Submit your issue
Diagnosis is free and secrets are redacted before AI processing. A GPU runs only after you approve the displayed hard cost cap.
Badgr Run Issue
Paste a Discord thread, Slack message, GitHub issue, Docker image, or error log. Badgr detects the workload, redacts secrets, and finds the cheapest safe GPU to verify it.
Five steps from messy conversation to a verified GPU command - no manual debugging.
Paste a Discord thread, Slack message, support ticket, error log, GitHub issue URL, Docker image, or repository URL.
AI extracts the workload, image, model, launch command, errors, and any details still needed.
Known configurations, multi-GPU flags, and large-model download requirements are checked before a test is proposed.
Review the proposed test, copy the exact command to run yourself, or continue to verification.
A private result page shows every execution stage, the verified exact command, actual cost, and teardown confirmation.
Any format your team uses to share GPU issues - Badgr handles the extraction.
Discord, Slack, email, or support tickets - paste the whole thread, Badgr extracts the signal.
Badgr safely fetches the issue title, body, and comments via the GitHub API.
Submit an image name. Badgr uses known image patterns and asks you to confirm the detected workload details.
Badgr reads Dockerfiles, launch scripts, and configs via the GitHub API file tree.
Paste a workflow JSON and Badgr extracts nodes, models, and missing custom nodes.
CUDA OOM messages, stack traces, startup failures - any error log becomes a structured reproduction plan.
Diagnosis is free and secrets are redacted before AI processing. A GPU runs only after you approve the displayed hard cost cap.
Yes - detection, secret redaction, AI extraction, static inspection, and template matching are all free. You only pay if you choose to run a GPU test, and the maximum charge is shown before you approve.
Badgr runs a pattern-matching redactor over your pasted text before sending anything to an external AI model. API keys, tokens, and passwords are replaced with [REDACTED]. Only the variable names are stored - never the values.
The price depends on the workload and GPU selected. Badgr shows an estimated range and a hard maximum (e.g. AUD $0.80–$2.40) before you approve. You’re only charged for actual runtime - unused reservation is released immediately.
For simple workloads costing ≤ $2 USD with no multi-GPU requirement and no large model download, you can request Badgr to sponsor the GPU test. A human reviews and reaches out - no GPU runs automatically.
Yes. On the approve screen, click “Run it yourself” to reveal the exact Badgr CLI command, recommended GPU, VRAM, and runtime cap. No payment or account needed for that.
vLLM serving, ComfyUI image generation, PyTorch training (torchrun / Deepspeed / Axolotl / TRL), lm-evaluation-harness, and generic Docker containers. More are added as verified templates accumulate.