Use Cara with VS Code, JetBrains, chat UIs, and scripts

Outcome

Run Cara locally with OpenAI-style endpoints enabled, then connect:

• VS Code / JetBrains (via Continue)
• Open WebUI
• LibreChat
• Scripts and automations (curl)

Prerequisites

• cara installed.
• ANTHROPIC_API_KEY or OPENAI_API_KEY.
• Optional client app: Continue, Open WebUI, or LibreChat.

1) Create config

Export a gateway token:

export CARAPACE_GATEWAY_TOKEN="$(openssl rand -hex 32)"

Windows (PowerShell) alternative:

$bytes = [byte[]]::new(32)
[System.Security.Cryptography.RandomNumberGenerator]::Fill($bytes)
$env:CARAPACE_GATEWAY_TOKEN = [System.BitConverter]::ToString($bytes).Replace('-', '').ToLower()

{
  "gateway": {
    "bind": "loopback",
    "port": 18789,
    "auth": {
      "mode": "token",
      "token": "${CARAPACE_GATEWAY_TOKEN}"
    },
    "openai": {
      "chatCompletions": true,
      "responses": true
    }
  },
  "anthropic": {
    "apiKey": "${ANTHROPIC_API_KEY}"
  }
}

2) Start Cara

CARAPACE_CONFIG_PATH=./carapace.json5 cara

3) Connect your app

Use these values in your app's model/provider settings:

• Provider type: OpenAI / OpenAI-compatible
• Base URL: http://127.0.0.1:18789/v1
• API key/token: same value as CARAPACE_GATEWAY_TOKEN
• Model: carapace (this is the fixed model name Cara exposes; the actual backend model is determined by your provider config)

VS Code / JetBrains (Continue)

In Continue model/provider settings, point the model to the values above.

Open WebUI / LibreChat

Add a custom OpenAI-compatible connection and use the same values above. Field names vary by UI, but the required pieces are always: base URL, API key, and model.

4) Script smoke test (curl)

Chat Completions:

curl -sS \
  -H "Authorization: Bearer ${CARAPACE_GATEWAY_TOKEN}" \
  -H "Content-Type: application/json" \
  -X POST http://127.0.0.1:18789/v1/chat/completions \
  -d '{
    "model": "carapace",
    "messages": [
      {"role":"user","content":"Say hello in one sentence."}
    ]
  }'

Responses:

curl -sS \
  -H "Authorization: Bearer ${CARAPACE_GATEWAY_TOKEN}" \
  -H "Content-Type: application/json" \
  -X POST http://127.0.0.1:18789/v1/responses \
  -d '{
    "model": "carapace",
    "input": "List three secure defaults for a local AI assistant."
  }'

5) Verify

• Your client app returns assistant output through Cara.
• curl calls return 200 JSON responses with assistant output.
• Unauthorized calls return 401 (auth enforcement working).

Next step

• Trigger Cara from other apps
• Add Discord or Telegram

Common failures and fixes

• Symptom: 404 on /v1/chat/completions or /v1/responses.
  • Fix: Ensure gateway.openai.chatCompletions / gateway.openai.responses are enabled.
• Symptom: app says endpoint not found.
  • Fix: Confirm base URL is http://127.0.0.1:18789/v1 (include /v1).
• Symptom: 401 Unauthorized.
  • Fix: Confirm bearer token matches gateway.auth.token.
• Symptom: 500 provider error.
  • Fix: Verify provider key/model config and check logs with cara logs -n 200.
• Symptom: app on another device cannot connect.
  • Fix: This recipe is local-only (127.0.0.1). If you need remote access, use a LAN/Tailnet setup with TLS and auth hardening.