Julep
Verified Oct 2, 2026 ActiveOpen-source Python framework for agents written as durable flows that resume after a crash, retry safely and log every step.
Listed on EffectHub ·
Pricing
- Pricing model
- 100% Free
The library is free under Apache-2.0 and installs with pip. The older hosted Julep platform's pricing is not documented anywhere we could reach, so no paid price is recorded.
Use tool ↗About Julep
Julep 3 lets developers describe an agent as a flow in ordinary Python: tools, pure functions, model calls, branches, fan-out, retries and timeouts compile into a fixed graph. Flows can be dry-run locally with fake model replies, then deployed with an optional Temporal layer so a run that crashes resumes where it stopped. Tool access is allow-listed, so the model cannot call anything that was not declared, and each step can be explained afterwards from the recorded run. It is Apache-2.0 licensed and currently published as a release candidate; version 1 of the project is kept on a separate branch with no migration path.
- API
- No
- Last verified
- Oct 2, 2026
Verified sources
- README: Julep 3 is durable composable flows, Apache-2.0, release candidate needing --pre, v1 kept on a separate branch with no migration pathgithub.com ↗
- PyPI release history shows 3.0.0 release candidates from July 2026pypi.org ↗
- julep.ai redirects to the memory.store productjulep.ai ↗
- Docs site describes the earlier hosted platform for agents with memory and long-running tasksdocs.julep.ai ↗
Is Julep legit?
We don’t label tools safe or unsafe. Below are the records we checked ourselves, each with the date we read it.
- Last checked
- Oct 2, 2026The site answered
- On the market since
- 2024The vendor's own material, read by us
- Operated by
- Julep AIOur research, card verified Oct 2, 2026
Pros & cons of Julep
Julep suits Python developers who want agent logic that is testable offline and survives crashes, and who are comfortable with a graph-style API. It is a release candidate and a ground-up rewrite, so anyone on the earlier hosted platform has to start over. Pick a mainstream framework if you want stable APIs and a large ecosystem today.
Pros
- Flows are plain Python, so they can be unit-tested with fake reasoners and no API key
- Tool allow-listing denies any call the model was not explicitly given
- Crash recovery and safe retries come from the optional Temporal layer
- Apache-2.0 licence and an active repository with several thousand stars
Cons
- Version 3 is still a release candidate and needs the pre-release flag to install
- Version 3 is a full rewrite with no migration path from version 1
- The company's main site now points to a different product, so commercial support and hosting are unclear
Best for
- Backend teams building long-running agent workflows that must retry and resume
- Developers who want to dry-run agents with deterministic fake model replies before deploying
Not for
- Teams that need a stable 3.0 release and a migration path from Julep 1
- People looking for a no-code agent builder or a hosted dashboard
Watch out: The julep.ai domain now redirects to the Memory Store product from the same team, and the docs site still carries the old platform's pages next to the new framework, so check which version a tutorial targets.
Top alternatives to Julep
Julep head-to-head
Side-by-side fact sheets against the competitors our research or their makers name — pricing, platforms, API and both verdicts on one page.