Fix any error. Fast.

The `x-api-key` header you sent doesn't match an active Anthropic API key — usually because the env var isn't loaded, the key was rotated or revoked, you're using a Workspace key in the wrong workspace, or a wrong-provider key (Bedrock or Vertex) was sent to the direct Anthropic API.

Your container starts, exits with a non-zero code (or is OOMKilled), the kubelet restarts it, it exits again — repeat. After several quick failures the kubelet enters CrashLoopBackOff, an exponential delay between restart attempts so you don't melt the node.

Your application tried to open a TCP connection to Postgres and the OS rejected it — Postgres isn't listening on the host:port you specified, or a firewall blocked the connection.

Your OpenAI organisation has run out of paid credit, hit its monthly hard limit, or hasn't added a payment method yet. Despite the 429 status, this is a billing problem — not a rate-limit problem — and retrying won't help.

V8's old-generation heap filled up and the garbage collector couldn't free enough space, so V8 aborts the process with a fatal allocation failure. Default heap is ~4GB on 64-bit; long-lived references (caches, listeners, closures, big arrays) prevent reclamation.

Postgres rejected your connection because the server has hit `max_connections`. Each connection consumes ~10MB of RAM plus a backend process; the limit exists to protect the host from memory exhaustion.

The API key you sent in the `Authorization: Bearer ...` header doesn't match any active Stripe key — usually because of a test/live mode mismatch, a deleted or rolled key, copy-paste corruption, or a missing environment variable falling through to `undefined`.

Redis hit its `maxmemory` limit and the configured `maxmemory-policy` is `noeviction` (the default), so it refuses every write. Reads still work; commands that allocate memory return `OOM command not allowed when used memory > 'maxmemory'`.

Anthropic's infrastructure is at capacity for the model you requested. This is server-side, not a problem with your code or your account — Claude is experiencing a traffic spike or capacity event and rejecting requests until load eases.

You exceeded one of Anthropic's per-minute caps for the model and tier — RPM (requests/min), ITPM (input tokens/min), or OTPM (output tokens/min). Anthropic enforces all three independently and you can hit any one without breaching the others.

The IAM principal (user, role, or assumed role) making the request does not have an `Allow` statement for the action and resource being called, or an explicit `Deny` somewhere in the evaluation chain blocks it.

The S3 bucket name in your request does not exist in the region your client is calling. Either the name is misspelled, the bucket was deleted, or your client is pointed at the wrong region for a bucket that exists elsewhere.