Connect Pi to ApiClo

Before you start

• Create a separate ApiClo API key and check balance and the selected model's current rates.
• Use the client version matching the linked documentation. Keep credentials out of project files.

Steps

1. Connection

   Set APICLO_API_KEY in the shell that launches Pi. Add the JSON provider below without overwriting your other providers.

2. Model and credentials

   Use an exact ApiClo model ID and declare image input only for a vision model. Open /model and choose apiclo/claude-sonnet-5 or your chosen entry.

3. Verification

   If an old credential overrides the new key, inspect Pi's provider-scoped authentication. Test a short reply first, then one tool call and an image in a separate request.

Examples

export APICLO_API_KEY="YOUR_APICLO_KEY"

$env:APICLO_API_KEY="YOUR_APICLO_KEY"

{
  "providers": {
    "apiclo": {
      "baseUrl": "https://apiclo.com/v1",
      "api": "openai-completions",
      "apiKey": "$APICLO_API_KEY",
      "models": [{"id": "claude-sonnet-5", "input": ["text", "image"]}]
    }
  }
}

Check the result

Send one small request and inspect ApiClo Logs for the exact model, outcome and charged usage. Test tool calls and images separately. Cache settings in the client do not prove a hit: check actual read/write token counts.

Troubleshooting

• A model can load from JSON yet remain unavailable until its credentials resolve. Unlike the OMP example, Pi's environment interpolation explicitly uses $APICLO_API_KEY.
• For 401, check the active key and credential source; for 404, check the request URL and model ID. Preserve the sanitized error body for support.

Compatibility and billing

ApiClo is an independent gateway. A client integration does not include that client's subscription or guarantee all its features. Compare ordinary input, cache reads, cache writes and output at current model rates; cache hits and savings are workload-dependent.

Sources

• Pi