Architecture in 5 minutes
AGILAB is a reproducible AI/ML experimentation workbench for engineering teams, bridging local interactive development, distributed execution, and result analysis.
Use this page for the mental model before opening the detailed architecture reference.
One control path
User
|
v
Streamlit UI, CLI wrappers, or notebook entry points
|
v
AgiEnv: settings, project selection, app paths, logs, local workspace
|
v
AGI facade: install, get_distrib, run, service actions
|
v
agi-node / agi-cluster: package workers, dispatch work, start local or SSH workers
|
v
Dask back-plane and optional MLflow tracking
|
v
Artifacts, run manifests, app pages, and ANALYSIS views
What each layer owns
Layer |
Responsibility |
Proof to look for |
|---|---|---|
UI / CLI / notebook |
Capture user intent and keep PROJECT, ORCHESTRATE, PIPELINE, and ANALYSIS on one visible path. |
|
|
Resolve app paths, settings, logs, environment variables, and per-user workspace files. |
|
|
Provide the public install, distribution, execution, and service actions used by both UI and examples. |
Example scripts in |
worker packaging |
Build and stage worker runtimes so manager code and worker code stay isolated. |
Worker copies under |
Dask / MLflow |
Dask is the worker dispatch back-plane; MLflow is an optional tracking system AGILAB can integrate with, not replace. |
Dask logs, MLflow runs, artifacts, and ANALYSIS pages. |
Boundary
AGILAB does not replace MLflow, Kubeflow, cloud deployment platforms, security gateways, or production rollback systems. It keeps the engineering experiment path reproducible and visible, then exports evidence and artifacts that those deployment-focused systems can consume.
Next pages
Quick-Start for the first proof.
AGILab Architecture for package and runtime details.
AGILab in the MLOps Toolchain for the MLOps boundary.
Compatibility Matrix for validated and documented public routes.