If you're running more than a handful of AI agents in production, you already know the dirty secret: most of them are hitting OpenAI or Anthropic directly, with no registered identity, no centralized policy, and no way to shut one down if it starts misbehaving.
That's fine when you have three agents. It's a nightmare when you have three hundred.
Which agent made that API call?
Which agent is allowed to use web search or file access?
What happens when one starts leaking PII into prompts, or gets hit with a prompt injection attack?
If something goes wrong at 2am, how do you turn it off right now — without redeploying?
Most teams don't have good answers to these questions. Kimss is a free gateway built to fix that, without asking you to replatform anything.
What it actually does
Kimss sits between your agents and the model provider. You don't change your architecture — you change one line of config.
python
from openai import OpenAI
client = OpenAI(
base_url="https://gateway.kimss.dev/v1", # <- the one line
api_key="your-existing-key",
)
Same for Anthropic's SDK — point base_url at the gateway and everything else in your code stays the same.
From that point on, every request flowing through the gateway gets:
- A registered identity
Every agent gets a discoverable identity mapped back to your company login. No more mystery traffic — you can see exactly which agent, which team, and which human is behind every call.
- Content safety at the edge
Requests and responses are screened for hate speech, violence, and prompt injection attempts before they hit your application logic.
- PII and secret scrubbing
The gateway strips sensitive data — PII, API keys, tokens accidentally pasted into a prompt — before it ever reaches the model provider.
- Workspace-level tool opt-ins
Want only certain agents to have web search, file access, or code execution? That's a policy on the gateway, not something buried in each agent's codebase.
- A real kill switch
If an agent goes rogue — infinite loop, data exfiltration attempt, whatever — you flip it off instantly from the control plane. No redeploy, no waiting on CI/CD.
Why "no replatforming" actually matters
A lot of "AI governance" tooling asks you to adopt a new framework, migrate your agent logic, or wrap everything in a proprietary SDK. Kimss doesn't. It's a drop-in proxy for the OpenAI and Anthropic client libraries you're already using. If your code can set base_url, it can run through Kimss today.
Who this is for
If you're an individual developer with two agents and a personal project, you probably don't need this yet.
If you're an engineering org where agents are multiplying faster than anyone can track them — across teams, across tools, across environments — this is the boring infrastructure piece you didn't know you were missing.
Try it
Kimss is free. Point your existing client at the gateway and you're done — no migration, no new SDK to learn.
Have you hit this problem in your own org — agent sprawl with no central control? Curious how other teams are handling agent identity and governance. Drop a comment below.
