Workflow · Production

Jockey: Twelve Labs' open-source conversational video agent built on LangGraph

The problem

Jockey v1.0, built on the legacy LangChain AgentExecutor, lacked the granular control and scalability needed for complex multi-step video processing workflows.

First attempt

Jockey v1.0 on the legacy LangChain AgentExecutor lacked flexible API support for custom cognitive architectures and could not provide the low-level node-level control required for complex video workflows.

Workflow diagram · grounded in source
1
User query submitted
trigger
“the decision-making process from the initial query input”
2
Query complexity determined
ai_action
“LangGraph's analysis to determine query complexity, and the subsequent routing to either a simple text response or a more complex chain of video processing steps”
3
Supervisor routes task
routing
“The Supervisor acts as the central coordinator, responsible for routing tasks between different nodes and managing the overall workflow. It receives user input and determines the next course of action, whether it's engaging the Planner f…”
4
Planner creates step-by-step plan
ai_action
“The Planner is called upon by the Supervisor to create detailed, step-by-step plans for complex user requests. This component is essential for breaking down intricate tasks into manageable steps that can be executed by the Workers.”
5
Instructor generates task instructions
ai_action
“The Instructor, which generates precise and complete task instructions for individual workers based on the Planner's strategy”
6
Workers execute video tasks
ai_action
“The Actual Workers, which are specialized agents that execute the instructions using their available tools. These include Video Search, Video Text Generation, and Video Editing capabilities.”
7
Final results presented to user
output
“presenting the final results to the user”
Reported outcome

Jockey v1.1 with LangGraph provides granular control over each workflow step, optimizes token usage, enables more accurate node response guidance, and supports scalable production deployment via LangGraph Cloud.

Reported metrics
Token usage efficiencyoptimizes token usage
Agent reliabilityincreasing the reliability of agents
Reported stack
LangGraphLangGraph CloudLangGraph StudioTwelve Labs APIsLLMs
Source
https://blog.langchain.dev/jockey-twelvelabs-langgraph/
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Jockey v1.1 with LangGraph provides granular control over each workflow step, optimizes token usage, enables more accurate node response guidance, and supports scalable production deployment via LangGraph Cloud.

What tools did this team use?

LangGraph, LangGraph Cloud, LangGraph Studio, Twelve Labs APIs, LLMs.

What results were reported?

Token usage efficiency: optimizes token usage; Agent reliability: increasing the reliability of agents (source-reported, not independently verified).

What failed first in this deployment?

Jockey v1.0 on the legacy LangChain AgentExecutor lacked flexible API support for custom cognitive architectures and could not provide the low-level node-level control required for complex video workflows.

How is this workflow AI workflow structured?

User query submitted → Query complexity determined → Supervisor routes task → Planner creates step-by-step plan → Instructor generates task instructions → Workers execute video tasks → Final results presented to user.