Skip to content

README.md

Latest commit

 

History

History
99 lines (75 loc) · 4.46 KB

README.md

File metadata and controls

99 lines (75 loc) · 4.46 KB

Agent Runtime

Execution supervisor for multi-agent sessions — privilege rings, saga orchestration, and governance enforcement

Part of the Agent Governance Toolkit

License Python

Important

Community Preview — The agent-runtime package on PyPI is a community preview release for testing and evaluation only. It is not an official Microsoft-signed release. Official signed packages will be available in a future release.

Note: This package was previously named agent-hypervisor. The agent-hypervisor package is still available for backward compatibility but will redirect to agent-runtime in a future release.

What is Agent Runtime?

Agent Runtime provides execution-level supervision for autonomous AI agents. While Agent OS handles policy decisions and AgentMesh handles trust/identity, Agent Runtime enforces those decisions at the session level:

  • Execution Rings — 4-tier privilege model (Ring 0–3) controlling what agents can do at runtime
  • Shared Sessions — Multi-agent session management with consistency modes (strict, eventual, causal)
  • Saga Orchestration — Compensating transactions for multi-step agent workflows
  • Kill Switch — Immediate termination with audit trail and blast radius containment
  • Joint Liability — Attribution tracking across multi-agent collaborations
  • Audit Trails — Hash-chained, append-only execution logs

Quick Start

pip install agent-runtime
from hypervisor import Hypervisor, SessionConfig, ConsistencyMode

# Create the runtime supervisor
hv = Hypervisor()

# Create a governed session
session = await hv.create_session(
    config=SessionConfig(consistency_mode=ConsistencyMode.EVENTUAL)
)

# Execute with privilege enforcement
result = await session.execute(
    agent_id="researcher-1",
    action="tool_call",
    tool="web_search",
    ring=2  # restricted privilege ring
)

Architecture

┌─────────────────────────────────────────────────────────────────────┐
│                        Agent Runtime                                 │
├─────────────┬──────────────────┬──────────────────┬─────────────────┤
│  Execution  │     Session      │      Saga        │    Liability    │
│   Rings     │   Management     │  Orchestration   │    Tracking     │
│             │                  │                  │                 │
│  Ring 0:    │  Create/join     │  Multi-step      │  Attribution    │
│   System    │  Consistency     │  Compensation    │  Vouching       │
│  Ring 1:    │  Checkpoints     │  Rollback        │  Slashing       │
│   Trusted   │  Merge/fork      │  Recovery        │  Quarantine     │
│  Ring 2:    │                  │                  │                 │
│   Standard  │                  │                  │                 │
│  Ring 3:    │                  │                  │                 │
│   Sandboxed │                  │                  │                 │
└─────────────┴──────────────────┴──────────────────┴─────────────────┘

Ecosystem

Agent Runtime is one of 7 packages in the Agent Governance Toolkit:

Package Role
Agent OS Policy engine — deterministic action evaluation
AgentMesh Trust infrastructure — identity, credentials, protocol bridges
Agent Runtime Execution supervisor — rings, sessions, sagas (this package)
Agent SRE Reliability — SLOs, circuit breakers, chaos testing
Agent Compliance Regulatory compliance — GDPR, HIPAA, SOX frameworks
Agent Marketplace Plugin lifecycle — discover, install, verify, sign
Agent Lightning RL training governance — governed runners, policy rewards

License

MIT — see LICENSE.