agentgate

The local model control layer for coding agents. Route, convert, fail over, and trace model requests locally.

Collateralized execution engine for AI agents: bond-and-slash accountability with Ed25519 identities, bounded exposure, and progressive trust tiers.

Smallest outsider-readable proof path through AgentGate + MCP Firewall: governed write_file, independent on-disk verification, and inspectable audit artifacts.

Adversarial test rig for AgentGate. Five attack stages from solo probes to 9-agent coordinated swarms.

A bonded file-transform agent governed by AgentGate's bond-and-slash model

Bounded human-to-agent delegation with checkpointed execution, dual-bond accountability, and a local append-only transparency log.

Evidra + AgentGateway: evidence intelligence layer & CKA/CKS certification for kagent — 75 real infrastructure scenarios, 3 models, A2A integration

Governance proxy for MCP tool calls with narrow independent verification for governed write_file and delete_file effects.

Design-stage: tests whether bond-and-slash can govern knowledge integrity when a saboteur agent poisons shared data.

Public installer for AgentGate: validation gates and token-budget guardrails for autonomous coding agents

Helm charts for deploying AgentGate AI governance services on Kubernetes.

本地优先的 AI Agent 控制平面：统一监控、审批、审计与安全 Worker 协议，支持 FastAPI、React、PostgreSQL 在 Windows 上运行。