Troubleshooting

401 INVALID_API_KEY

The assigned route has no eligible model for the requested capability or response_profile
Open AI Providers to verify provider connection health, and Routes to check model coverage for the endpoint
Try a simpler request (plain chat without tools) to isolate capability mismatch
Remember that vision, tools, audio, and video each require their own model capability on the route — see the capability matrix in Models & providers

Add credits or switch the route to BYOK
Free workspaces may have managed onboarding caps for chat, image, and embeddings

Expected behavior. Remove model, provider, route, and policy from SDK calls.

Run one of:

latentkit login

or:

export LATENTKIT_API_KEY="lk-..."

For CI and AI agents, prefer LATENTKIT_API_KEY. For humans, latentkit login opens the browser approval flow.

Logs, traces, usage, costs, providers, routes, keys, and connections are workspace-scoped commands. Pass your workspace id:

latentkit logs --tenant t_123 --days 1

Use latentkit whoami --json and latentkit apps --json to inspect the account/app context.

Your login may still be valid, but the current session does not include the account context needed by that command.

Update the CLI to the latest version.
Run latentkit logout, then latentkit login again.
Confirm that the expected workspace app is selected with latentkit apps.
If the account is still unknown, contact support and include the CLI version; do not send the local profile file or access token.

Do not use NEXT_PUBLIC_LATENTKIT_API_KEY, VITE_LATENTKIT_API_KEY, or frontend env prefixes for secrets
Move the LatentKit call into a backend route, server action, serverless function, Laravel controller, FastAPI route, or similar server-side layer
The browser should call your app endpoint, for example /api/chat

Confirm LATENTKIT_API_KEY is present in the server environment; the package reads it through config/latentkit.php
Clear and rebuild cached configuration after changing environment values: php artisan config:clear && php artisan config:cache
Run php artisan latentkit:check to inspect the connected app, route, credits, and latest request without printing the key
Log the normalized exception request ID when a direct SDK or Laravel AI request fails; do not log provider response bodies

Use with LatentKit(...) or async with AsyncLatentKit(...) so the underlying HTTP client closes cleanly
Set a request timeout at your framework boundary if the request can be user-triggered
Catch LatentKitAPIError and log exc.status_code, exc.code, and exc.request_id

Match the workspace, API key and selected range. Usage charts/history are workspace-wide; its Headline metrics selector only narrows requests, spend and tokens. A key Overview covers that key's traffic.
Requests count completed calls; Recent Requests also includes errors. Upstream fallback attempts are separate trace steps, not extra customer requests. Purpose usage excludes calls without a purpose, and degraded attempts can include errors.
24h is a rolling window. 7d/30d use UTC calendar days; charts use UTC while request timestamps use your local timezone. Data outside retention is unavailable.
BYOK spend is an estimate paid to your provider; Platform Access customer charges deduct LatentKit credits. Provider cost estimates and customer charges are different measurements.
If provider cost was not recorded, the chart reports unavailable coverage rather than claiming zero cost. Partial charts show recorded estimates only; a missing latency sample is a gap.
Recent Requests starts with 10 rows. Use Load more for older rows, or Refresh requests for a new snapshot. Filters search the full selected range, and a loading failure offers Retry.