Docs

Github

Dashboard

Examples

Discord

Book demo

Open Source

Github

★

4.7K

Deploy serverless
AI workflows
at scale

in minutes, not weeks.

Open-source API to build
infinitely scalable, durable and long running AI pipelines
using agents, tasks and tools

Deploy your first workflow

Get a demo

CUSTOMER STORIES

Vidyo built personalised marketing intelligence for thousands of users using posting trends and current top trends.

Iterate and deploy AI features super fast by bringing data from different sources in one place.

CUSTOMER STORIES

Vidyo built personalised marketing intelligence for thousands of users using posting trends and current top trends.

Iterate and deploy AI features super fast by bringing data from different sources in one place.

CUSTOMER STORIES

Vidyo built personalised marketing intelligence for thousands of users using posting trends and current top trends.

Iterate and deploy AI features super fast by bringing data from different sources in one place.

Connect with any AI model,
API or data source

Seamlessly integrate with your existing stack and favorite AI models.

LLM Platforms

Languages

Integrations

Any REST API

Build with Julep

Create an agent

Define agents that can talk to users inside of a session

Add tools

Equip agents with tools - web search, API calls, or custom integrations

Define your tasks

Define multi-step processes in YAML with decision trees, loops, and parallel execution

Deploy

Execute production-grade workflows with one command

agent = julep.agents.create(

name="Spiderman",

about="AI that can crawl the web and extract data",

model="gpt-4o-mini",

default_settings={

"temperature": 0.75,

"max_tokens": 10000,

"top_p": 1.0,

}

)

Create an agent

Define agents that can talk to users inside of a session

Add tools

Equip agents with tools - web search, API calls, or custom integrations

julep.agents.tools.create(

agent_id=agent.id,

name="my_weather_tool",

integration={

"provider": "weather",

"setup": {

"openweathermap_api_key": "OPENWEATHERMAP_API_KEY"

}

)

Define your tasks

Define multi-step processes in YAML with decision trees, loops, and parallel execution

iimport yaml

task = julep.task.create("""

name: Weather Reporting Task

main:

- tool: spider_crawler

arguments:

url: _.url

- prompt: |

Here’s a crawl of {{inputs[0].url}}:

Create a short summary (~100 words) of the website.

""") - city

Deploy

Execute production-grade workflows with one command

execution = julep.executions.create(

task_id=task.id,

input={

"url": "https://spider.cloud"

}

)

agent = julep.agents.create(

name="Cristopher",

about="An agent specialized in weather reporting",

model="gpt-4o-mini",

)

Create an agent

Define agents that can talk to users inside of a session

Add tools

Equip agents with tools - web search, API calls, or custom integrations

Define your tasks

Define multi-step processes in YAML with decision trees, loops, and parallel execution

Deploy

Execute production-grade workflows with one command

agent = julep.agents.create(

name="Spiderman",

about="AI that can crawl the web and extract data",

model="gpt-4o-mini",

default_settings={

"temperature": 0.75,

"max_tokens": 10000,

"top_p": 1.0,

}

)

Real people, Real results

Julep to me sits in the intersection of what Zapier did for simplifying workflows, and what Vercel did for simplifying shipping. The new possibilities are endless. I am excited abut what all can be done with this tech at ES.
Suryansh Tibarewal
CEO, EssentiallySports
For Vidyo, we shipped 6 months worth of product development in 6 hours. GAME CHANGER!!
Vedant Maheshwari
CEO, Vidyo.ai
We tried building our ai stack where we needed simple automated computer use. we spent months trying to get it right, but our development time got cut to a couple weeks because most of the infra orchestration was handled by julep. all that we had to do was refine our prompts
Madhavan
CEO, Reclaim Protocol
"Same experience as reviewing designs on Figma with my designer. Sat with my developer once, and we kept brainstorming, testing, prototyping and shipping at once. I got control back in my hands as a non-developer."
Akshay Pruthi
CEO, Calm Sleep
Working with Julep accelerated Ukumi, by enabling us to focus on our core strengths and letting Julep navigate the complex web of calls to ai models we could do more with less.
Sachin Gaur
Cofounder, Ukumi AI

Build AI pipelines
in minutes.
Scale to millions.

Rapid Prototyping

"From idea to demo in minutes" using built-in RAG, in-house state management and external tools.

