Foundation
Typed identity, temporal state, and proof-oriented primitives define what counts as a valid runtime fact.
Platform
A control plane for agents that can have consequences
MAPLE separates cognition from authority. Models can reason, but only commitments that pass policy and capability evaluation can become real-world effects.
Layer Cake
Each layer exists to make production agent systems legible
Kernel
Event fabric, commitment gate, worldline ledger, memory engine, and operator bus enforce consequence semantics.
Supply chain
Packages, registry, and model operations let teams version and move cognitive services safely.
Platform services
Guard, Foundry, and Fleet add governance, eval, rollout, and cost-aware orchestration.
Service Topology
Gateway, runtime, Guard, and worldline storage are separate on purpose
User -> maple-gateway -> maple-control -> maple-registry
| |
v v
maple-runtime -> maple-model
|
v
maple-worker -> maple-guard
worldline-ledger records gateway, runtime, worker, and guard receiptsRequest Lifecycle
Consequence is a multi-stage runtime decision, not a direct tool call
Step 1
Gateway authenticates the tenant and starts a trace.
Step 2
Control resolves the right package, model policy, and environment bindings.
Step 3
Runtime forms intent and submits a typed commitment.
Step 4
Guard evaluates capability, policy, risk, and approval requirements.
Step 5
Worker executes only authorized consequences and records receipts.
Step 6
Observer and WorldLine persist the full provenance chain for replay.
Need the technical deep dive?
The docs section covers architecture overview, worldline model, commitment boundary, and the runtime invariants in detail.