Continue with EXAIV: Custom Model Configuration for the Open-Source Coding Assistant
Continue is the open-source AI coding assistant for VSCode and JetBrains, and it lets you declare any OpenAI-compatible endpoint as a model source. This guide covers the config file, model fields, verification steps, and how it compares with the Cursor route.
Continue with EXAIV: Custom Model Configuration for the Open-Source Coding Assistant
Continue is one of the most popular open-source AI coding assistants for VSCode and JetBrains, and its defining trait is "you configure everything" — models are declared in a config file, which naturally supports OpenAI-compatible endpoints. Point it at EXAIV and your IDE gains a coding assistant running on domestic models (Kimi, GLM, and more), with configuration living in a local file you can version-control.
Setup steps
Install the extension
Search for Continue in the VSCode marketplace (or JetBrains plugin marketplace) and install it.Open the config file
Click the gear icon in the Continue panel to open the config file (usually under the .continue folder in your home directory).Add model entries
Add an OpenAI-compatible model entry per the sample below, filling in apiBase, apiKey, and model.Save and verify
Save the config and start a chat in the Continue panel to confirm connectivity.
Config sample (YAML):
models:
- name: Kimi K3 (EXAIV)
provider: openai
model: kimi-k3
apiBase: https://exaiv.com/api/ai/v1
apiKey: YOUR_EXAIV_KEY
roles:
- chat
- name: GLM (EXAIV)
provider: openai
model: glm-5.3-flash
apiBase: https://exaiv.com/api/ai/v1
apiKey: YOUR_EXAIV_KEY
roles:
- chat
- edit
model must be a real catalog ID
The model field must be a real ID from EXAIV's /models catalog (like kimi-k3). Do not enter HuggingFace weight names. Available IDs follow the live catalog.
Verification and debugging
| Symptom | Check | Fix |
|---|---|---|
| 401 | Is the apiKey correct | Regenerate the key |
| 404 | Does apiBase end at /api/ai/v1 | Correct the path |
| Model not found | The model field value | Verify against /models |
| Chat works but completion does not | Do the roles include that capability | Add the right roles to the model |
Which model handles which capability
Continue can assign different models to different roles (chat / edit / autocomplete). A common pattern: a capable model for chat and edits, a separate lightweight model for autocomplete — completion requests are frequent, so a lightweight model is friendlier on both cost and latency.
Continue versus the Cursor route
| Dimension | Continue | Cursor |
|---|---|---|
| Openness | Fully open-source | Commercial product |
| Configuration | Local config file, versionable | Graphical interface |
| IDE support | VSCode + JetBrains | Standalone editor |
| Time to first run | Requires writing config | Faster via GUI |
| Best for | Teams wanting config control across IDEs | Teams wanting out-of-the-box feel |
Both support custom OpenAI-compatible endpoints. Choose by team habit: config control favors Continue, experience favors Cursor.
FAQ
Where is Continue's config file?
Can autocomplete use EXAIV?
Does the config support JSON as well as YAML?
Can I commit the config file to a team repository?
Where is pricing?
Related reading
- Cursor with EXAIV — the GUI-based alternative
- Qwen Local Coding Workflow — local coding assistant setups
- Kimi K3 and GLM Integration Guide — the full API tutorial
- Qwen Cost Optimization — cost control for high-frequency completion requests
- Pricing — live pricing and plans