AI Workforces
as Code.
Declare AI workforces. Compile infrastructure. Deploy to real systems. Operate with confidence.
CAPABILITIES
RESOURCES
REAL SYSTEMS
ONE DECLARATIVE MODEL
Define missions, agents, capabilities, resources and policy in versioned files.
Resolve references and compile declarations into a deterministic manifest.
Review the diff, then reconcile desired state against the control plane.
Execute work through governed runtimes, resources and capabilities.
Infrastructure should be declared.
Not assembled by hand.
Passerel turns a repository into an operational AI workforce. The same project that describes the workforce becomes the basis for review, deployment and repeatable operations.
kind: Mission metadata: name: customer-support spec: agents: - triage - specialist capabilities: - crm.read - email.reply resources: - customer-db - knowledge
PASSEREL target: prod Mission customer-support + create Agent triage + create Agent specialist + create Resource customer-db ~ update Capability email.reply + create ──────────────────────────────────── Plan complete. 4 to create, 1 to update, 0 to destroy.
One model.
Two planes.
Keep lifecycle management separate from runtime execution. Passerel connects declared intent to governed execution without collapsing everything into one service.
Compile · Plan · Apply
PROJECT → MANIFEST → PLAN → APPLIED STATE
Runtime · Execution
INGRESS → TASK → RUN → CAPABILITY → RESOURCE
A vocabulary for
AI workforces.
Small, typed primitives give teams a shared language for designing and operating AI systems.
Mission
The operating boundary for a workforce and its authorized work.
Agent
A declared worker with a role, harness and runtime configuration.
Capability
A governed action an agent can invoke through Passerel.
Resource
A real system or data source exposed through an adapter.
Goal
A durable objective that gives work direction and completion criteria.
Policy
Explicit constraints around access, execution and operations.