Coding agents and IDEs

Continue.dev

Route Continue's chat, edit, and apply requests through Merge Gateway from VS Code or JetBrains

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. An organization key from gateway.merge.dev/api-keys works everywhere; a project API key, created from a project’s API keys tab, additionally scopes every request to that project so its routing policy and budget apply automatically. See Projects.
  • Decide which Continue roles (chat, edit, apply, summarize, autocomplete, embed, rerank) you want each model to serve

Configure Continue

1

Open your Continue config

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.

2

Add a Gateway model entry

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

~/.continue/config.yaml
1name: My Config
2version: 0.0.1
3schema: v1
4models:
5  - name: Claude Opus 5 (Merge)
6    provider: openai
7    model: anthropic/claude-opus-5
8    apiBase: https://api-gateway.merge.dev/v1/openai
9    apiKey: ${{ secrets.MERGE_GATEWAY_API_KEY }}
10    roles:
11      - chat
12      - edit
13      - apply
14    capabilities:
15      - tool_use
16      - image_input
17
18  - name: GPT-5.5 (Merge)
19    provider: openai
20    model: openai/gpt-5.5
21    apiBase: https://api-gateway.merge.dev/v1/openai
22    apiKey: ${{ secrets.MERGE_GATEWAY_API_KEY }}
23    roles:
24      - chat
25      - edit
26      - apply
27    capabilities:
28      - tool_use
29      - 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).

3

Reload Continue and select the model

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.<vendor>.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.

Add a model entry with model: default_routing and Gateway hands each request to your routing policy, which picks the vendor and model per request. The policy comes from the key’s project (project API key) or the org default; if neither has a policy, requests fail with model_required.

The simplest way to pin all Continue usage to a project is a project API key. To switch projects per model entry instead, use Continue’s requestOptions.extraBodyProperties to send project_id (the project’s UUID) in the request body. Tags ride along the same way.

1    requestOptions:
2      extraBodyProperties:
3        project_id: 4f8b2c6e-9d1a-4e7a-b3f5-2c8d0a6e1b47
4        tags:
5          - key: team
6            value: platform

Next steps

Routing policies

Let Gateway pick the best vendor for each Continue request

Projects

Attach Continue usage to a specific project

Cost governance

Cap spend and restrict which models Continue can hit