The platform
Authority stays in the path.
Three named systems keep identity, policy, and evidence in the path of agent action. All of it is under active construction, and we say what is verified and what is not.
- AgentForge
- The strategic platform for governed agent identity, execution, orchestration, and evidence.
- AgentLabs Auth
- Multitenant identity, roles, policies, and audit foundations for people, services, and agents.
- AgentLabs Notify
- A notification foundation for timely, attributable communication across governed systems.
Why the foundation matters
Capability becomes useful when accountability survives execution.
Models can reason and tools can act. Enterprise systems still need to know who initiated the work, what authority applied, which boundaries held, and what happened next.
- Known actor
Identity
Every person, agent, and system has a known operating identity.
- Explicit scope
Authority
Permissions and policy establish explicit, contextual boundaries.
- Bounded work
Action
Agents execute through governed tools and enterprise systems.
- Reviewable history
Evidence
Each decision and action leaves a durable, reviewable record.
Reasoning may be probabilistic. Authority is not.
The platform
Three systems, one governed foundation.
AgentForge runs agents. AgentLabs Auth decides who every principal is and what it may do. AgentLabs Notify carries attributable communication out.
- AgentForgeThe strategic platform for governed agent identity, execution, orchestration, and evidence.Owns: Agent identity, runtime isolation, capabilities, supervision, and cleanup.Read more
- AgentLabs AuthMultitenant identity, roles, policies, and audit foundations for people, services, and agents.Owns: Tenant identity, roles, policies, and access audit.Read more
- AgentLabs NotifyA notification foundation for timely, attributable communication across governed systems.Owns: Attributable communication and system events.Read more
How AgentForge works
Agents run as untrusted workloads. Capabilities are granted by intersection. Missing grants fail closed.
The runtime treats agent code, model output, tools, and external inputs as untrusted by default. Credentials and platform authority never sit inside the agent's reach.
Four layers of an agent
- 01
Agent
A persistent identity with a role, permissions, and scope.
- 02
Agent image
An immutable, content-addressed package of what the agent runs.
- 03
Agent instance
A deployed configuration that persists across restarts.
- 04
Agent process
An ephemeral execution inside an isolated environment, destroyed when the work ends.
Execution state is always visible
Product interfaces expose what an agent is doing rather than hiding it behind a chat box. Identity, scope, tool use, policy boundaries, and the audit trail are visible objects.
What is built so far
The runtime and isolation boundary come first.
The runtime and isolation boundary are being built first, with each slice gated by demonstrated enforcement. Gateway, control plane, scheduling, and the public API and CLI follow on that foundation.
AgentForge is under active construction. No part of it is available for production use, and no isolation or security guarantee is claimed ahead of its own verification gates. One rule gates every slice: no demo claim may exceed the enforcement actually implemented.
Read about AgentForgeThe next step
Bring your enterprise constraints into the platform conversation.
Discuss identity, authority, workflow, integration, and evidence requirements with AgentLabs.