Skip to content

APS Execution Environment

APS defines how agents execute, independent of runtime, OS, or host platform.

Goal: Run any APS agent anywhere.

Execution Modes

Mode Description Use Case
Local Execute on host Dev, prototyping
Sandbox Execute inside isolated container Security, enterprise
Remote Execute via APS executor service Cloud/hybrid Orchestration

APS does not mandate sandboxing — but provides hooks for it.

aps run # local aps exec docker # future aps exec remote # future

Core Contract

Inputs

  • stdin: JSON request
  • env: runtime variables
  • cwd: agent root
  • $PYTHONPATH or language equivalent auto-set

Outputs

  • stdout: streaming or final JSON
  • stderr: logs only (never protocol)

Environment Variables

Variable Purpose
APS_AGENT_ROOT Path to unpacked agent
APS_AGENT_ID id from manifest
APS_AGENT_VERSION version
APS_RUNTIME_KIND python, node, etc
APS_STREAM_MODE true/false
APS_TEMP_DIR Ephemeral workspace

Runtime Rules

Rule Required
No network unless policy allows ✅ future policy file
Agent cannot modify its package ✅ required
Working dir = agent root ✅ enforced
All state in temp dir ✅ future
Final stdout must contain JSON frame ✅

Streaming Protocol (preview)

Lines until final JSON response:

{ "status": "...", "outputs": {...} } ```

Future extension: APS-FRAME: JSON\n