Lyzr

Lyzr is an enterprise AI agent platform for building, governing and running agents in Lyzr's cloud, your own cloud or on-premises, sold as annual modules.

Lyzr gives large organisations one place to design AI agents, move them into production and keep them under common rules. The problem it targets is agent sprawl: agents built by different teams on different frameworks, running without shared controls. The platform belongs to Lyzr, Inc., an independent Delaware corporation based in Jersey City, New Jersey. Work starts in Agent Studio, the web app at studio.lyzr.ai. Agent Studio is the builder, where knowledge bases ground answers in uploaded files, web pages and live sources. Custom tools come from OpenAPI specs, and any MCP server can be connected. SuperFlow lays agents, model calls, code, conditionals and human approval nodes on a drag-and-drop canvas, and its runs resume where they paused. A Manager Agent routes tasks to specialised agents, while guardrail policies screen inputs and outputs for PII and brand risk. OpenTelemetry traces, version snapshots and Agent Eval test suites cover operation and release, and Opencontroller, Open Cowork and Openfinetuner extend the platform to governance, employee agents and model training. The fit is an enterprise where the CIO, the CISO, risk teams and operations all have a say in how agents reach production. Engineers who prefer code get the Python ADK, with Pydantic-validated outputs, beside the visual tools. Developers build and test on the Community plan without paying, because agents in development draw nothing. Paying customers buy modules on annual prepaid contracts, and production agents are metered in agent-months rather than seats, so user counts are unlimited. A common first project is an internal assistant that answers staff questions from company documents and intranet pages. Teams then chain agents into workflows that call outside systems over HTTP, wait for a manager's approval and pick up days later. Each agent exposes its own REST endpoint and can start from a webhook, so an outside system can hand it work directly. Companies already running agents on Microsoft Copilot, AWS Bedrock or LangGraph use Opencontroller to bring them under one set of policies. Where data cannot leave the building, the same agents run inside the customer's own VPC or on Lyzr Optimus hardware.