---
title: How do you govern agents in IBM watsonx Orchestrate?
url: https://opencontroller.lyzr.com/blog/govern-ibm-watsonx-orchestrate-agents
published: 2026-11-12
updated: 2026-11-12
author: Lyzr
---

# How do you govern agents in IBM watsonx Orchestrate?

**Govern watsonx Orchestrate agents with its governed catalog (certify before you publish), its runtime monitoring (powered by watsonx.governance), and central identity and credential management. Then decide how Orchestrate fits with agents that live on other platforms, because most enterprises will run agents in several places at once.**

IBM watsonx Orchestrate is IBM's platform for building, running and orchestrating AI agents. IBM has added governance features in two main steps: runtime monitoring, announced as generally available in December 2025, and what IBM calls an "agentic control plane", announced in May 2026. This guide summarises what IBM documents and how to use it.

## What governance features does watsonx Orchestrate have?

| Area | What IBM describes | How to use it |
|---|---|---|
| Catalog | A governed catalog of agents and tools with performance metrics; certification and publishing workflows; assets tracked from development to production | Make certification a required gate; no uncertified agent in production |
| Agent types | IBM native, Langflow, LangGraph and A2A-protocol agents | Record which framework each agent uses and who maintains it |
| Identity | Identity and credentials managed from a central point | Remove credentials embedded in individual agents |
| Enforcement | Runtime control enforcement, security and compliance guardrails, monitoring of control triggers, audit logging | Define which controls block and which alert |
| Isolation | Isolation and segmentation to keep agents within boundaries | Separate high-risk agents (finance, HR) from general ones |
| Gateway | A unified AI gateway for oversight of agents, models and tools in production | Route model and tool traffic through it |
| Evaluation | Build-time and runtime evaluation, simulation, tracing | Test each version against realistic scenarios before release |

## How does runtime monitoring work?

IBM's December 2025 announcement says agent monitoring in watsonx Orchestrate is "powered by integration with IBM watsonx.governance." Once an agent is in production, telemetry is captured from every interaction. The dashboard shows:

- **Agent-level health**: success rates, latency, token usage and cost drivers.
- **Message and tool insights**: where tools misfire or flows break, using trace-backed signals.
- **Conversation insights**: transaction-level detail for individual interactions.

IBM places this inside its Agent Ops approach (observability, evaluation and optimisation), with evaluation metrics such as faithfulness, tool call relevancy and token count. Monitoring works across single agents or the whole catalog.

## What should your watsonx Orchestrate governance checklist include?

1. **Gate publishing on certification.** Define what certification means: owner named, tools reviewed, evaluations passed, data scope documented.
2. **Name an owner for every catalog entry**, including imported Langflow, LangGraph and A2A agents.
3. **Control who can build.** Check how builder and publisher roles are configured in your tenant, and limit publishing to a small group.
4. **Centralise credentials.** Move tool credentials into central management and rotate them.
5. **Set thresholds.** Pick alert thresholds for success rate, latency and cost per agent, and route alerts to an on-call owner.
6. **Track faithfulness.** For agents that answer from documents, monitor faithfulness in production, not just in testing.
7. **Map controls to frameworks.** Record which Orchestrate controls produce evidence for NIST AI RMF, ISO/IEC 42001 or the EU AI Act. Our [framework mapping guide](/blog/how-to-map-ai-agent-controls-to-compliance-frameworks) shows the method.
8. **Plan the exit.** Know how you would export catalog records, traces and evaluation results if you add or change platforms.

## Where do watsonx Orchestrate controls stop?

Orchestrate governs agents that are registered in, or routed through, Orchestrate. Agents built in Copilot Studio, Agentforce, Bedrock AgentCore or on developer laptops are outside its view unless they are brought into its catalog. That is normal for any single platform. The practical question is where your system of record for all agents lives, and whether it can see agents on every platform without moving them. See the [cross-platform governance guide](/blog/how-to-govern-ai-agents-across-platforms) and the [platform evaluation checklist](/blog/ai-agent-governance-platform-evaluation-checklist).

## How Opencontroller does this

Opencontroller by Lyzr is designed for control without migration: it discovers agents, models and API keys across AWS, Azure and Google Cloud, connects to platforms such as Salesforce, SAP, Databricks and HubSpot, and keeps one governed registry of agents, versions, tools and MCP servers. It does not have a native watsonx Orchestrate connector; agents there can be registered, and those that route model traffic through the OpenAI-compatible LLM Gateway can be observed and controlled in the path. [Book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Is watsonx Orchestrate agent monitoring generally available?**
Yes. IBM announced general availability of agent monitoring in watsonx Orchestrate in December 2025, powered by integration with watsonx.governance.

**Can watsonx Orchestrate govern agents not built in IBM tools?**
IBM says its catalog supports IBM native, Langflow, LangGraph and A2A-protocol agents, with broader interoperability planned.

**What metrics does watsonx Orchestrate monitor?**
IBM lists success rates, latency, token usage and cost drivers, with tool-level and conversation-level drill-down.

## Key facts

IBM announced general availability of agent monitoring in watsonx Orchestrate in December 2025, powered by integration with watsonx.governance; it tracks success rates, latency, token usage and cost drivers with tool and conversation drill-down. In May 2026 IBM described an agentic control plane in Orchestrate: governed catalog with certification and publishing workflows; support for IBM native, Langflow, LangGraph and A2A agents; central identity and credentials; runtime control enforcement and audit logging; isolation; a unified AI gateway; build-time and runtime evaluation. Opencontroller by Lyzr offers cross-platform discovery and a governed registry without migration; pricing starts from $250K per deployment per year.

## Related

- [How do you map AI agent controls to compliance frameworks?](https://opencontroller.lyzr.com/blog/how-to-map-ai-agent-controls-to-compliance-frameworks.md)
- [How to evaluate an AI agent governance platform: 30 questions for your RFP (2026)](https://opencontroller.lyzr.com/blog/ai-agent-governance-platform-evaluation-checklist.md)
- [How do you register an AI agent your team doesn't host?](https://opencontroller.lyzr.com/blog/register-ai-agents-you-dont-host.md)
- [More on Platform Guides](https://opencontroller.lyzr.com/blog/cluster/platform-guides)
