🤖 Agent Runtime

The Agent Control & Provenance layer for autonomous AI agents: execution-provenance graphs, lethal-trifecta detection, an inline MCP proxy, delegation-chain enforcement — plus per-agent identity, session budgets, tool policies, and real-time monitoring.

Overview

The Agent Runtime is a control plane purpose-built for autonomous AI agents. Unlike basic API gateways that treat every request the same, CloudVera understands agent sessions — giving you granular control over what agents can do, how much they can spend, and what tools they can access.

Session Budgets

Set hard limits on what agents can spend within a session. When a budget is exhausted, the session is paused or terminated.

Tool Call Policies

Control which external tools and functions your agents can invoke, require approval for the dangerous ones, and keep agents on-mission.

Control & Provenance

The layer a request-scoped proxy can't reach: CloudVera sees the agent's decision loop across a whole session, not one request at a time. A typed execution-provenance graph is the substrate; detection and enforcement are built on top.

Browser Rendering

Give agents the ability to browse the web with built-in security controls.

RAG Knowledge Bases

Give each agent its own searchable knowledge base using Retrieval-Augmented Generation.

Agent SDK Quickstart

The @cloudvera/agent-sdk wraps /api/ai-gateway/agents with typed clients + an approval polling helper. Install, instantiate, register the agent, run a session — five lines per stage. Auth accepts either a dashboard JWT (interactive) or a `cv_` personal access token (issue at Settings → API Keys; scope `agents:write` to operate, `admin` to review approvals or use the kill switch). PATs identify as `apikey:<keyId>` for 4-eyes accounting on approvals.