Production Ready

"Go live instantly with managed infrastructure" and features such as long-running tasks, automatic retries and error handling

Modular Design

"Build features like Lego blocks" by connecting to any external API, switch between LLMs and add custom tools

Infinite Scale

Handle millions of concurrent users with automatic scaling and load balancing

Future Proof

"Add new AI models anytime" "Integrate with any new tool or API"

Complete Control

"Full visibility into AI operations" "Manage costs and performance easily"

Questions?
Answers.

What is Julep?

Julep is a serverless AI workflow platform that lets data science teams build, iterate on, and deploy multi-step AI pipelines using a simple, declarative YAML-based configuration.

What is Julep?

Julep is a serverless AI workflow platform that lets data science teams build, iterate on, and deploy multi-step AI pipelines using a simple, declarative YAML-based configuration.

What is Julep?

Julep is a serverless AI workflow platform that lets data science teams build, iterate on, and deploy multi-step AI pipelines using a simple, declarative YAML-based configuration.

Who is Julep for?

Julep is designed for data scientists and ML engineers—especially in enterprise settings—who want to quickly move from AI prototypes to production without the need for heavy backend engineering.

Who is Julep for?

Julep is designed for data scientists and ML engineers—especially in enterprise settings—who want to quickly move from AI prototypes to production without the need for heavy backend engineering.

Who is Julep for?

Julep is designed for data scientists and ML engineers—especially in enterprise settings—who want to quickly move from AI prototypes to production without the need for heavy backend engineering.

How does Julep's approach differ from typical AI development?

While most platforms focus on prompt engineering and chaining LLM calls, Julep brings software engineering discipline to AI development. Through our 8-Factor Agent methodology, we treat AI components as proper system elements: - Prompts as Code: Track prompts separately from application code, enabling systematic improvements and proper versioning. - Clear Tool Interfaces: Define explicit interfaces for all tool interactions, making capabilities modular and maintainable. -Model Independence: Treat model providers as external, replaceable resources to avoid vendor lock-in and enable easy switching. -Context Management: Explicitly define how application and user state is managed and reduced. -Ground Truth Examples: Maintain clear examples of expected prompt results for validation and testing. -Structured Reasoning: Separate processes into deliberative (planned, multi-step) and impromptu (quick response) reasoning. -Workflow-Based Model: complex processes as clear workflows rather than chains of prompts. -Full Observability: Save execution traces for debugging, monitoring, and continuous improvement.

How does Julep's approach differ from typical AI development?

How is Julep different from agent frameworks?

While LangChain is great for creating sequences of prompts and managing model interactions, Julep is built for creating persistent AI agents with advanced task capabilities. Think of LangChain as a tool for prompt chains, while Julep is a complete platform for building production-ready AI systems with complex workflows, state management, and long-running tasks.

How is Julep different from agent frameworks?

What types of workflows can I build with Julep?

You can build multi-step AI workflows that include decision-making, loops, parallel processing, state management, retries, and integrations with external tools and APIs.

What types of workflows can I build with Julep?

You can build multi-step AI workflows that include decision-making, loops, parallel processing, state management, retries, and integrations with external tools and APIs.

What types of workflows can I build with Julep?

You can build multi-step AI workflows that include decision-making, loops, parallel processing, state management, retries, and integrations with external tools and APIs.

How does Julep handle scaling and fault tolerance?

Julep automatically scales to handle thousands to millions of executions and is built with robust fault tolerance, state management, and long-running execution capabilities to ensure reliable, production-grade workflows.

How does Julep handle scaling and fault tolerance?

Is Julep secure and enterprise-grade?

Yes. Julep is built with enterprise needs in mind. We offer private deployments, robust security measures, and compliance features that ensure our platform is reliable and production-ready.

Is Julep secure and enterprise-grade?

Yes. Julep is built with enterprise needs in mind. We offer private deployments, robust security measures, and compliance features that ensure our platform is reliable and production-ready.

Is Julep secure and enterprise-grade?

Yes. Julep is built with enterprise needs in mind. We offer private deployments, robust security measures, and compliance features that ensure our platform is reliable and production-ready.

What are Julep’s pricing models?

We offer a generous free tier for experimentation. For paid plans, we have two main models: a usage-based pricing model (charging per workflow step executed) for our cloud platform, and dedicated enterprise/on-prem deployments for large organizations.

What are Julep’s pricing models?