Orkes Conductor
Battle Tested Platform for Agent and Workflow Orchestration
Orchestrate agents, services and human decisions in one durable execution. Keep your code and frameworks. Conductor handles the state, coordination and recovery.
See how it works- Agent decision Plan the deployment Selects the deployment tool
- Required human approval Approve the release Execution waits for approval
- Defined service call Deploy the service Runs your existing deployment code
Execution state saved at every step
Developer first
Build agents and workflows your way.
Use your coding agent with Conductor Skills, write with SDKs or build in the visual editor. Keep your existing services and agent frameworks.
Build with your coding agent.
Build, run and debug workflows, agents and workers from Claude Code, Codex, Cursor and other coding agents.
Get Conductor SkillsBuild a deployment agent that requires human approval before deploying payments-api.
from conductor.ai.agents import (
Agent, AgentRuntime, tool
)
@tool(approval_required=True)
def deploy_to_production(service, version):
# Your existing deployment function.
return rollout(service, version)
agent = Agent(
name="production_deploy_agent",
model="openai/gpt-4o-mini",
tools=[deploy_to_production],
instructions="Deploy requested releases.",
)
with AgentRuntime(config) as runtime:
runtime.deploy(agent)
runtime.start(
agent, "Deploy payments-api v2.4.1"
)
runtime.serve(agent)
The execution graph records agent turns, tool calls, approvals and retries.
Build or bring your agents.
Use the Conductor Agent SDK, or bring agents built with LangGraph, OpenAI Agents SDK or Google ADK. Agents and workflows can call each other.
Bring your agentBring determinism to your agents.
The agent selects the deployment tool. The approval rule in your code controls when that tool can run.
Durable execution
A failed step doesn’t erase the work.
Conductor saves execution state as work progresses. Retry a failed task or rerun from a chosen step using the outputs already recorded.
Long-running processes can wait for a human, an event or an external system, then continue with their state intact.
Explore execution and recovery- Plan the deployment Agent decision recordedSaved
- Approve the release Human approval recordedSaved
- Deploy the service Retry after the service is availableRetry needed
Earlier steps and their results stay in place.
Visibility and control
Understand the execution.
Act on what you find.
Inspect inputs, outputs, agent decisions and tool calls in one execution history. Use the Orkes Assistant to explain failures and suggest fixes.
See the Orkes AssistantProven in production.
Built on Conductor, originally created and open sourced at Netflix. Orkes adds the governance, operational capabilities and support teams need for business-critical agents and workflows.
Govern every execution.
Manage access with RBAC and SSO. Keep secrets out of code, require approvals and retain audit trails across agents and workflows.
Build on an open foundation.
Conductor and its SDKs are open source. Build with your languages and services, with the visibility and control of Orkes Conductor.
Explore Conductor OSSWork with the team behind Conductor.
Get enterprise support and guidance from Orkes, from your first deployment through ongoing production operations.
Explore Orkes CloudLinkedIn runs its agent platform on Conductor.
Agents save progress, pause for human approval and hand work to other agents on the same durable execution engine.
Twilio Segment
“We stopped maintaining a workflow engine and started building on one.”
Seven handoffs.
One Conductor workflow.
Bumble runs trust-and-safety reports through Conductor, combining AI steps and human moderation with automatic retries and recorded decisions.
Orkes Conductor
Build agents and workflows
your business
can depend on.
Start building for free, or see Conductor in action with your team.