Skip to content
Book a demo
In development · early accessAI agent governance

Control what your AI tools send to models.

An LLM gateway between your people, your agents and every model provider. Route, redact, cap and record every call, and stop prompt injection and data leaks before they reach a model.

Join early accessSee what ships today
1 endpoint
for every model your company uses
Before the model
secrets and personal data redacted on the way out
Every call
logged with who, which agent and why

Point your tools at the gateway. The policies do the rest.

  1. 01

    Point

    Swap the base URL in your SDKs, IDEs and agents. Provider keys stay in the gateway and never sit on laptops.

  2. 02

    Decide

    Which models each team may use, which tools an agent may call, how much a team may spend.

  3. 03

    Inspect

    Every prompt and response is checked for secrets, personal data and injection attempts, in both directions.

  4. 04

    Record

    An audit trail of every call: user, agent, model, tokens, cost and the verdict.

What the gateway checks before a prompt reaches a model.

Identity, data, injection, tools and budget. A prompt that passes all five goes through; the rest are masked, held or stopped.

LLM firewall

Prompt injection, jailbreaks and data exfiltration patterns are caught on the way in and on the way out.

Redaction, not refusal

Card numbers, keys and personal data are masked before the prompt leaves, so work goes on.

Tool-call policies

Agents may read the wiki and open tickets, but not push to main or run a shell, unless you say so.

Model allow-list

Approved providers and models per team, with self-hosted models for the sensitive work.

Spend you can see

Budgets and rate limits per team, agent and key, with an alert before the bill.

Evidence for AI rules

Usage, policies and incidents mapped to the NIST AI Risk Management Framework.

Built for what you run.

Routes to

  • Anthropic
  • OpenAI
  • Gemini
  • Mistral
  • Ollama
  • AWS
  • Azure
  • Any OpenAI-compatible endpoint

Part of one platform.

AI extension securityAssay already checks the skills, plugins and MCP servers your agents load.
Runtime detectionSee what an agent actually executed on the machine.
Continuous complianceGateway logs become evidence for your AI controls.

AI agent governance is in development. This page shows where it is going; details will change as we build it with early-access customers. Today's product is on the platform pages.

NextApplication security

Want it first? Build it with us.

Early-access customers shape what ships: tell us what you run and what you need it to do.

Join early accessSee the platform