SwarmForge

01AI Runtime Governance

Control what AI can do. Protect what AI can see.

SwarmForge adds runtime governance and data protection to the AI systems and agents you already use. See what they do, control what they can do, protect the data they access, and keep evidence of what happened.

  • Keep your models. Keep your agents. Keep your workflows.
  • SwarmShield protects what AI can see.
  • The Control Plane governs what AI can do.
Explore the platform
SwarmForge architecture showing SwarmShield protecting enterprise data, SwarmForge Runner executing an AI agent, the Control Plane governing policies and actions, and the CLI integrating existing systems.Open the full-size architecture

02The problem

AI creates two new control boundaries.

AI becomes operational when it can see enterprise context and take actions in real systems. Both boundaries need explicit control.

  • Can it see something it shouldn't? Protect the data boundary with SwarmShield.
  • Can it do something it shouldn't? Govern the action boundary with the Control Plane.
  • Protect data. Control actions.
Editorial metaphor: a silver execution path from a development structure stops at a brass authority plane before an intact production-data lattice. It is not product proof.

03The platform

One governed AI stack. Four focused components.

Adopt the parts you need. Integrate with what you already have.

  • SwarmShield — data protection boundary. What can AI see?
  • Control Plane — governance and oversight. What can AI do?
  • CLI — developer integration. Runner — native reference runtime.
Explore SwarmForge
Generated editorial illustration for “See the governed system in motion.”: SwarmForge Platform is the referent implementation for this architecture. It makes the control relationships inspectable without requiring your workflow to move into Platform. It is not product proof.

04SwarmShield · Data Protection Boundary

Use AI without giving it everything.

SwarmShield helps reduce unnecessary exposure by protecting sensitive information before it reaches models, agents or external services.

  • Minimise the context exposed for each task.
  • Mask personal data and protect credentials at the AI boundary.
  • SwarmShield provides technical controls that can support GDPR and privacy-by-design programmes. Compliance remains an organisational and legal responsibility.
Generated editorial illustration for “Context enters through a visible boundary.”: The Platform reference surface shows model-context tools and server configuration as part of the governed lifecycle, alongside the data and model choices your team keeps. It is not product proof.

05Control Plane · Governance & Oversight

Governance happens at runtime.

Policies matter most when they can affect execution. The Control Plane gives operators visibility and control over agent behaviour as it happens.

  • Allow or deny actions against explicit policy.
  • Require human approval before material effects.
  • Trace, audit and observe what happened.
Explore the Control Plane
Generated editorial illustration for “Not on your own is a real decision.”: Approval and halt are explicit state machines, not a generic yes-or-no button. It is not product proof.

06CLI · Developer Integration

Start with what already exists.

The SwarmForge CLI provides the developer path for integrating governance into existing agents, workflows and tools.

  • No forced rewrite or platform migration.
  • Instrument and connect incrementally.
  • Designed to integrate with existing agent and workflow architectures.
See the integration path
Generated editorial illustration for “Keep your stack where it runs.”: A paid design-partner integration will keep Python and LangGraph in the customer application while the framework-neutral Control Plane remains outside its framework layer. It is not product proof.

07Runner · Reference Agent Runtime

A reference runtime for governed agents.

SwarmForge Runner demonstrates how agents can execute natively inside a governed architecture, with policy, traceability and control designed into the runtime.

  • Execute and reason inside the native governed stack.
  • Use the Runner when you want the reference runtime.
  • Use the CLI and Control Plane when you want to govern systems you already have.
Generated editorial illustration for “A workflow is more than a model call.”: The reference Platform surface connects agent nodes, tools and execution controls so a governed lifecycle can be inspected as a system. It is not product proof.

08From decision to action

A model can make a mistake. It doesn't have to become an action.

SwarmForge places enforceable controls between AI decisions and their effects.

  • Data → SwarmShield → AI or agent.
  • Policy decision → Allow · Deny · Approve.
  • Action → traceability and evidence.
Explore the architecture
Generated editorial illustration for “A decision is not an observed effect.”: The contract treats permission, external execution and observation as separate facts. It is not product proof.