does deliberately. Autocomplete is a fill-in-the-middle job rather than a chat one, and we have not verified FIM behaviour on these models, so this page does not tell you to switch it on.

Elsewhere

Other tools

The same base URL and key work anywhere that takes an OpenAI-compatible endpoint.