Opencontroller by Lyzr
Opencontroller blog
Answers to the questions enterprises ask about governing AI agents
Agent governance basics
All in this topicShould AI agent governance be centralised or federated?
Centralised, federated or hub-and-spoke AI agent governance: a comparison table, which decisions to centralise, and signals for choosing.
How do you explain AI agent governance rules to employees?
A comms plan for AI agent rules: messages by audience, a one-page rules template, rollout steps, FAQ and metrics that show adoption.
What is the difference between an AI agent registry and an agent marketplace?
A registry is the system of record for control; a marketplace is a catalogue for adoption. Comparison table and how to connect them.
What should third-party risk teams ask SaaS vendors about their AI agents?
A 19-question addendum for vendor AI agents: models and sub-processors, data use, permissions, shutdown, logs and testing, with scoring.
How should governance differ for customer-facing and internal AI agents?
Customer-facing agents need stricter content, disclosure and handoff controls; internal agents need tighter access. A side-by-side baseline.
Should system prompts be version controlled and reviewed like code?
Yes. A prompt-as-config workflow: store prompts in a repo, review changes, gate on evals, roll out gradually, and know which edits need re-approval.
What is the difference between AgentOps, MLOps and LLMOps?
MLOps manages trained models, LLMOps manages prompts and LLM apps, AgentOps manages agents that act. A table of artefacts, controls and owners for each.
How do you govern AI agents built by citizen developers?
A checklist for governing low-code AI agents built outside IT: environments, data policies, sharing limits, ownership and a promotion path to production.
Agent identity and access
All in this topicHow can AI agents call Amazon Bedrock from outside AWS without long-term keys?
Credential options for agents and gateways calling Amazon Bedrock from outside AWS: IAM Roles Anywhere, web identity federation and short-term Bedrock API keys.
What roles should an AI agent governance console have?
A role matrix for AI agent governance consoles: platform admin, security reviewer, agent owner, finance viewer and auditor, with separation-of-duties rules.
Runtime control
All in this topicShould AI agents be restricted to business hours or change windows?
Some agent actions should only run when people can respond. A decision table by action type, timezone pitfalls, an on-call rule and exception handling.
When should a customer service AI agent hand off to a human?
A numbered checklist of handoff triggers for customer service AI agents, the context to pass to the human, and how to measure handoff quality.
How do you detect an AI agent acting as a user without saying so?
Signals that reveal an AI agent using a person's account: token type, client ID, timing, API path. Plus the policy that requires agents to carry their own identity.
How do you set content safety thresholds for customer-facing AI agents?
Set harm thresholds per channel and audience using category and severity scales, test them on real traffic, and review false positives weekly.
Tools and MCP
All in this topicWhat are Client ID Metadata Documents in MCP authorization?
CIMD lets an MCP client use an HTTPS URL as its OAuth client_id. How the flow works, how it compares with Dynamic Client Registration, and why it helps allow-listing.
How should MCP servers store API keys and secrets?
MCP configs often hold plaintext tokens. Where secrets end up, why OAuth beats static keys, secret-manager patterns, and how to scan configs for leaks.
Should you use official vendor MCP servers or community-built ones?
Official vendor MCP servers usually win on auth, maintenance and accountability. How to verify which is which, and when a community server is acceptable.
What can real MCP security incidents teach enterprises?
Three documented MCP incidents: a malicious npm server, an mcp-remote RCE and an MCP Inspector flaw. What happened, and the controls each one argues for.
What is the difference between function calling and MCP?
Function calling is how a model asks for a tool; MCP is how tools are discovered, transported and authorised. A layer-by-layer comparison and what to govern.
Which agent APIs let you require approval before an MCP tool runs?
OpenAI Responses and Microsoft Foundry Agent Service support require_approval for remote MCP tools; Anthropic's MCP connector has allow/deny lists only. Compared.
Gateways and cost
All in this topicHow do you upgrade a self-hosted LLM gateway without breaking agents?
Checklist for upgrading a self-hosted LLM gateway: rollout pattern, config and key compatibility, SSO re-checks, provider smoke tests and rollback.
Should you cap max tokens for AI agents at the gateway?
Cap output tokens, input context, calls per task and spend per agent at the gateway. How to set the numbers and spot truncation.
How do you detect anomalies in AI and LLM spend?
Budgets catch overspend in days; anomaly detection catches a runaway agent in an hour. Baselines per agent version, five signals, thresholds and a worked example.
What should an LLM gateway log, and what should it never store?
Log metadata on every request; treat prompts and responses as sensitive data that is sampled, redacted and access-controlled. A field-by-field table.
How does long-context pricing change the cost of AI agents?
Some models charge more per token once a prompt passes 200K tokens. Current examples, a worked agent cost calculation, and six ways to stay under the line.
What are virtual keys in an LLM gateway?
