Configure your first proxy from the dashboard.

Step 1

Download the proxy and copy your token

Sign in to the customer portal, download the covered mw-proxy build from Dashboard → Downloads, and copy the current token from Dashboard → Licenses.

chmod +x ./mw-proxy

The customer portal manages purchases, downloads, and issued tokens. Configuration happens in the separate dashboard served by the local proxy.

Step 2

Run ModelWarden

./mw-proxy

With no flags, ModelWarden uses the platform config directory. If the config does not exist, it creates a valid starter file automatically and starts on 127.0.0.1:3758.

• Linux: ~/.config/modelwarden/config.json or $XDG_CONFIG_HOME/modelwarden/config.json
• macOS: ~/Library/Application Support/modelwarden/config.json
• Windows: %APPDATA%\modelwarden\config.json

Step 3

Open the local dashboard

Open http://127.0.0.1:3758/admin/, then go to Config.

1. Paste the token into License token.
2. Set Heartbeat endpoint to https://modelwarden.ai/api/v1/license/heartbeat.
3. Leave the machine ID blank on first setup; the proxy creates a stable local activation ID.
4. Save the configuration and confirm the license card shows the expected tier and entitlements.

Dashboard saves are validated against the config schema and your license before being written to disk.

Step 4

Add an upstream provider and route

Open Providers → Add Provider and enter:

• A unique provider name.
• The upstream API endpoint, provider format, and API key.
• An optional fixed upstream model.
• The route/model key your client will request, such as gpt-5.4 or claude-sonnet.

Every tier can create a basic one-provider route. Professional and Enterprise can add more providers to the same route and manage ordering and failover from Routing.

If you want contextual judge review, open Judge, add a judge provider from the configured provider inventory, and turn off Heuristics only under Config.

Step 5

Connect your client

The Providers and Routing pages can copy configuration using the routes currently saved by the proxy. The examples below assume a route named gpt-5.4 and an optional proxy API key exposed as MODELWARDEN_API_KEY.

{
  "provider": {
    "modelwarden": {
      "npm": "@ai-sdk/openai-compatible",
      "options": {
        "baseURL": "http://127.0.0.1:3758/v1",
        "apiKey": "{env:MODELWARDEN_API_KEY}"
      },
      "models": {
        "gpt-5.4": { "name": "gpt-5.4" }
      }
    }
  }
}

export ANTHROPIC_BASE_URL=http://127.0.0.1:3758
export ANTHROPIC_API_KEY="$MODELWARDEN_API_KEY"
claude --model gpt-5.4

model = "gpt-5.4"
model_provider = "modelwarden"

[model_providers.modelwarden]
name = "ModelWarden"
base_url = "http://127.0.0.1:3758/v1"
env_key = "MODELWARDEN_API_KEY"
wire_api = "responses"

Step 6

Verify the full path

curl http://127.0.0.1:3758/health
curl http://127.0.0.1:3758/v1/models

Send one normal client request, then confirm it appears under Audit. Use Providers to check upstream health and Debug on entitled profiles to test a tool call before enforcing it.

Deploying more than one machine?

Use the Enterprise fleet configuration workflow for explicit config paths, systemd, permissions, rollout checks, and rollback.