Opencontroller by Lyzr
Opencontroller blog
Answers to the questions enterprises ask about governing AI agents
Agent governance basics
All in this topicHow do you govern open-weight models like DeepSeek or Qwen running on-premises?
Check each model's licence, load only safe weight formats, size GPUs honestly, evaluate before approval and route agents via a gateway. A 10-point checklist.
How do you govern AI agents built by vendors and system integrators?
Give vendor-built agents an isolated workspace, a named internal owner and contract rights to logs, model disclosure and shutdown. A checklist for each stage.
Discovery and shadow agents
All in this topicHow do you register an AI agent your team doesn't host?
Register agents you do not host through a manual record, platform connector, endpoint or gateway path. The minimum record, and what you can observe at each depth.
How do you govern coding agents on developer laptops with MDM?
Use MDM to push managed settings to each coding agent and deploy a device agent that inventories tools, MCP servers and skills. File paths and privacy rules.
Runtime control
All in this topicWhat should a production monitoring view for AI agents show?
A panel-by-panel spec for monitoring AI agents in production: status, severity-ranked issues, deviation, performance, quality, cost per agent and owner, posture.
Should AI agents do financial calculations themselves?
No. Let the agent plan and explain, and let deterministic code do the arithmetic. How to design calculation tools, thresholds and evidence for finance agents.
Tools and MCP
All in this topicWhat requirements should an enterprise set for managing AI agent skills?
A requirements list for AI agent skills management in six lifecycle areas, from inventory to retirement, with an acceptance test for each requirement.
How do you govern an MCP server that turns natural language into SQL?
Give text-to-SQL MCP servers a read-only database role, timeouts, row-level security and full query logging. Treat tool annotations as hints, not controls.
Gateways and cost
All in this topicEvidence and frameworks
All in this topicHow do you update AI agent policies when regulations change?
A five-step process to update AI agent policies when rules change: log, diff the mapping, update policy as code, re-test and file evidence. EU AI Act Omnibus example.
What evidence should you keep when a human approves an AI agent's output?
A human review record should capture what the agent proposed, its confidence and sources, who reviewed it, what they decided and when. Field list inside.
What SOC 2 evidence do auditors want for AI agents?
SOC 2 Trust Services Criteria for AI agents: what auditors test under CC6 access, CC7 operations, CC8 change and CC9 vendors, with sample evidence for each.
How does DORA apply to AI agents in financial services?
How DORA's ICT risk, incident reporting, testing and third-party articles apply to AI agents at EU banks, insurers and investment firms, with evidence for each.
What does the EU AI Act require for AI agents, and when?
EU AI Act duties for AI agents by role and risk class, with the dates now in force after the Digital Omnibus (Regulation (EU) 2026/1744).
What evidence does ISO/IEC 42001 expect for AI agents?
ISO/IEC 42001 Annex A control areas A.2 to A.10 with agent-specific evidence: inventories, impact assessments, event logs, monitoring and supplier records.
How does the NIST AI RMF apply to AI agents?
How the NIST AI RMF Govern, Map, Measure and Manage functions translate into AI agent controls, with subcategories such as GOVERN 1.6 and MANAGE 2.4.
How do you use MITRE ATLAS to threat-model AI agents?
A worked MITRE ATLAS threat model for a tool-using AI agent, with real agent technique IDs such as AML.T0080 and AML.T0086 and a detection idea for each.
What evidence maps to the OWASP Top 10 for Agentic Applications?
One control and one evidence artefact for each OWASP agentic risk, ASI01 Agent Goal Hijack to ASI10 Rogue Agents, ready for an audit or security review.
Buying and comparison
All in this topicPlatform Guides
All in this topicHow do you govern UiPath agents?
Govern UiPath agents with Automation Ops AI Trust Layer and Studio Web policies: guardrails, model locks, regions and design limits. Checklist.
How do you govern agents in IBM watsonx Orchestrate?
Govern watsonx Orchestrate agents with catalog certification, watsonx.governance-powered monitoring and central credentials. Checklist and limits.
How do you govern AI agents in Oracle Fusion Applications?
Oracle Fusion agents inherit role-based security and approvals. What to add: builder limits, template reviews, Credential Store rules and evidence export.
How do you govern AI agents in Workday?
How the Workday Agent System of Record governs Workday, customer and partner agents, and the extra controls HR and finance data need.
How do you govern Atlassian Rovo agents?
Rovo agent creation is open to everyone by default. How to restrict builders, limit agent access and fix Jira and Confluence permissions first.
How do you govern AI agents built in n8n?
Govern n8n AI agents: find instances, run the n8n security audit, restrict risky nodes, route models through a gateway and stream AI node logs.
How do you govern LlamaIndex agents in production?
Govern LlamaIndex agents through tools, model endpoints, index sources and the instrumentation module with OpenTelemetry export. Production checklist.
