Frequently asked questions
How does Denpex get access to my training logs?
+
You have two options. (1) Paste logs into the web console — no install. (2) Install the single-file Python agent, which wraps your training command and ships the last 500 log lines on a non-zero exit. We never install a daemon or modify the training process.
Will adding Denpex slow down my training?
+
No. The agent is completely passive during training. It only activates on job failure. Zero overhead on forward / backward passes, zero overhead during normal operation.
Can it diagnose failures that already happened?
+
Yes. Paste logs from any past failure into the console. A free account gets 3 diagnoses per day with no credit card, instant result.
Do you store our training logs?
+
No. The agent masks PII/PHI client-side before egress by default; in strict mode it ships only anonymized failure signatures, so raw logs never leave. Raw logs are processed in memory and never written to durable storage. For zero egress, DENPEX_LOCAL=1 runs the deterministic engine entirely on-host — nothing leaves the cluster at all.
What if Denpex can't diagnose it?
+
If no pattern match exists with ≥60% confidence, Denpex falls back to an LLM-inferred diagnosis, clearly marked. You always know the confidence level of your diagnosis.
Does it work with spot instances and preemptible GPUs?
+
Yes. Spot preemption is classified separately from software failures. Denpex distinguishes between PREEMPTION events and actual training failures, so your metrics aren't polluted by cloud provider interruptions.
How does the team knowledge base work?
+
Every time an engineer confirms a fix (thumbs up in the dashboard or via the CLI), the failure signature and resolution are added to your organization's private knowledge base. Future identical failures match against confirmed fixes first.
Can Denpex compare my current failed run against a previous successful one?
+
Yes. Cross-run comparison is available on Team and Scale. When a run fails, Denpex automatically pulls the config, framework versions, and gradient norm trajectory from your last successful run of the same job and diffs them.
Can I self-host / run air-gapped?
+
Yes. DENPEX_LOCAL=1 runs the full deterministic engine — regex patterns, Xid→node-action router, cross-rank cascade, and the 3500+-class encyclopedia — entirely on-host, with zero egress (no logs, signatures, or heartbeats leave). It's a single Python file plus a bundled pattern DB; no server, no internet, no API key. The LLM fallback for novel failures is cloud-only. White-label / OEM is available on Data Center.
Is there a free trial?
+
Yes. Every new account gets 3 free diagnoses with no credit card. We also publish trial codes via partners and at conferences — see /redeem.