AGILab in the MLOps Toolchain

AGILab focuses on the early experimentation phase of AI projects—roughly Technology Readiness Level 3 (TRL‑3), where teams validate concepts, explore algorithms, and collaborate on data preparation. This page explains how AGILab fits alongside the broader MLOps landscape.

Where AGILab helps

  • Rapid experimentation: templates, web pages, and CLI mirrors reduce the friction of testing new ideas without scaffolding bespoke dashboards.

  • Multi-algorithm workflows: built-in orchestration (AGI.run / AGI.get_distrib) lets engineers cycle through multiple models using the same datasets and environment setup.

  • Distributed execution without DevOps: Dask-based scheduling, SSH helpers, and worker packaging (agi_cluster, agi_env) allow TRL‑3 teams to scale out experiments without managing Kubernetes or cloud stacks.

  • Offline productivity: optional offline assistants (Mistral/GPT‑OSS paths, plus local documentation assets when configured) can keep experimentation usable in air-gapped environments once the dependencies and local model/runtime setup are installed.

What AGILab does not aim to cover

  • Production deployment (TRL‑6+): model serving, CI/CD, feature stores, online monitoring, or model drift detection belong to the deployment-focused side of MLOps (tools such as Kubeflow, MLflow Serving, Sagemaker, etc.).

  • Enterprise governance: compliance workflows, audit trails, or retraining policies are intentionally out of scope. AGILab’s strength is rapid iteration before promoting assets to hardened pipelines.

Positioning vs. other tools

Positioning vs. other tools

Phase

AGILab focus

Examples of complementary tools

Ideation / TRL‑2

Not covered (use notebooks, small prototypes)

Whiteboards, notebooks, lightweight sandboxes

Experimentation / TRL‑3

Primary target – templated projects, cluster automation

AGILab + data catalogues + experiment trackers

Validation / TRL‑4

Hand off to deployment-stack as soon as requirements stabilise

MLflow, Weights & Biases, Seldon, Kubeflow

Deployment / TRL‑6+

Out of scope

CI/CD, serving frameworks, APM, feature stores

Suggested workflow

  1. Use AGILab to prototype algorithms, reuse app templates, and validate data processing. Capture run history via ~/log/execute/<app>/.

  2. Once an approach stabilises, prepare the project artefacts for your target environment and integrate it with your organisation’s deployment toolchain (MLflow, Kubeflow, internal devops stack). In a source checkout, this commonly uses tools/run_configs and src/agilab/apps/<app>; in packaged installs, use the generated app package artifacts available from your installer.

  3. Track long-running metrics and governance artifacts using your preferred MLOps platform; AGILab does not replace those systems.

See also