# AI Badgr

> AI Badgr is the compute API for running AI workloads. Submit GPU jobs, serve open models, and route AI API calls, with spend caps, health checks, teardown, logs, and receipts.

AI Badgr is the compute API for running AI workloads. Developers submit GPU jobs, serve open-source models, and route AI API calls through a single API — with spend caps, startup health checks, automatic teardown, streaming logs, billing, and receipts on every run.

## What AI Badgr does
- Compute API: submit jobs (image.generate, custom.run, model.serve, train.lora, audio.transcribe) with a type, input, and max budget
- GPU job execution: runs Python scripts, containers, or eval pipelines on H100, A100, L40S, RTX 4090, and more
- Open model serving: deploys Llama, Mistral, Qwen, and other open-source models as OpenAI-compatible endpoints
- AI API gateway: routes OpenAI, Claude, Gemini, and open-model requests with retries, fallback, and receipts
- Spend caps: every job has a max_cost policy — the workload stops automatically when it hits the limit
- Automatic teardown: Badgr tears down completed or failed workloads so billing never runs away

## Core pages
- [Compute API for AI workloads — run GPU jobs, serve models, route AI APIs](https://aibadgr.com)
- [GPU compute — run GPU jobs and serve open models without managing infrastructure](https://aibadgr.com/gpu)
- [Badgr Capacity — live GPU route marketplace with max-cost protection](https://aibadgr.com/capacity)
- [OpenAI compatible API documentation](https://aibadgr.com/docs)
- [AI API gateway for production AI APIs](https://aibadgr.com/ai-api)
- [Why AI Badgr was built — about](https://aibadgr.com/about)
- [OpenAI to AI Badgr 2-line migration](https://aibadgr.com/proof)
- [OpenAI API alternative with retries and fallback](https://aibadgr.com/openai-alternative)
- [OpenRouter alternative routing and outage handling](https://aibadgr.com/openrouter-outage)
- [Llama API, Qwen API, Mistral API, and open-source LLM API](https://aibadgr.com/open-source-llm-api)
- [LLM API gateway for multi-model AI provider routing](https://aibadgr.com/llm-api-gateway)
- [Affordable production AI API with provider abstraction](https://aibadgr.com/affordable-ai-api)
- [Reliable Claude-compatible AI workflow automation](https://aibadgr.com/claude-alternative)
- [AI API request tracking and cost debugging](https://aibadgr.com/bill-calculator)
- [AI job infrastructure with execution receipts](https://aibadgr.com/receipt-rescue)
- [AI provider routing and outage fallback guidance](https://aibadgr.com/outage)
- [AI provider routing comparison guides](https://aibadgr.com/comparison)
- [Paid video, audio, PDF, and image tools](https://aibadgr.com/tools)
- [Compute APIs for AI jobs](https://aibadgr.com/apis)

## Route marketplace (dynamic pages)
Badgr Capacity is a live GPU route marketplace. Provider identity is never exposed — every route is a Badgr route with GPU class, region, price, availability, startup time, and reliability. Pages are generated from real route data, not static content:
- [Live capacity search with filters](https://aibadgr.com/capacity) — filter by GPU, region, model, workload, price, availability
- GPU landing pages: https://aibadgr.com/capacity/gpu/{gpu} — e.g. https://aibadgr.com/capacity/gpu/h100, https://aibadgr.com/capacity/gpu/a100-80gb, https://aibadgr.com/capacity/gpu/l40s
- GPU + region pages: https://aibadgr.com/capacity/gpu/{gpu}/{region} — e.g. https://aibadgr.com/capacity/gpu/h100/europe (regions: us, europe, asia)
- Model serving pages: https://aibadgr.com/serve/models/{model} — e.g. https://aibadgr.com/serve/models/qwen-7b, https://aibadgr.com/serve/models/llama-70b, https://aibadgr.com/serve/models/deepseek
- Workload pages: https://aibadgr.com/run/{workload} — e.g. https://aibadgr.com/run/vllm-endpoint, https://aibadgr.com/run/comfyui, https://aibadgr.com/run/lora-training, https://aibadgr.com/run/batch-inference
- Specific job pages: https://aibadgr.com/run/{workload}/{job} — e.g. https://aibadgr.com/run/comfyui/flux, https://aibadgr.com/run/lora-training/sdxl-lora, https://aibadgr.com/run/vllm-endpoint/qwen-7b, https://aibadgr.com/run/batch-inference/classify-csv
- Individual route listings: https://aibadgr.com/listings/{route_id} — e.g. https://aibadgr.com/listings/h100-europe-a

## npm packages & CLI tools
Free, open-source, local-first CLIs and libraries. Each installs from npm, runs without signup, and (where applicable) prints a JSON report:
- [badgr-auto — local AI proxy that cuts token and routing costs](https://aibadgr.com/docs/badgr-auto)
- [badgr-cli — run GPU jobs and serve models from one command](https://aibadgr.com/docs/badgr-cli)
- [badgr-agent — CI pipeline check runtime for build failure triage](https://aibadgr.com/docs/badgr-agent)
- [badgr-mcp — MCP server for provisioning GPU jobs from coding agents](https://aibadgr.com/docs/badgr-mcp)
- [badgr-agent-resume — resumable step checkpoints for agent runs](https://aibadgr.com/docs/badgr-agent-resume)
- [badgr-llm-guard — guarded AI API client with retries and spend caps](https://aibadgr.com/docs/badgr-llm-guard)
- [badgr-gpu-doctor — GPU VRAM and CUDA preflight checks](https://aibadgr.com/docs/badgr-gpu-doctor)
- [badgr-mcp-doctor — MCP server connection and tool diagnostics](https://aibadgr.com/docs/badgr-mcp-doctor)
- [badgr-config-doctor — env and config preflight checks](https://aibadgr.com/docs/badgr-config-doctor)
- [badgr-ci-triage — CI log failure triage and next-step suggestions](https://aibadgr.com/docs/badgr-ci-triage)
- [badgr-eval-check — repeatable assertions on LLM output](https://aibadgr.com/docs/badgr-eval-check)
- [badgr-security-scan — offline secret and prompt-injection scanner](https://aibadgr.com/docs/badgr-security-scan)
- [gpu-price-finder — compare live GPU pricing routes from the terminal](https://aibadgr.com/docs/gpu-price-finder)
- [gpu-monitor — flag idle GPUs and estimate wasted spend](https://aibadgr.com/docs/gpu-monitor)
- [ai-prod-guard — production guard for AI apps: cost limits, retries, fallbacks](https://aibadgr.com/docs/ai-prod-guard)

## Best fit queries
- Compute API for AI workloads and GPU jobs
- Run GPU jobs with spend caps and automatic teardown
- Serve open-source models as OpenAI-compatible endpoints
- AI job infrastructure with health checks, logs, and receipts
- AI API gateway with fallback routing and provider abstraction
- LoRA fine-tuning and batch inference API

## API compatibility
The Compute API uses POST /v1/jobs with a job type, input, and policy (max_cost, max_runtime_minutes). The AI API is OpenAI-compatible — swap the base URL and your existing SDK keeps working. Both APIs return structured receipts with cost, runtime, logs, and output.