A virtual key is a gateway-issued credential that maps to a provider key plus an owner, budget and model allow-list. How they work and how to rotate them.
How much does retrieved context add to AI agent cost?
Retrieved chunks are often the biggest share of an agent's input tokens. A worked example of chunks × tokens × turns, and the levers that cut it.
When should AI agents use priority or flex processing tiers?
Flex, standard, priority and batch tiers trade price for latency and reliability. Which agent workloads fit each, with real OpenAI and Gemini tier behaviour.
Evidence and frameworks
All in this topicWhat does FINRA expect from broker-dealers using AI agents?
FINRA's 2026 oversight report names AI agent risks and considerations. How supervision, communications and recordkeeping rules apply.
What does Australia's Guidance for AI Adoption expect for AI agents?
Australia's Guidance for AI Adoption sets six essential practices. How each maps to AI agents, with the evidence to keep.
What does OSFI Guideline E-23 mean for AI agents at Canadian banks?
OSFI E-23, effective 1 May 2027, includes AI/ML in its model definition. What that means for inventorying, rating and monitoring AI agents.
Are AI agent conversations business records you must keep?
Some agent transcripts are business records, some are just logs. How to tell them apart, and how retention schedules, legal hold and eDiscovery apply.
Platform Guides
All in this topicHow do you track Claude Code usage across a team with the Analytics API?
What the Claude Code Analytics Admin API returns per user per day, what it misses (subscriptions, Bedrock, Vertex), and how to build a governance report.
What does the OpenAI Responses API store, and how do you turn it off for agents?
What the OpenAI Responses API retains by default (30-day application state), how store=false and encrypted reasoning work, and what ZDR changes for agents.
How do you add guardrails to LangChain agents with middleware?
Which LangChain built-in middleware act as guardrails (PII, human-in-the-loop, call limits, fallback), how hooks work, and what still needs central control.
How do you restrict network egress for AI agent pods on Kubernetes?
Default-deny egress for AI agent pods: NetworkPolicy YAML, DNS rules, why FQDN rules need CNI extensions, and routing model calls through a gateway service.
How do you control which terminal commands Copilot agent mode can run in VS Code?
VS Code settings that control agent terminal commands: autoApprove rules, global auto-approve, URL approvals, sandboxing and enterprise enforcement.
How should you scope RBAC for AI agents that operate Kubernetes clusters?
How to give an AI ops agent Kubernetes access safely: dedicated ServiceAccount, namespaced Roles, verbs to avoid, secrets, exec and audit logging.
How do you govern agents built with the Claude Agent SDK?
How Claude Agent SDK permissions work (hooks, deny, ask, modes, allow, canUseTool), which settings matter for production agents, and what to govern centrally.
What does Agent Engine Threat Detection in Security Command Center detect?
Security Command Center's agent threat detection (preview, Premium) watches Agent Engine runtimes and audit logs. Detector list, gaps and how to route findings.
How do SAP AI Core resource groups isolate AI agent workloads?
SAP AI Core resource groups are the isolation boundary for deployments and executions, capped at 50 per tenant. How to map them to teams, environments and agents.
How do you control which Databricks users and agents can query model serving endpoints?
Databricks serving endpoints use CAN VIEW, CAN QUERY and CAN MANAGE ACLs, plus AI Gateway rate limits per user, group or service principal. A setup guide.
Which Azure OpenAI deployment type keeps agent data in region?
Global, Data Zone and Standard/Regional deployments process prompts in different places. Table of all types, the Azure Policy to block Global, and agent checks.
Which CrewAI settings cap agent iterations, rate and execution time?
CrewAI agents expose max_iter (default 20), max_rpm, max_execution_time and max_retry_limit (default 2). What each caps, what it misses, and sensible values.
How do you use LangGraph recursion limits to stop runaway agents?
LangGraph recursion_limit caps super-steps per run and raises GraphRecursionError. How to set it, pick a value, handle the error and pair it with gateway caps.
How does environment routing keep Copilot Studio makers out of the default environment?
Power Platform environment routing sends Copilot Studio makers to personal managed developer environments. Prerequisites, rule logic, inherited settings and gaps.
How do you secure Langflow before agents built in it reach production?
Patch CVE-2025-3248, disable auto-login, set the secret key, restrict CORS and network access, and inventory flows and credentials in Langflow.
Should AI agents on Google Cloud use service account keys?
Avoid service account keys for agents. Orgs created from 3 May 2024 block them by default; use attached identities, federation or Agent Engine identity.
How do you restrict which Claude Code plugins developers can install?
Use strictKnownMarketplaces, blockedMarketplaces, enabledPlugins and disableSideloadFlags in managed settings to control Claude Code plugins, plus the gaps.
Which secrets can the GitHub Copilot coding agent see?
Copilot cloud agent sees only Agents secrets; COPILOT_MCP_ values reach MCP servers only; Actions secrets are not exposed. How to scope them safely.
Put the framework into practice