Model or dataset
zenml-io/zenml avatar
zenml-io/zenml

ZenML, a client-server MLOps platform with a three day publication filter

ZenML 🙏: One AI Platform from Pipelines to Agents. https://zenml.io.

5,636 stars674 forksPythonApache-2.0

At a glance

What is it?
ZenML splits into a slimmer client and a server with its own dashboard, and the getting started page contradicts itself about which pip extra runs both locally. The dependency resolver quietly drops anything published in the last three days, and the development compose file ships admin/password.
Who is it for?
ZenML fits a company with more than one place to run pipelines and a real need for run history, because the client-server split is what lets the same code run locally and against a deployed server. Check four things before you commit.
Can I use it commercially?
Yes. Apache-2.0 is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
Is it still maintained?
Yes. The repository last received commits 2 days ago.
What is it written in?
Mainly Python, according to GitHub's language statistics.

Answers come from the project's GitHub data, last synced on November 15, 2026, and from our analysis. They are not legal advice.

From daily snapshots

Star history

  • zenml-io/zenml5,636 ★ · +56
Star history
Repository2026-09-102026-11-15
zenml-io/zenml5.6k5.6k

56 Stars in the last 66 days, measured from the daily GitHub snapshot. Compare with other projects

Editorial analysis

Package versions, Python support, and licensing

`pyproject.toml` declares version `0.98.0`, which is also the newest published release, from 2026-10-14, and the last push to `main` is dated 2026-10-30. Elsewhere in the manifest, `requires-python` is `>=3.10,<3.15`, the classifiers list Python 3.10 through 3.14, and `Typing :: Typed` is declared. The classifier also reads `Development Status :: 5 - Production/Stable` on a zero major version, and the license field says Apache-2.0, matching the `LICENSE` file at the root. The build is not a plain setuptools one either: the build backend is uv_build, and the package builds two module names, `zenml` and `zenml_cli`, which is how the CLI ships inside the same distribution as the library.

A three day publication filter, and a resolution failure that does not mention it

The build section configures the resolver to refuse anything published in the last three days:

toml
[tool.uv]
# Supply chain security: ignore Python packages published in the last 3 days.
exclude-newer = "3 days"
# Don't exclude our own package which is installed in some CI jobs.
exclude-newer-package = { zenml = false }

The manifest describes the window as a supply-chain precaution. It can give maintainers time to identify a compromised release, but it is not a guarantee that malicious packages will be detected within three days. The escape hatch is documented too, `uv add <pkg> --exclude-newer "0 days"`. What makes this worth reading is the next note in the file: when a version is filtered out, uv does not mention it, resolution just fails with `no compatible version found`, and the file says to keep that in mind when debugging. So the failure mode you will actually see names no cause. ZenML exempts itself from the window because its own package is installed in some CI jobs. The build backend is pinned narrowly as well, `uv_build >= 0.8.17, <0.9.0`.

The same page says zenml[server] and then zenml[local]

The quickstart installs one thing:

bash
# Install ZenML with server capabilities
pip install "zenml[server]"  # pip install zenml will install a slimmer client

# Initialize your ZenML repository
zenml init

# Start local server or connect to a remote one
zenml login

Then the architecture section, lower on the same page, gives a different extra for the same outcome. Local development is `pip install "zenml[local]"`, described as running both client and server locally, while production is `pip install zenml` plus `zenml login <server-url>` against a separately deployed server. The package manifest shows that these extras are different: `server` includes `local` and adds server dependencies. Check the installation guide for the workflow you intend to use and the manifest for the version you install.

The development compose file ships admin, password, and a MySQL root password

The repository's compose file opens with a credentials anchor and a database service that both use the same placeholder value:

yaml
x-admin-creds: &admin-creds
  ZENML_DEFAULT_USER_NAME: admin
  ZENML_DEFAULT_USER_PASSWORD: password

services:

  # Database
  db:
    image: mysql:8.4
    container_name: db
    ports:
      - 3306:3306
    environment:
      MYSQL_ROOT_PASSWORD: password

So the default administrator is `admin` with the password `password`, and the database root password is `password` as well, with port 3306 published to the host. The server service sets `ZENML_SERVER_SERVER_URL: http://localhost`, plain HTTP, plus `ZENML_SERVER_AUTO_ACTIVATE: true`, `ZENML_SERVER_DEPLOYMENT_TYPE: docker`, `ZENML_LOGGING_VERBOSITY: DEBUG`, `ZENML_ANALYTICS_OPT_IN: false`, and `ZENML_SECRETS_STORE_TYPE: sql`, which is where secrets land when the store type is sql. Fine for a laptop, dangerous the moment this file reaches a shared host.

The default compose path builds the server from source, not from the published image

The published server image is in the file, commented out, and the active path is a local build:

yaml
    # image: zenmldocker/zenml-server:latest
    build:
      context: ./
      dockerfile: docker/zenml-server-dev.Dockerfile
      args:
        INSTALL_DEBUG_TOOLS: true
      target: local-runtime

So a fresh `docker compose up` compiles the server from the working tree, with debug tools switched on, rather than pulling a released image. The build targets `local-runtime`, which the comment says runs the API against mounted source and refreshes on change, and the service confirms that by mounting `./src:/zenml/src`. It also pins `platform: linux/amd64`, maps host port 80 to container 8080, health checks `http://localhost:8080/health`, and waits for the database health check before starting. The tree carries `alembic.ini` and a `db` service on MySQL 8.4, which is the server-side cost the client-only install avoids.

