A SOURCE-BASED TOUR
Click a stage, reviewer, engine component or model to explore it.
Save the visual map ↓
ADAM: an AI experiment command centerSource-based map of ADAM workflows, lifecycle reviewers, shared execution services, and model architectures.PROJECT FIELD GUIDE / 01 OCT 2026ADAMA local AI experiment command center.From an idea to a reviewed dataset, a trained model, and evidence for the next run.LOCALDesktop workspaceREVIEWEDTraining approval + reviewEXTENSIBLEDeclared model pluginsREPEATABLESaved jobs + experiments01 / THE EXPERIMENT LOOP01Idea → planChat or guided modelcreation02Collect + reviewImages, clips, captions,decisions03Review + approvePreflight and workloadchecks04TrainOne shared sequential jobqueue05GenerateSeeds, previews,image/video history06Learn + repeatCompare runs and saverecipesEVIDENCE FEEDS THE NEXT RUNEVEDataset reviewerLocal visual similarity; uncertainimages stay reviewable.ORIONBefore trainingChecks workload and settings; warningscan require approval.ATLASDuring trainingWatches loss, heat and disk; criticalconditions pause work.NOVAAfter trainingChecks sample health; artistic qualitystill needs your eye.02 / ONE ENGINE, MULTIPLE WORKSPACESDESKTOP + REMOTEShared execution pathPlannerTool registryJob managerTool executorDURABLE MEMORYJobs + logs / Datasets + models / Experiment database / Generation settings + reviews03 / DIFFERENT WAYS TO MAKE AN IMAGEDDPM + FlowNoise → imageDenoising or an integrated flow.PixelRowRows → imageBuilds the canvas top to bottom.INRFlowCoordinates → RGBPixel-space flow, no VAE.Neural Cellular AutomataSeed → growthLearns local cellular update rules.ALSO IN THE LABSDXL LoRA · Wan video LoRA · Oasis action world model + playerSource inspection: adam/ + models/; animations in the companion guide explain mechanisms, not model quality.
UNDER THE HOOD

FOUR VISUAL MECHANISMS

Same canvas. Different ways to build it.

Illustrative animation only. No model inference, no quality comparison, no training.

EXPLORE ADAM

A lab with memory.

  • One experiment loop. Collect and review a dataset, approve a plan, train, generate samples, compare outcomes and preserve useful recipes.
  • Four explainable responsibilities. EVE proposes visual dataset decisions. ORION reviews planned workloads. ATLAS supervises active training. NOVA checks available sample health. Human review remains part of the process.
  • Different model architectures. Connected DDPM, Flow and SDXL LoRA tools sit alongside experimental PixelRow, INRFlow and custom Neural Cellular Automata plugins.
  • Image, video and playable workflows. Wan 2.1 video LoRA has a dedicated clip, caption, training and generation workspace. Oasis supports action-conditioned world models and a player.
  • Local by default. Optional Ollama, web context and token-protected Remote access have explicit settings. Remote access is disabled in the clean release configuration.

About this guide

ADAM began as a Jarvis-inspired assistant for AI workflows and grew through ongoing experiments. This guide explains the source release dated 01 October 2026. The animations illustrate concepts; they are not inference outputs or quality benchmarks.

Project, source code and setup instructions →