IDE Integration

Nova exposes an OpenAI-compatible API on the LLM Gateway (http://localhost:8001/v1). Any tool that speaks the OpenAI chat completions protocol works out of the box.

Continue.dev (VS Code / JetBrains)

Quick start

1. Install the Continue extension
2. Open the config: Cmd+Shift+P > Continue: Open config.json
3. Add an entry to the models array:

{
  "models": [
    {
      "title": "Nova (Claude Sonnet)",
      "provider": "openai",
      "model": "claude-max/claude-sonnet-4-6",
      "apiBase": "http://localhost:8001/v1",
      "apiKey": "unused"
    }
  ]
}

apiKey is required by Continue’s JSON schema but Nova ignores its value. apiBase is the only thing that matters — it redirects traffic from api.openai.com to Nova.

Recommended model set

Add multiple entries to switch models from the Continue sidebar:

{
  "models": [
    {
      "title": "Nova - Sonnet (fast)",
      "provider": "openai",
      "model": "claude-max/claude-sonnet-4-5",
      "apiBase": "http://localhost:8001/v1",
      "apiKey": "unused"
    },
    {
      "title": "Nova - Sonnet (latest)",
      "provider": "openai",
      "model": "claude-max/claude-sonnet-4-6",
      "apiBase": "http://localhost:8001/v1",
      "apiKey": "unused"
    },
    {
      "title": "Nova - Opus (most capable)",
      "provider": "openai",
      "model": "claude-max/claude-opus-4",
      "apiBase": "http://localhost:8001/v1",
      "apiKey": "unused"
    },
    {
      "title": "Nova - GPT-4o",
      "provider": "openai",
      "model": "openai/gpt-4o",
      "apiBase": "http://localhost:8001/v1",
      "apiKey": "unused"
    }
  ]
}

Verify available models

curl http://localhost:8001/v1/models | jq '.data[].id'

Returns all registered model IDs.

With API key auth enabled

If REQUIRE_AUTH=true, create a key first:

curl -X POST http://localhost:8000/api/v1/keys \
  -H "X-Admin-Secret: your-admin-secret" \
  -H "Content-Type: application/json" \
  -d '{"name": "continue-dev", "rate_limit_rpm": 120}'

Use the returned key value (shown once) as the apiKey field in the config above.

Cursor

Same approach — Cursor supports custom OpenAI-compatible endpoints:

1. Settings > Models > Add model
2. Set Base URL to http://localhost:8001/v1
3. Set API Key to any placeholder
4. Use any Nova model ID as the model name

Aider (terminal)

aider \
  --openai-api-base http://localhost:8001/v1 \
  --openai-api-key unused \
  --model claude-max/claude-sonnet-4-6

Raw API (curl / scripts)

curl http://localhost:8001/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-max/claude-sonnet-4-6",
    "messages": [{"role": "user", "content": "Hello from Nova"}]
  }'

How it works

IDE / tool
    |  POST /v1/chat/completions  (OpenAI format)
    v
LLM Gateway  :8001
    |  translates OpenAI -> Nova internal format
    |  forwards to registered provider (Anthropic, OpenAI, Ollama, ...)
    v
Provider API
    |  response
    v
LLM Gateway
    |  translates provider response -> OpenAI format
    v
IDE / tool  <-- looks identical to talking directly to OpenAI

The translation lives in llm-gateway/app/openai_compat.py.