Fresh 2026

Continue.dev

How to add Gateway model entries to Continue's config.yaml for chat, edit, and apply flows.

Continue.dev accepts any OpenAI-compatible endpoint via a config.yaml model entry. Add one block pointed at Merge Gateway and every Gateway model, including OpenAI, Anthropic, Google, and Kimi, becomes available across Continue's chat, edit, and apply flows in VS Code or JetBrains.

Before you start

• Grab an API key from gateway.merge.dev/settings/api-keys
• Decide which Continue roles (chat, edit, apply, summarize, autocomplete, embed, rerank) you want each model to serve

Configure Continue

Open ~/.continue/config.yaml (create it if it doesn't exist). In VS Code, you can also open it from the Continue side-panel → gear icon → Configure.

Each Gateway model is one entry under models:. Use provider: openai (Continue's OpenAI-compatible driver) with apiBase pointed at Gateway.

name: My Config
version: 0.0.1
schema: v1
models:
  - name: Claude Opus 4.6 (Merge)
    provider: openai
    model: anthropic/claude-opus-4-6
    apiBase: https://api-gateway.merge.dev/v1/openai
    apiKey: ${{ secrets.MERGE_GATEWAY_API_KEY }}
    roles:
      - chat
      - edit
      - apply
    capabilities:
      - tool_use
      - image_input

  - name: GPT-5.2 (Merge)
    provider: openai
    model: openai/gpt-5.2
    apiBase: https://api-gateway.merge.dev/v1/openai
    apiKey: ${{ secrets.MERGE_GATEWAY_API_KEY }}
    roles:
      - chat
      - edit
      - apply
    capabilities:
      - tool_use
      - image_input

Continue supports ${{ secrets.NAME }} interpolation. Set MERGE_GATEWAY_API_KEY in the Continue Hub secrets (or inline as a literal for local-only setups).

Reload the Continue extension (VS Code: command palette → Developer: Reload Window; JetBrains: restart the IDE). Open the Continue panel, pick one of your new Gateway models, and send a test message. The request will show up in your Gateway dashboard.

Caveats

Continue's roles list is a hint. Actual behavior depends on the model's capabilities on the vendor route Gateway picks. Before assigning a model to autocomplete, confirm it's a fast model (usually a smaller one) and that the vendor supports streaming.

Add tool_use to capabilities only if vendors..capabilities.supports_tool_calling is true for the route Gateway will use. Check via GET /v1/models.

You can wire dedicated models for autocomplete (e.g. a small fast model), embed, or rerank as separate entries. Continue will use each model for the role it's assigned to. Gateway's /v1/openai/embeddings endpoint is available for the embed role.

To pin a routing project, use Continue's requestOptions.extraBodyProperties to send project_id in the request body. To attach tags, use requestOptions.headers to send X-Merge-Tags.

    requestOptions:
      extraBodyProperties:
        project_id: prj_abc123

Next steps

Let Gateway pick the best vendor for each Continue request

Attach Continue usage to a specific project

Cap spend and restrict which models Continue can hit