Skip to main content
Both SDKs are distributed org-internal. Neither is on PyPI or public npm, so the install steps differ from a normal public package.
Internal release. Confirm with your Atlan contact that you have repository or GitHub Packages access before you start. atlan-ai has no license assigned and must not be redistributed outside Atlan.

​
Python

atlan-ai requires Python 3.10 or later. Each release is a committed, checksummed artifact in the atlanai-sdk-python repository under releases/<version>/. Install from a checkout or from any copy of the artifact:
In a uv or requirements.txt setup, pin the artifact by path:
Verify the artifact before installing it:

​
LangChain and LangGraph

atlan_ai.langchain needs langchain-core in your application. It is imported lazily, so the core SDK has no framework dependency:

​
TypeScript and JavaScript

@atlanai/sdk requires Node 20 or later, and also runs on Cloudflare Workers with nodejs_compat. It publishes to the GitHub Packages npm registry. Point the @atlanai scope at GitHub Packages once per project, in .npmrc:
Authenticate with a token that has read:packages — a personal access token locally, or NODE_AUTH_TOKEN in CI. Keep the token out of .npmrc in version control; use your environment or your CI secret store.
pnpm add and yarn add work the same way. The published package is compiled JavaScript (ESM and CJS) with type declarations, so plain JavaScript projects need no TypeScript toolchain.

​
Verify the install

Printing a version confirms the package resolves. It does not confirm that export works — that needs credentials and a reachable Gateway. Continue to the quickstart, which ends in a verification step against Agent Registry.

​
Next steps

Quickstart

Initialize a client and report your first trace.

Configuration

Credentials, environment variables, and export modes.