This is the documentation of Gwenlake Atlas, the platform that gets AI projects to production: it connects to your systems, prepares your data, runs your models and agents, and writes the results back into your tools. Under your control. This documentation covers the two ways to program it:
  • The Python library (gwenlake). A single Gwenlake client exposes your catalog (projects, datasets, files, connections, SQL) and the inference models, and a transforms layer lets you write dataset-to-dataset pipelines and trainings as decorated functions.
  • The REST API at https://api.gwenlake.com/v1, for every other language and for direct HTTP calls.

Quickstart

Install the library, authenticate and run your first query in five minutes.

Core concepts

Organizations, projects, datasets, connections, models and lineage.

Transforms

Reproducible pipelines between datasets, with lineage recorded for you.

API reference

Every public endpoint, with an interactive playground.

​
What you can do

List projects and datasets, upload and download files, run SQL on your datasets with DuckDB or on a connection’s native engine. See Projects, Datasets, Files and Queries.
Declare inputs and outputs with decorators and let the platform run, schedule and trace your code. See Transforms, Incremental transforms, Data branches and Running.
OpenAI-compatible chat completions, embeddings and reranking, served by Gwenlake. See Chat and Embeddings.
Define an agent as a file in your repository, bind it to your datasets, ontologies and mediasets, and talk to it over sessions. See Agents.
Long trainings that checkpoint, survive preemption and resume where they stopped. See Models.

​
Get help

The library is open source on GitHub. For questions about the platform or your account, write to info@gwenlake.com.