1. Atlas Documentation¶
A self-hosted, pre-integrated gen-AI, ML, and data platform — one Docker Compose stack
Chat, RAG, agents, distributed compute, and a full data platform — source-configurable services wired together out of the box and selectable among the deployment modes each supports.
Atlas is a self-hosted engineering platform that bundles 30+ services — LLM inference and a gateway, vector and graph databases, workflow and DAG automation, distributed compute, object storage, notebooks, and observability — behind a Kong gateway and an adaptive FastAPI backend.
1. Capabilities¶
Atlas organizes 57 service families into 7 tracks. Each track pre-selects a working subset of the platform for one class of workload; the setup wizard prompts for track-scoped services and force-disables the rest.
Generative AI · RAG
Retrieval-augmented generation — vectors, graph, reranker, doc ingest, web search, workflow automation.
Generative AI · Engineering
Agentic apps + workflows with voice, vision, and search.
ML Engineering
Distributed training/inference + notebooks + experiment storage.
Trading / Financial Research
Read-only financial research and paper portfolios in notebooks; no live trading.
2. Quick Start¶
./start.sh
./start.sh --track gen-ai-rag
./start.sh --llm-provider-source ollama-container-gpu
Interactive wizard by default; CLI flags skip prompts for the values you set. Full flow, flags, and troubleshooting: Quick Start.
3. Platform Topology¶
Kong routes every *.localhost host; LiteLLM is the single path for local and cloud model traffic.
Per-flow diagrams (data/RAG, LLM provider routing, observability, security boundary, bootstrapper lifecycle): Architecture.
4. Documentation Map¶
Reference
Env vars, ports, manifest fields, and the 53 SOURCE-configurable service surfaces.
5. Setup Surface¶