Eleven customer logos, and every one of them is an empty anchor

The page says ZenML is used by thousands of companies, then shows the featured ones as a grid of links. The eleven linked companies are Airbus, AXA, JetBrains, Rivian, WiseTech Global, Brevo, Leroy Merlin, koble.ai, Playtika, NielsenIQ, and Enel. There is a note under the grid asking to email [email protected] if you want to be featured. The navigation strip lists Projects, Roadmap, Changelog, Report Bug, a ZenML Pro signup, Blog, and Docs.

The example gallery and repository layout

The recommended starting point is the quickstart, `examples/quickstart/`, described as covering pipelines, steps, artifacts, snapshots, and deployments. After that, a numbered list of seven more examples. Four of the seven live in this repository: the agent architecture comparison, deploying classical ML models, deploying agents, and E2E batch inference. The other three point at a separate `zenml-projects` repository, covering an LLM RAG pipeline, a deep research agentic workflow, and a fine tuning pipeline. Meanwhile the tree carries twenty two example directories, including agent framework integrations, agent outer loop, agentic human in the loop pipelines, a baseten robot policy, computer vision, end to end NLP, hierarchical document search, hydra config management, lakefs data versioning, LLM fine tuning, an MLOps starter, optuna hyperparameter tuning, a sandbox harbor, a sandbox pydantic ai, a software factory, and a weather agent. The rest of the repository is arranged the same way, with `AGENTS.md`, `CLAUDE.md`, `CLA.md`, a `.cursor/` and a `.claude/` directory, `.coderabbit.yaml`, `.typos.toml`, `lychee.toml` for link checking, `.trivyignore` and `trivy-secret.yaml` for scanning, a `.gitbook.yaml` for the docs site, `changelog.json` beside `RELEASE_NOTES.md`, `docker/`, `helm/`, `infra/`, and four Google Cloud Build files covering pull requests, nightly releases, release preparation, and the release itself.

The MCP server lives in another repository and holds your server URL and API key

The chat section points at `zenml-io/mcp-zenml`, a separate repository, and describes what it does: query pipelines, analyze runs, and trigger deployments using natural language through Claude Desktop, Cursor, or any MCP compatible client. Setup is four steps. Download a `.dxt` file, drag it into Claude Desktop settings, add your ZenML server URL and API key, then start chatting. So the third party in that chain holds a credential for your ZenML server and can trigger production deployments with it, which is a wider grant than reading run metadata would need. The example prompts in the page are correspondingly broad, asking which pipeline runs failed this week, for accuracy metrics across churn models, and to trigger the latest fraud detection pipeline with production data.

Editorial conclusion

ZenML fits a company with more than one place to run pipelines and a real need for run history, because the client-server split is what lets the same code run locally and against a deployed server. Check four things before you commit. Pick the extra deliberately, since the page names both `zenml[server]` and `zenml[local]` for a local run. Expect resolution failures that name a package as merely having no compatible version, because that is how the three day supply chain filter reports itself. Never reuse the compose file's admin and password outside a laptop. And remember the server needs a relational database with migrations, which the client does not.

Frequently asked questions

How do I install ZenML?

The quickstart uses `pip install "zenml[server]"` for server capabilities, noting that `pip install zenml` installs a slimmer client, then `zenml init` and `zenml login`. The architecture section instead gives `pip install "zenml[local]"` for running client and server locally, and `pip install zenml` plus `zenml login <server-url>` for a separately deployed server.

What is ZenML used for?

It is aimed at ML or AI engineers in a company setting, on traditional ML use cases, LLM workflows, or agents. You write workflows, called pipelines, that run on any infrastructure backend, called stacks, embed ordinary Python logic inside them, and get containerization, run tracking with metrics, logs and metadata, and integration with tools such as MLflow, Langgraph, Langfuse, Sagemaker and GCP Vertex.

Does ZenML need a database?

The server does. The development compose file runs MySQL 8.4 with a published port 3306 and a health check, the repository carries `alembic.ini` for migrations, and the server environment sets `ZENML_SECRETS_STORE_TYPE: sql`. The slimmer client installed by `pip install zenml` has no such requirement.

Which Python versions does ZenML support?

`requires-python` is `>=3.10,<3.15`, and the classifiers list 3.10, 3.11, 3.12, 3.13 and 3.14, with a `Typing :: Typed` marker. The build backend is uv_build, pinned to `>= 0.8.17, <0.9.0`.

Is ZenML open source?

The manifest declares `license = "Apache-2.0"` and the repository root carries a `LICENSE` file, and the package classifiers include the Apache Software License entry. The page also links to a ZenML Pro signup, so an open source core and a commercial offering sit side by side.

Official sources

  1. License: Apache-2.0
  2. Project website
  3. README
  4. Releases
  5. zenml-io/zenml on GitHub
Add this badge to your README

If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.

Add this badge to your README

markdown
[![Hysen Labs](https://hysenlabs.com/badge/zenml-io-zenml.svg)](https://hysenlabs.com/projects/zenml-io-zenml)

Star history chart for your README

markdown
[![Star history](https://hysenlabs.com/chart/zenml-io-zenml.svg)](https://hysenlabs.com/projects/zenml-io-zenml)

Was this useful?