One control layer, not another point tool
You don't run Aegis instead of your favorite tools - you run it underneath them. Aegis manages the whole bill - cutting cost, protecting quality, and showing you the savings.
Only Aegis both actively cuts spend and provides verified savings audit receipts.
Frequently asked questions
How intelligence, caching, and savings share billing operate.
How is Aegis different from gateways, routers, and observability tools?+
Point tools give you one slice - a way to call many models, or a passive spend report. Aegis sits under your entire AI stack: it actively lowers the bill through microsecond deterministic downscaling, exact-match caching, and lossless context compression, enforces hard budgets, protects quality, and provides an audited breakdown of savings for finance.
I already use Cursor / Claude Code / OpenAI SDK - where does Aegis fit?+
You don't change your code or workflow. You simply set your tool's API base URL to Aegis (http://localhost:8080/v1 or https://api.aegis.dev/v1). Aegis handles prompt classification, cache lookups, and failovers transparently behind the scenes.
Can Aegis automatically use a more cost-effective model when a task is simple?+
Yes. Our sub-millisecond classifier evaluates prompt structure, ambiguity, and reasoning depth. Simple queries, categorizations, and extractions route to ultra-fast models (like GPT-4o mini or Haiku), saving up to 94% on those calls without sacrificing user experience.
Will my conversation stay consistent - or could the model change mid-thread?+
You retain total control. You can lock specific models permanently, use routing hints, or let Aegis optimize only when confidence is 100% assured. High-reasoning and coding tasks are never downgraded.
Do I pay twice for the same question?+
Not for an exact repeat - a byte-identical prompt returns instantly from cache at $0.00 cost. Fuzzy matching on differently-worded but semantically identical questions is built and tested, and is rolling out as we finish provisioning the on-box model it depends on.
Are my prompts and data private?+
Yes. Aegis supports per-tenant zero-retention: prompt and response payloads are evaluated in volatile memory and never stored to disk. We also provide VPC and on-premises deployment options.
How does the savings share pricing work?+
You pay a transparent base plan fee plus a verified percentage of savings generated. For every request, we record what it would have cost on the requested model vs what it actually cost on the routed model. If we generate zero savings in a month, you pay $0 in share fees.
Can I bring my own API keys (BYOK)?+
Yes. On Pro, Team, and Enterprise plans, you can bring your own Anthropic, OpenAI, or Google keys. You pay your providers directly with zero token markup.
Ready to cut your inference bill?
Get started in under 5 minutes with zero code changes.