Continue
Continue (the VS Code / JetBrains assistant) reaches MartLLM through its
openai provider with a custom apiBase. Step-by-step guide with live model
picks: Continue setup.
Fastest: one command
npx uberllm init --tool continue
This connects your account in the browser, saves the key to a file only you can
read and adds two MartLLM models to ~/.continue/config.yaml, with the key in ~/.continue/.env. The file is backed up first, and
npx uberllm init --undo puts it back. Details: uberllm init. The
steps to do it by hand are below.
Setup
Put your key in ~/.continue/.env (the editor extensions do not read shell
variables):
ULL_API_KEY=ull_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
Then add a model to ~/.continue/config.yaml:
name: My Config
version: 1.0.0
schema: v1
models:
- name: DeepSeek V4.1 Flash (MartLLM)
provider: openai
model: deepseek/deepseek-v4.1-flash
apiBase: https://api.martllm.com/v1
apiKey: ${{ secrets.ULL_API_KEY }}
useResponsesApi: false
useLegacyCompletionsEndpoint: false
capabilities: [tool_use]
roles: [chat, edit, apply]
Key fields:
- provider:
openai(MartLLM speaks the OpenAI API). - model: any catalog model id, e.g.
deepseek/deepseek-v4.1-flash. - apiBase:
https://api.martllm.com/v1— overrides the default OpenAI URL. - apiKey: your
ull_key, here read from the.envsecret. - useResponsesApi: false: Continue otherwise sends
gpt-5*ando*model names to the Responses API. MartLLM serves that route too (Responses API), so this line only keeps every model on chat completions, the route this config was tested on. - useLegacyCompletionsEndpoint: false: autocomplete otherwise calls
/v1/completions, which MartLLM does not serve (404).
Reload Continue and select the model from the model dropdown.
Multiple models and autocomplete
Add one entry per model. For autocomplete, add a fast, cheap model with
roles: [autocomplete] and the same two false settings:
- name: Autocomplete (MartLLM)
provider: openai
model: qwen/qwen3-coder
apiBase: https://api.martllm.com/v1
apiKey: ${{ secrets.ULL_API_KEY }}
useResponsesApi: false
useLegacyCompletionsEndpoint: false
roles: [autocomplete]
Each call is sent to an available third-party provider for the model. See Models & routing.