Reduce AI spend by 30% in less than 30 days
The self-hosted AI gateway from the creators of Kubernetes
What can you do with Stacklok’s AI gateway?
We’re so glad you asked! You can:
- Set and enforce AI cost controls by individual, agent, project or team, so that you have a forecastable line item
- Tie every agent request to a real person or named agent through your identity provider
- Establish one location to rotate keys; one place to audit calls; one surface to secure
- Unify both AI and MCP gateways with a singular control point
- Select a model for every task based on your prioritizations for security, cost and performance
“Stacklok built our AI gateway with us; they were clearly invested in successful outcomes.”
Let’s get a little more technical. Here’s what separates Stacklok’s AI gateway from alternatives.
Self-hosted
Alternatives: SaaS AI gateways require you to route data through a third-party cloud, which is a no-go for enterprises with strict data residency requirements.
Stacklok: Our solution runs self-hosted in your environment, governed by your policies, with zero risk of vendor lock-in. Your data and telemetry always stays behind your firewall.
Policy layers
Alternatives: Many AI gateways focus almost entirely on the LLM-call layer, routing requests across model providers and tracking costs.
Stacklok: We enforce policy at three separate checkpoints: (i) registry (who can discover and install servers), (ii) MCP gateway (which tool calls are permitted), (iii) AI gateway (which models should answer a prompt).
Accountability
Alternatives: A lot of AI gateways stop at API key-level authentication and therefore static shared credentials.
Stacklok: Our AI gateway integrates with your IdP so it ties every agent action to a verified user identity (or named agent), so ‘Agent A’ only sees what ‘User A’ is allowed to see via a real OAuth token exchange.
Judgment
Alternatives: Most AI gateways are a black box to which you delegate model routing decisions (potentially introducing security risks).
Stacklok: We allow you to use your judgment to define routing based on cost v. performance v. security. Your team accumulates experience and learning that extends to future models.
Fun fact:
Stacklok is a charter member of the Agentic AI Foundation and a member of the Security Committee.
Want to talk through your options?