Integrations
Continue
Continue in VS Code and JetBrains with AI Tokens: a config.yaml with the base URL, the key in a .env file, the context window and tools for Agent mode.
Last updated: 2026-10-11
Settings
Continue reads its models from config.yaml (in ~/.continue/). It has no
provider of its own for AI Tokens: use provider: openai with apiBase.
name: AI Tokens
version: 0.0.1
schema: v1
models:
- name: Qwen3.8 27B
provider: openai
model: qwen3.8-27b
apiBase: https://api.aitokens.ch/v1
apiKey: ${{ secrets.API_KEY }}
roles: [chat, edit, apply]
capabilities: [tool_use]
defaultCompletionOptions:
contextLength: 262144
maxTokens: 8192
- name: Qwen3-Coder 30B
provider: openai
model: qwen3-coder-30b
apiBase: https://api.aitokens.ch/v1
apiKey: ${{ secrets.API_KEY }}
roles: [chat, edit, apply]
capabilities: [tool_use]
defaultCompletionOptions:
contextLength: 32768
maxTokens: 8192
${{ secrets.API_KEY }} is read from a .env file: ~/.continue/.env, or
.continue/.env and .env in your project, which come first.
API_KEY=sk-at-…
The extension does not read your shell's environment variables, so
export API_KEY=… in a terminal is not enough.
capabilities: [tool_use] is what turns on Agent mode for a model that
Continue does not know. Add image_input only on a model that reads images,
such as qwen3.6-35b.
Reasoning off
qwen3.8-27b reasons before every answer. For quick edits you can add a second
entry of the same model with reasoning switched off, through
requestOptions.extraBodyProperties:
- name: Qwen3.8 27B, no reasoning
provider: openai
model: qwen3.8-27b
apiBase: https://api.aitokens.ch/v1
apiKey: ${{ secrets.API_KEY }}
roles: [edit, apply]
defaultCompletionOptions:
contextLength: 262144
maxTokens: 8192
requestOptions:
extraBodyProperties:
reasoning_effort: none
Limits
- No autocomplete. Continue's tab autocomplete needs fill-in-the-middle,
which AI Tokens does not serve. Do not give these models the
autocompleterole. topKis refused. The API does not knowtop_kand answers400: leavetopKout ofdefaultCompletionOptions.maxTokensup to 8192. A higher value is refused with422.