JetBrains AI Assistant with Verdictan
Add your Verdictan gateway to JetBrains AI Assistant as an OpenAI-compatible provider, and the features you assign to it send their model requests through your policies. Get chat working first, then assign other features one at a time.
127.0.0.1:41002 with at least one provider target (Gateway setup for IDEs); an API token for AI Assistant (API tokens for IDEs); one model ID from the gateway's /v1/models response; a JetBrains IDE with the AI Assistant plugin.Configure JetBrains AI Assistant
- Open Settings → Tools → AI Assistant → Providers & API keys.
- Add an OpenAI-compatible endpoint with the URL
http://127.0.0.1:41002/v1. - Enter the Verdictan API token as the API key, not the gateway token or a provider key.
- Run the connection test.
- Assign a returned model to chat. Once chat works, assign it to the other features you want governed.
AI Assistant assigns custom models per feature group. Configure inline completion and next-edit models separately in Providers & API keys → AI Completion → OpenAI Compatible; they do not inherit the chat provider assignment. A completion model must support the required completion or edit-prediction protocol, and a chat model usually does not. Check that the gateway and target support that request family before enabling it. See JetBrains custom models.
Add the MCP endpoint (conditional)
MCP is optional and separate from the model connection. It adds Verdictan tools, such as history search.
Verdictan's MCP endpoint requires a Bearer token, and the JetBrains MCP guide does not describe a protected field for one. Before you add https://<published-hostname>/mcp, check that your installed version or your organization's MCP settings can send an authenticated header. A server added with only a URL cannot authenticate. See Connect an MCP client for the endpoint requirements.
Verify
-
In AI Assistant, pick the Verdictan model and send
Reply with Hello. Do not edit files or run commands. -
In a terminal, read the recent decision events:
verdictan events tail --since 10m --event-type decision -
Check that a matching event appears. Delivery can take a few seconds, so run the command again if it is not there yet.
-
If you added MCP, check that the server is connected and lists
history_search. Ask it to search the last ten minutes without changing anything. An empty result is valid when nothing matches.
Troubleshooting
| Symptom | Cause | Fix |
|---|---|---|
| The connection test fails | The URL is not http://127.0.0.1:41002/v1, or the gateway is down | Check the URL, and run curl --fail http://127.0.0.1:41002/healthz. |
401 in the connection test | The API key is the gateway token or a provider key | Enter the API token. |
| A feature doesn't list the Verdictan model | The model lacks a protocol or capability that feature needs | Pick a capable model, or leave that feature on another provider. |
| Inline completion returns nothing | The model does not support fill-in-the-middle | Assign a model that does. |
| Only chat events appear | Only features assigned to the Verdictan model use the gateway | Assign the model to every feature you want governed. |
| MCP does not authenticate | Your version has no protected Bearer header field | Use the model path only, and check newer JetBrains MCP documentation. |
For endpoint and event problems, see IDE integration troubleshooting.
Next steps
- Continue with Verdictan: Continue in JetBrains, with roles and secrets in
config.yaml. - Configure policies for IDE traffic: controls for coding assistants.