Run agents securely
Run an approved agent in an isolated microVM with governed authority.
Explore this feature →AIrspace / Model, tool & peer gateways
A2A: Under Development
Control model calls, tools and peer communication
LLM, MCP and A2A address different actions while sharing the need for attributable authority.
What this means for your team
- Large language model (LLM) Gateway: govern supported model access, provider credential custody and configured consumption limits.
- Model Context Protocol (MCP) Gateway: constrain servers and tools and bind account selection to the actual launching user.
- Agent-to-Agent (A2A) Gateway: Under Development for internal and external peer discovery, delegation and delivery.
- Use configured network controls with the gateways to govern where an agent can connect.
A practical example
The same Agent launched by a different person must not borrow the first person’s connected SaaS account.
How it works
Select a step to read its explanation
- The gateway identifies both the specific running agent and the person who launched it. That context makes it possible to evaluate the call against the right user and agent, rather than treating all runs as the same caller.
- The gateway checks whether this user and agent are currently allowed to access the requested model or tool, and applies the relevant scope and policy before the request proceeds.
- For a model request, the LLM Gateway connects to the configured provider. For a tool request, the MCP Gateway connects to the approved server and selects the appropriate account. Agent-to-agent communication through the A2A Gateway is under development.
- Apply the configured restrictions, such as which model or tool may be used, which connected account is appropriate, and any applicable usage limits. The precise controls depend on the approved configuration.
- Associate the decision and supported usage records with the run. Authorized reviewers can use this evidence to investigate what was allowed or refused and understand consumption.
1 / 5
This example illustrates the journey. Select any step to explore it at your own pace.
Explore more AIrspace features
Images & workers
Prepare approved agent images and connect the hosts that run them.
Explore this feature →Live sessions & recordings
Follow a live run and review its recording with the right access.
Explore this feature →AI Agent Session Recording Analyzer
Under Development
Investigate recorded agent sessions through evidence-linked findings, timelines and clips.
Explore this feature →Agent threat intelligence
Under Development
Correlate evidence from independent observers to investigate agent behavior.
Explore this feature →