Create

AI Factory

Describe a dataset, model, simulation, or workflow and Factory drafts it for you to review and publish. An optional accelerator — you can always upload manually instead.

When to use it

AI Factory is for the moments when you know the shape of the asset you want but don't want to wire the scaffolding yourself. It lowers the floor on tasks that would otherwise take hours of yak-shaving: a synthetic dataset spec, an RL environment template, a labeling rubric, a training pipeline config.

Factory is not a replacement for expert authorship. For state-of-the-art models, custom collection rigs, or anything where the value is in the data you alone can capture, upload directly.

The user flow

Prompt input. Describe what you want in plain language. Examples: “Generate labeled traffic intersection data for autonomous vehicles.”
Analysis. Factory classifies the request into an asset type, infers domain and modalities, and estimates scale.
Clarification. If the prompt leaves material decisions on the table, Factory asks one to four targeted questions. Questions with discrete answer spaces show option chips.
Build plan. Factory drafts a concrete plan — goal, outputs, ordered steps, risks, and success criteria — along with a difficulty, time, and cost estimate. You approve, redraft, or cancel.
Generation. Factory produces every file in the plan: schema, manifest, scripts, configs. The progress page polls for live status.
Review. A tabbed reviewer shows every file, the generated README and docs, the metadata suggestions, and a safety report. You can edit title, license, and price, regenerate from this stage, or publish.
Publish. Approved jobs become real marketplace assets owned by you, immediately licensable.

Supported asset types

Type	What Factory generates
Synthetic vision dataset	Schema, manifest, sampling plan, generation script
Robotics dataset	Episode schema, modality manifest, collection plan, ROS bag template
RL environment	Runnable Gym/Gymnasium scaffold, observation/action specs, reward shaping
AI workflow	Composable pipeline definition with prompts, transforms, and gates
Training pipeline	Config, hyperparameters, eval harness, README
Data labeling project	Annotation schema, instructions, quality rubric, calibration set

Safety

Every generated asset goes through the same automated pipeline as manual uploads (format scan, malware scan, license compatibility) plus a Factory-specific safety review. The review classifies the artifact as pass, review, or block:

Pass — publishable as a public listing immediately.
Review — publishable but flagged for human moderation; the listing enters pending_review status.
Block — cannot be published. Regenerate with revised inputs or appeal.

A human is always in the loop

Factory never publishes silently. Every job stops at the review stage for explicit approval, and the safety verdict is shown alongside the publish controls.

API

Everything in the UI maps to an API endpoint.

POST /v1/factory/jobsbash

curl https://www.nodedata.ai/api/v1/factory/jobs \
  -H "Authorization: Bearer $NODE_DATA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "Create synthetic defect images for manufacturing quality control."
  }'

GET /v1/factory/jobs/:idbash

# Poll the job to follow the pipeline
curl https://www.nodedata.ai/api/v1/factory/jobs/$JOB_ID \
  -H "Authorization: Bearer $NODE_DATA_API_KEY"

# Status transitions:
# queued → analyzing → awaiting_clarification → planning →
# awaiting_approval → generating → review → published

POST /v1/factory/jobs/:id/clarify — submit answers
POST /v1/factory/jobs/:id/approve — approve the plan and start generation
POST /v1/factory/jobs/:id/regenerate?stage=plan|build
POST /v1/factory/jobs/:id/publish — convert into a marketplace listing

Pricing

Factory generation runs on platform credits at cost. Approve a plan only after reviewing the estimate in the right sidebar. Generation cost is separate from the marketplace platform fee — the 25/75 split on sales is unaffected.

Limits

5 concurrent jobs per account on the default plan
50 jobs/day on Open, unlimited on Creator and Enterprise
Generated artifacts capped at 20 MB total for the text-shaped asset types; larger outputs queue through dedicated workers