API troubleshooting

First check your KeenKind API key, endpoint, model ID, and tool version. Save the error code, time, and request ID when available.

Find the cause by error

Symptom or code	First step
401 / 403	Check that the key is complete and valid and that the account can use the model. A website login cookie cannot replace an API key.
404 / model not found	Check the full request path and duplicated /v1, then confirm the actual ID in the model catalog.
INSUFFICIENT_CREDITS	Check available balance or credit in Balance and top-ups.
402 / API_KEY_QUOTA_EXCEEDED	Check the key’s spending limit in API keys, even if the account has balance.
429 / rate_limit_exceeded	Reduce concurrency and wait before retrying. Follow Retry-After when provided.
400 / context_length_exceeded	Shorten history and attachments or reduce maximum output, then try a smaller task.
400 / 422 parameter error	Restore default parameters and confirm support for reasoning intensity, images, or other extensions.
503 / UPSTREAM_AUTHENTICATION_FAILED	Contact support to check the server channel. Replacing your own valid key repeatedly will not resolve it.
503 / UPSTREAM_ENDPOINT_UNAVAILABLE	Save the error code and request ID and contact support to check the upstream endpoint.

Timeout, interrupted stream, or incomplete answer

1. Open usage records and check requests for the relevant time and model.
2. Check whether the tool is still running before submitting the same task again.
3. Save the error, time, and request ID. Ask support to investigate if the outcome is uncertain.
4. Once a retry is appropriate, verify with a smaller task.

A connection failure does not mean the server did not execute the request. Partial output may be billed for confirmed usage.

Chat works, but files or code tasks fail

Distinguish the model connection from local tool execution:

• Confirm the model and API support tool calling.
• Open a project in the coding tool and allow access to the relevant files.
• Expand the failed tool call to distinguish a model parameter error from a local permission, command, or file-path error.
• First verify a task that reads the README, then try edits or long runs.

Requests still use the old service

Select the new provider and model. If using environment variables, launch the tool from the terminal where you set them. An already-running app usually does not inherit a new terminal’s environment. See Tool setup.

What to send to support

Use the support entry and include:

Tool name and version:
Operating system:
Time, including time zone:
Model ID and protocol:
Base URL:
HTTP status / error code:
Request ID, if available:
Steps to reproduce:

The request ID is usually available in the X-Inference-Request-ID response header. If your tool does not show it, provide the time instead. Remove API keys, cookies, and private project content from screenshots or logs.