How do you govern agents built with Google's Agent Development Kit?
Govern Google ADK agents with plugins, before_tool_callback checks, a clear agent-vs-user identity choice and exported traces. Checklist included.
How do you govern agents built with Microsoft Agent Framework?
Govern Microsoft Agent Framework agents with agent, function and chat middleware, OpenTelemetry traces and an estate-wide registry. A practical checklist.
Do Azure AI Content Safety Prompt Shields govern AI agents?
Azure Prompt Shields detect user prompt and document attacks. What they cover for AI agents, where they sit, their limits, and what agent governance adds.
What does Databricks AI Gateway govern for agents?
Databricks AI Gateway and Unity AI Gateway for agents: usage tracking, inference tables, rate limits, guardrails, fallbacks, and the limits that matter for agents.
How do you govern agents built with AWS Strands Agents?
Control points for AWS Strands Agents: model providers, tools and MCP, hooks, multi-agent patterns, AgentCore Runtime and built-in OpenTelemetry traces.
What is Microsoft Entra Agent ID and how does it help govern agents?
Microsoft Entra Agent ID gives AI agents their own identities, blueprints, sponsors, access packages and Conditional Access. What it governs and where it stops.
How do you govern agents in Amazon Quick (formerly Quick Suite)?
Amazon Quick agent governance: custom permissions, Deny by Default, connector and MCP controls, approval gates and logging, plus the gaps you must close yourself.
What does the Salesforce Einstein Trust Layer cover for agents?
The Einstein Trust Layer protects LLM calls with grounding, masking, toxicity checks, audit and zero retention. What it covers for Agentforce agents, and what it does not.
What does Google Cloud Model Armor do for AI agents?
Google Cloud Model Armor filters, enforcement modes and integrations for agents and MCP servers, and the agent risks it does not cover on its own.
How do you audit Claude Enterprise usage?
What Claude Enterprise audit logs, the Compliance API, analytics dashboards and APIs, SCIM and retention settings each give you, and how to combine them.
What do Amazon Bedrock Guardrails cover, and what do they not?
Amazon Bedrock Guardrails policy types, tiers, enforcement and pricing, compared with the inventory, identity and cost controls agent governance also needs.
How do you govern agents built with the OpenAI Agents SDK?
Where governance controls go in OpenAI Agents SDK apps: guardrails, tool approvals, MCP tool filters, trace processors, and gateway and identity outside it.
How do you govern agents in Gemini Enterprise?
Gemini Enterprise admin toggles for agent creation and sharing, approval for shared agents, custom agent registration, audit logs and Model Armor coverage gaps.
How do you control which MCP servers GitHub Copilot can use?
The GitHub Copilot MCP policy, registry-only mode, enterprise MCP allowlists, cloud agent MCP config and firewall, and the gaps each one leaves open.
How do you govern Snowflake Cortex Agents?
How Snowflake Cortex Agent privileges, caller roles, tool access, MCP connectors, guardrails, traces and cross-region settings work, with a hardening checklist.
How do you govern ServiceNow AI agents?
ServiceNow AI agent roles, ACLs, run-as identity, supervised tool execution and execution logs, and how to bring these agents into a cross-platform inventory.
How do you govern AI agents in a Microsoft-first enterprise?
A coverage map of Entra Agent ID, Purview, Defender, Copilot Studio, Foundry and Agent 365 for AI agents, and the gaps a Microsoft-first company still has.
How do you govern agents in Microsoft 365 Copilot?
Where Microsoft 365 admins see every Copilot agent, approve org-published agents, control sharing, block or delete agents, and what sits outside that view.
How do you send AI agent and LLM telemetry to Datadog?
Three OTLP paths into Datadog, the env vars for direct export, which gen_ai attributes to keep, and how to handle prompt capture for AI agents.
What Snowflake privileges does an AI governance tool need to monitor Cortex usage?
The Snowflake database roles, Cortex views, agent privileges and a least-privilege service-user script for monitoring Cortex AI usage and cost.
How do you govern HubSpot Breeze agents?
HubSpot AI data settings, Agent builder permissions, credit and run limits, Agent Inbox, audit log and the remote MCP server for governing Breeze agents.
How do you govern AI agents running on Kubernetes?
Per-agent ServiceAccounts and workload identity, default-deny egress, Restricted pod security, quotas, admission policy and new agent projects on Kubernetes.
How do you govern CrewAI agents in production?
CrewAI agent and task defaults, MCP tool filters, guardrails, per-crew identity, gateway routing and telemetry settings for production crews.
How do you govern LangGraph agents in production?
Custom auth handlers, per-user thread scoping, interrupt() approvals, gateway routing with base_url, OpenTelemetry traces and retention limits for LangGraph.
Put the framework into practice