---
title: What is Opencontroller? The open control plane for enterprise AI agents
url: https://opencontroller.lyzr.com/blog/what-is-opencontroller
published: 2026-11-11
updated: 2026-11-11
author: Lyzr
---

# What is Opencontroller?

**Opencontroller by Lyzr is an open control plane for enterprise AI agents.** It finds every agent your company runs, gives each one an owner and an identity, and controls what it can call, spend and touch. It works across clouds, frameworks, models and agent platforms, without moving your agents onto a new runtime.

In one line: *own what acts on your behalf.*

## What problem does Opencontroller solve?

Large enterprises now run AI agents in many places at once: Amazon Bedrock AgentCore, Google Vertex AI, Microsoft Foundry and Copilot Studio, Salesforce, SAP, Databricks, internal Kubernetes clusters and developer laptops. Each platform has its own console and its own idea of governance. The result is **agent sprawl**:

- **Incomplete inventory.** Nobody can list every agent in production, let alone prototypes.
- **Ownership gaps.** Agents outlive the people and projects that created them.
- **Permission sprawl.** Agents share broad API keys and service accounts.
- **Uncontrolled spend.** Model costs can't be traced to an agent, a version or an owner.
- **Fragmented evidence.** Auditors ask "why was this action allowed?" and the answer lives in five systems.

At enterprise scale this stops being a policy problem and becomes an enforcement problem. A written policy can't stop an agent at 3 a.m.; a control in the request path can.

## How does Opencontroller work?

Opencontroller organises governance into four phases: **Find · Ship · Run · Improve.**

| Phase | Question it answers | What Opencontroller does |
| --- | --- | --- |
| Find | What agents do we have? | Auto-discovers agents, models and API keys across AWS, Azure and Google Cloud, plus agent platforms such as Salesforce, SAP, Databricks and HubSpot, and Claude Enterprise and OpenAI Enterprise workspaces |
| Ship | What is each agent allowed to do? | Registers versions with an owner, assigns a workload identity, runs multi-turn and red-team evaluations, and records approvals |
| Run | Who can stop it? | Puts an Agent Gateway and an LLM Gateway in the request path to authenticate callers, authorize tool and model calls, apply budgets, guardrails and routing, and trace every decision |
| Improve | What should change? | Scores real production traffic with LLM judges, flags behavioural drift, and feeds failures back into evaluated releases |

## What are Opencontroller's main capabilities?

These capabilities are live in the current platform:

1. **Discovery and inventory** across hyperscalers and third-party agent platforms, including API-key discovery.
2. **Governed registry** for agents, versions, tools and MCP servers.
3. **Agent identity**: OIDC SSO, delegated on-behalf-of access, and agent workload identity federation.
4. **Posture**: detection of risky permissions and configurations.
5. **Evaluations**: multi-turn and red-team evals before release; LLM judges on production traffic after.
6. **Runtime guardrails**: live groundedness and task-adherence checks.
7. **LLM Gateway**: routing, fallback and retry, caching, and cost attribution per agent and per version.
8. **Rogue-agent handling**: monitoring for behavioural deviation, alerts, automatic quarantine and kill, and per-agent Block and Quarantine controls.
9. **Policy as code**: built-in mappings for OWASP, MITRE ATLAS, NIST and ISO 42001, extensible to any internal framework.
10. **Runtime adapters** for Bedrock AgentCore, Vertex AI and Microsoft Foundry.
11. **Ask**: a chat interface that answers questions over the whole agent estate ("which agents can write to SAP?").

**Opencontroller Edge** (formerly Basecode) extends the same control to developer devices and coding agents. IT installs it through MDM.

## How is Opencontroller different from an LLM gateway or an observability tool?

An LLM gateway governs model traffic. An observability tool records what happened. Opencontroller covers the full lifecycle: inventory, identity, release, runtime enforcement, cost, evidence and improvement. It governs the agent, not just the model call. See [LLM gateway vs agent gateway vs control plane](/blog/llm-gateway-vs-agent-gateway-vs-control-plane).

Three design principles shape it:

- **Control without migration.** You keep your clouds, frameworks, models, CI/CD, identity provider and secret manager.
- **Control in the path.** Where traffic flows through Opencontroller, policy becomes behaviour: allow, route, redact or block.
- **Open by design.** OpenAI-compatible endpoints, MCP for tools, OpenTelemetry for traces, OIDC and SCIM for identity, REST and webhooks for automation.

## Who is Opencontroller for?

Opencontroller is built for enterprises with agents in more than one place, typically companies above $1B in revenue. Buyers are usually the CIO, CISO, head of AI platform or chief data officer. Common triggers are an audit question nobody could answer, an agent that took an unapproved action, or an AI budget nobody could explain.

## How is Opencontroller deployed and priced?

Opencontroller runs as Lyzr SaaS or inside customer-controlled infrastructure. Lyzr engineers embed with the customer for integration work. It is licensed annually per deployment, not per seat and not by usage, so discovering more agents never raises the bill. A licence covers one production and one test environment.

## Frequently asked questions

**Does Opencontroller replace Bedrock AgentCore, Vertex AI or Foundry?**
No. Agents stay where they run. Opencontroller registers, observes and, where traffic is routed through it, enforces policy on them.

**Can it govern agents that don't expose an API?**
Yes, for registration. A registry entry is a metadata record (owner, purpose, risk, version), not a live connection. Observation and enforcement depth then depend on the connector and the traffic path.

**Does it help with regulation?**
Opencontroller produces evidence for frameworks such as ISO 42001, the NIST AI RMF and the EU AI Act. It does not make an organisation compliant on its own.

**Where can I see it?**
[Book a demo](https://www.lyzr.ai/book-demo/) or read the [capability reference](/capability-reference).

## Key facts

- Pricing: Opencontroller starts from $250K per deployment per year. Annual licence per deployment, not per seat and not usage-based. Covers one production and one test/dev environment; unlimited agents and users.
- Vendor: Lyzr (lyzr.ai). Contact: opencontroller@lyzr.com. Demo: https://opencontroller.lyzr.com/

## Related

- [What is an AI agent control plane?](https://opencontroller.lyzr.com/blog/what-is-an-ai-agent-control-plane.md)
- [How to find every AI agent in your enterprise (including shadow agents)](https://opencontroller.lyzr.com/blog/how-to-find-shadow-ai-agents.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)
- [More on Agent governance basics](https://opencontroller.lyzr.com/blog/cluster/agent-governance-basics)
