Mohamed Abomosallam@mohamedaymnabomosallam66962

AI agents don't crash. They loop. A broken tool call doesn't throw; it retries the same prompt, with the same fat context, at full API price, at 2 a.m., on your key. Spend meters like LiteLLM or Helicone tell you what you spent, after the fact. Nobody watches agent behavior at runtime and acts. MAAT is that judgment layer: a drop-in gateway (adoption is one line, the base_url) that watches every workflow. Progress-aware loop detection kills runaway retry loops without killing healthy multi-step agents. a growing conversation is a working agent; only stagnant repetition is a loop. Per-workflow budgets enforce graduated consequences: warn, then downgrade, then kill. Enforcement lives outside the model's context, so a prompt-injected agent cannot talk its way past it. Observe mode runs the same judgments in shadow to tune thresholds on production traffic risk-free, and every kill produces a one-click incident report with the measured burn rate projected forward. Built on AMD end to end: the gateway runs live on an AMD Developer Cloud MI300X droplet; the primary tier is Fireworks AI serverless (Kimi K2.6 at real prices); and near-budget workflows degrade to Gemma 3 4B served by vLLM on the same MI300X via ROCm — AMD-hosted Gemma at $0, so agents degrade instead of dying. Verified on real traffic: a retry loop carrying ~4.9k tokens per call was killed on its third identical attempt — $0.0124 burned in 8.3 seconds, a measured pace of ~$322 over an unattended weekend. Gateway overhead is ~7 ms p50, and the full test suite (guard logic plus HTTP integration) passes. Try it live during the judging window: the dashboard at http://129.212.191.62:8080 is public, and the README includes a deliberately public, budget-capped demo key. a leaked key bounded by MAAT costs at most its budget. That is the product.

To participate in the hackathon, simply click Sign up with AMD. If you are not already a member of the AMD AI Developer Program (ADP), you will need to create an account. ADP members can access AI tutorials, experts, and community support. In addition, new members can claim a free month of DeepLearning.ai pro, $100 in AMD GPU cloud credits, and $50 in Fireworks AI API credits. All hackathon participants will receive on hackathon launch additional compute and API credits for their projects. Ready to build what's next on AMD? Sign up with AMD below to secure your spot.