AI Frameworks · julep-ai

julep

Julep is a Python framework for building durable, composable AI agents as dataflows that can crash, resume, and retry safely. It provides a decorator-based authoring model (@flow) that compiles to an intermediate representation, with optional runtime support via Temporal or DBOS for production durability.

Talk to DEV.co Discuss Implementation

Source: GitHub — github.com/julep-ai/julep

6.6k

GitHub stars

972

Forks

Python

Primary language

Apache-2.0

License (OSI-approved)

Key facts

Objective fields from the source. Values we can't verify are shown as “Unknown” rather than guessed.

Field	Value
Repository	julep-ai/julep
Owner	julep-ai
Primary language	Python
License	Apache-2.0 — OSI-approved
Stars	6.6k
Forks	972
Open issues	3
Latest release	Unknown
Last updated	2026-07-05
Source	https://github.com/julep-ai/julep

What julep is

Julep v3 is a ground-up rewrite offering define-by-construction dataflows using @flow decorators and registered tools/reasoners. The core is dependency-free; optional extras add Temporal/DBOS durability, OpenTelemetry/Langfuse observability, multi-provider LLM support, and WASM-sandboxed pure functions. A CLI (julep) provides agent discovery, validation, local execution, and deployment across a module.

Quickstart

Get the julep source

Clone the repository and explore it locally.

terminalbash

1git clone https://github.com/julep-ai/julep.git2cd julep3# follow the project's README for install & configuration

Need it deployed, integrated, or customized instead? DEV.co ships production installs.

Best use cases

Long-running agent workflows with fault tolerance

Use Julep for AI agents that must handle crashes, network failures, or step retries gracefully. The Temporal/DBOS integration provides durable execution and automatic resume-on-failure without manual recovery logic.

Multi-step orchestration with tool-call safety

Build complex agent flows where tools must be explicitly registered and authorized. The frozen IR and registration pattern prevent models from calling unintended tools, reducing misuse risk in production scenarios.

Local-first agent development and testing

Leverage the dependency-free core and dry_run() for rapid iteration without API keys or external services. The @flow definition-time execution model and CLI tooling streamline agent design before production deployment.

Implementation considerations

RC status: v3 ships as --pre release; production use warrants thorough testing, monitoring of upstream changes, and clear rollback plans.
Temporal/DBOS setup required for durability: base install is authoring/compile only. Evaluate Temporal or DBOS operational overhead for your infrastructure.
Tool and reasoner registration is explicit; ensure all models, API keys, and tool definitions are correctly wired before deployment.
Local dry_run() with fake reasoners is fast for testing, but production behavior depends on LLM provider latency and external tool availability.
CLI tooling assumes a directory-based agent module structure; refactor existing codebases to align with Julep's layout expectations.

When to avoid it — and what to weigh

Simple, stateless chatbot or request-reply patterns — Julep's dataflow and durability model add complexity unnecessary for straightforward prompt-response use cases. Lightweight alternatives may be more suitable for basic conversational AI.
Mature, stable production systems already deployed — Julep v3 is a ground-up rewrite with no migration path from v1, and ships as a release candidate (--pre flag required). Adoption in locked-down production environments requires significant vetting and carries upgrade risk.
Teams without Python expertise or avoiding new SDKs — Julep requires Python and introduces a custom flow syntax and CLI. Organizations standardized on different languages or agent frameworks will face friction and ramp-up overhead.
Strict dependency minimalism in embedded or lightweight environments — While the core is dependency-free, production use typically requires Temporal, DBOS, or observability extras that introduce heavyweight dependencies. Lightweight deployments may find this limiting.

License & commercial use

Apache-2.0 (Apache License 2.0) is a permissive, OSI-approved license allowing commercial use, modification, and distribution with minimal restrictions (patent protection clause, state changes, and license preservation required).

Apache-2.0 permits commercial use and distribution. No commercial license, vendor lock-in, or usage fees are indicated. Derivative works and proprietary applications are allowed if license and copyright notices are retained. Confirm with legal review for risk-sensitive deployments.

DEV.co evaluation signals

Editorial assessment — not user reviews. Directional, with an explicit confidence level.

Signal	Assessment
Maintenance	Active
Documentation	Adequate
License clarity	Clear
Deployment complexity	Moderate
DEV.co fit	Good
Assessment confidence	Medium

Security considerations

Tool registration pattern enforces explicit authorization, reducing risk of model-triggered unintended function calls. WASM sandboxing (wasm extra) isolates pure function execution. Credential management (LLM API keys, Temporal, DBOS) must be secured via environment variables or secret stores—Julep does not provide built-in secrets management. No security audit or disclosure policy is documented.

Alternatives to consider

LangGraph (LangChain)

Stateful graph-based orchestration with tighter LLM ecosystem integration. Mature, well-documented, with built-in memory and state persistence. No Temporal dependency; simpler for teams already in LangChain.

CrewAI

Higher-level multi-agent framework with built-in role/task abstractions and tool orchestration. Easier onboarding for non-engineers; less control over durability and flow architecture.

Temporal + custom Python

Direct use of Temporal for durability without a framework layer. Maximum control and minimal abstraction; requires more boilerplate and operational expertise but avoids vendor-specific syntax.

Software development agency

Build on julep with DEV.co software developers

Start with pip install --pre julep. Run the quickstart without an API key, then explore durability with Temporal or DBOS for production workloads.

Talk to DEV.co

Related open-source tools

Surfaced by semantic similarity across the DEV.co open-source index.

agents-cli Agently hermes-agent

Related on DEV.co

Explore the category and the services that help you build with it.

AI Frameworks AI application development API development Cloud deployment Talk to DEV.co

julep FAQ

Is Julep v3 compatible with v1 agents?

No. Julep v3 is a ground-up rewrite; v1 and v3 are different products. v1 is preserved on the v1 branch; there is no migration path.

Can I use Julep without Temporal or DBOS?

Yes, for local authoring and testing. The base install includes define-by-construction (@flow) and dry_run(). Temporal/DBOS extras are optional and required only for durable, fault-tolerant production execution.

How do I prevent models from calling unregistered tools?

Register all tools and reasoners explicitly with @tool/@flow decorators and pass them to deploy(..., tools=[...], reasoners=[...]). The frozen IR ensures only registered tools are exposed; unregistered calls are denied.

What LLM providers are supported?

The providers extra enables multi-provider support via any-llm. Supported providers include Anthropic, OpenAI, and others. Check any-llm documentation for the full list; custom provider integration requires @tool decorators.

Software development & web development with DEV.co

From first prototype to production, DEV.co delivers software development services around tools like julep. Our software development agency staffs experienced software developers and web developers for custom software development, web development, integrations, and ongoing support across ai frameworks and beyond.

Ready to build resilient AI agents?

Start with pip install --pre julep. Run the quickstart without an API key, then explore durability with Temporal or DBOS for production workloads.

Talk to DEV.co Schedule a Consultation