Atlas v1 documentation

Understand the system before you operate it.

A guided map of the v1 implementation for evaluators, developers, administrators, security teams, and operators.

Run the deterministic demo →Inspect the source →Review implemented controls →

12 topics

Atlas v1

Overview

Atlas concepts, product boundaries, and the v1 evidence model.

Product tour
Personas
Scope and limitations

Atlas v1

Getting started

Run the deterministic demo or bring up the complete local stack.

Quick start
Demo mode
First governed analysis

Atlas v1

Architecture

Modular monolith, control plane, workers, data stores, and public edge.

Module boundaries
Data model
Release architecture

Atlas v1

Security

Identity, tenancy, authorization, query governance, agents, and supply chain.

Security model
Threat controls
Responsible disclosure

Atlas v1

Connectors

SDK contracts, read-only access, credential references, retries, and provider status.

PostgreSQL
Snowflake
BigQuery

Atlas v1

Analytics

Certified metrics, analytical methods, visualization, reports, and automation.

Methods
Semantic layer
Reports

Atlas v1

Agents

Planning, execution, tools, memory, evidence, budgets, safety, and evaluation.

Agent system
Evaluation
Provider boundaries

Atlas v1

Evidence

Citation lineage, verification, artifact integrity, and failure behavior.

Evidence ledger
Verification
Artifact formats

Atlas v1

SDKs

Typed Python and TypeScript clients with safe retries and pagination.

Python SDK
TypeScript SDK
Authentication

Atlas v1

API

Versioned contracts, errors, idempotency, limits, and health endpoints.

API conventions
Error model
Health and readiness

Atlas v1

Deployment

Containers, Compose, Kubernetes, Terraform, observability, release, and DR.

Production topology
Migrations
Rollback and restore

Atlas v1

Administration

Organizations, workspaces, identity, governance, billing, and operations.

Onboarding
Enterprise identity
Billing lifecycle

Quick start

Run the complete local platform.

Atlas fails closed unless the required database, Redis, identity, artifact, and observability configuration is present. Start from the checked-in example, never from invented production values.

cp .env.example .env
# Replace every required development-only value.
docker compose --env-file .env config
docker compose --env-file .env up --build

curl --fail http://localhost:8000/health/ready
open http://localhost:3000/demo

Typed SDKs

Call only an authorized Atlas API origin.

Python and TypeScript clients apply bounded timeouts, safe retries, cancellation, redacted errors, and pagination-loop detection. The public production candidate does not expose a customer API.

import os
from atlas_ai_sdk import AtlasClient

with AtlasClient(
    "https://your-authorized-atlas-origin.example",
    token=os.environ["ATLAS_TOKEN"],
) as atlas:
    workspaces = atlas.list_workspaces()

import { AtlasClient } from "@atlas-ai/sdk";

const atlas = new AtlasClient({
  baseUrl: "https://your-authorized-atlas-origin.example",
  token: process.env.ATLAS_TOKEN!,
});
const workspaces = await atlas.listWorkspaces(signal);

External validation status

Some production setup cannot be completed from source code.

Snowflake and BigQuery require authorized live accounts.
OAuth providers require registered production applications.
Email requires an authorized provider and SPF/DKIM review.
Customer use requires production infrastructure and enterprise operations.