Architecture

Sagewai is a control plane plus a worker fleet. The control plane plans, persists, and dispatches work; the worker fleet executes it. Every workflow run flows through this split. This page introduces the model. The four chapters that follow (Runtime topology, Security tiers, Execution modes, Sandbox backends) drill in.

Three invariants

Three statements that the entire platform is structured to keep true:

1. The worker is the only executor

The control plane (admin server, CLI, autopilot) plans and dispatches — it never runs a workflow step. Workers consume work; they do not allocate it. This means you can scale execution capacity (workers) independently from planning capacity (control plane), and a saturated fleet does not stall planning.

2. The execution mode is per-step, not per-deployment

A single workflow run can plan in Mode 0 on the worker, run an isolated tool call in Mode 1, hit a customer database in Mode 2, and dispatch a CLI agent like Claude Code in Mode 3 — each step picks its own mode. You don't pay sandbox cost for cheap planning steps and you don't lose isolation for sensitive ones.

3. The sandbox is the trust boundary (when present)

When a step runs in Mode 1+, the sandbox container is the security boundary. Customer secrets exist inside it, env-injected from the Sealed Identity backend. The worker host knows secret KEY NAMES but never holds plaintext values. The control plane sees neither.

Two backend surfaces

Sagewai has two pluggable backend types. They are independent — choose each to fit your infrastructure:

Backend typeWhat it picksExamples
Sandbox backendwhere the sandbox container, pod, or function livesDocker, Kubernetes, Lambda
Identity backendwhere Sealed Identity values come fromBuiltin (file-based), Vault, 1Password, AWS Secrets Manager, SOPS, Bitwarden

A single deployment can mix: Kubernetes sandbox + Vault identity, or Docker sandbox + Builtin identity, or anything else. See Sandbox backends for the sandbox decision guide. Identity backend selection is covered in the Sealed-ii decomposition (link forthcoming as that ships).

When you need this section

Read this section if you want to:

  • Understand how a workflow run actually executes
  • Decide which sandbox backend to deploy on
  • Pick the right execution mode for a workflow step
  • Reason about credential isolation and audit
  • Evaluate Sagewai for a security-sensitive deployment

If you just want to write your first agent, jump to the Quickstart. The architecture is consistent underneath, but you do not need to internalise it to start building.

Canonical reference

These four user-facing pages mirror the four authoritative architecture documents in the platform repo:

Those internal docs are the contract between implementation threads; this section is the user-facing render. They stay in sync via the docs(architecture) PR-title convention — when an internal doc changes, the corresponding page here is updated in the same PR.