Start Here

amberflo is the system of record for ai usage and ai spend it sits behind your ai gateway, meters every request in real time, turns usage into cost, and attributes that cost to the teams, applications, or business units that generated it if you are running ai at scale and do not have accurate, attributable, near real time cost visibility, you are already flying blind this guide introduces the core parts of the amberflo ai platform and how they fit together quickstart connect an ai gateway to amberflo today, amberflo provides first party support for litellm additional gateways are coming soon, including kong, apigee, and cloudflare how it works you deploy or configure an ai gateway in front of your ai providers and models the gateway forwards structured usage events to amberflo amberflo meters those events, enriches them with metadata, and converts usage into cost quickstart docid 9yiergqwbde8hklezyioe quickstart docid\ xpnvzezmm cemyw7euslp once connected, you immediately start seeing real usage and spend data in the product there is no batch processing and no waiting for cloud billing exports ai spend ai spend gives you a clear, real time view of how much your ai workloads actually cost what you can see month to date, quarter to date, and year to date spend top spend by business unit, application, or team top spend by model and provider spend trends over time provider level and model level breakdowns how you explore you can filter and slice the data by dimensions such as business unit or team provider or model environment (prod, staging, dev) time range modality or request type this is not estimated billing data it is derived directly from real ai usage flowing through your gateway ai observability amberflo provides full observability into every ai request this is not just cost reporting it is operational visibility into how ai is actually being used what you can inspect individual ai requests and responses token usage per call who made the request and when request latency and execution time failed calls and error types usage volume and trends over time why this matters ai failures, retries, and latency directly impact both cost and user experience observability lets you identify runaway usage debug failures and misconfigurations understand performance bottlenecks correlate cost spikes with real behavior this data is captured at request time, not reconstructed later from billing files alerting and automation amberflo lets you create alerts based on ai usage and spend in near real time alert types absolute usage or cost thresholds percentage increases or drops sudden spikes in usage notification and action alerts can trigger email notifications for awareness webhooks for automation webhooks enable advanced workflows such as throttling or blocking traffic triggering downstream processes integrating with internal tooling or incident systems this is how teams move from passive reporting to active control department chargebacks (business units) all ai usage and cost in amberflo can be attributed to business units a business unit can represent a department an application a team a cost center any internal ownership boundary you care about what you get for each business unit, amberflo can generate a complete monthly breakdown that includes usage by model cost by provider total spend detailed line item usage these reports can be used for internal chargebacks or showbacks, creating real accountability for ai spend across the organization ai catalog and custom rates the ai catalog is a centralized inventory of ai models and pricing what it includes all available models public list prices from ai providers support for multiple providers and regions custom negotiated rates if you have negotiated custom pricing with a provider such as openai, azure, or others, you can enter your actual contracted rates amberflo will then calculate spend using your real pricing, not public list prices this ensures accurate internal reporting correct chargebacks no surprises when invoices arrive this capability is critical for enterprises and is rarely supported elsewhere