TaughtHuman decisions for better models

Taught keeps human decisions attached to what they change.

Taught is a human-review and custom-model platform for teams that need to explain where training evidence came from, how a candidate was evaluated, and what version served a prediction.

Request early access Browse open releases
Evidence registerThe product follows one durable idea.

Source, suggestion, human action, dataset membership, model promotion, and serving identity should remain connected rather than becoming separate operational stories.

  1. Content
    Preserve the source

    Review begins with attributable image, audio, or video evidence.

  2. People
    Record human judgment

    Independent actions remain distinct from model assistance and rewards.

  3. Models
    Gate improvement

    Frozen datasets and evaluation decide whether a candidate advances.

  4. Use
    Name what served

    Compatible deployments return endpoint, model, and usage receipts.

Why build the whole evidence chain?

Annotation tools, training systems, and inference providers each answer part of the model lifecycle. Taught is designed to preserve the relationships between those parts so a team can inspect decisions without reconstructing them from disconnected exports and logs.

Human authority
Assistance can accelerate work, but model proposals never become human votes.
Reproducibility
Task revisions, dataset manifests, splits, and run artifacts make model changes inspectable.
Capability honesty
Labeling, proposals, training, export, and inference are reported separately for each task and deployment.
Open or private
Organizations control authority while verified open-source publishers can release checkpoints with evidence.

What Taught is building toward

  1. 01
    Clear work for people

    Contributors answer bounded questions about real content under project rules.

  2. 02
    Governed evidence for teams

    Owners define tasks, quorum, rewards, datasets, training, and promotion authority.

  3. 03
    Accountable models for products

    Applications and agents use exact promoted versions when hosted serving is enabled.

  4. 04
    One product truth

    Every public page distinguishes implemented capability, deployment availability, and future platform scope.

How to read capability claims on this site

  • Images, audio, and video have native review contracts; image object detection is the mature end-to-end task.
  • Text, document, and PDF are platform targets, not native labeling claims.
  • Hosted inference, public endpoints, MCP, A2A, and redemption depend on explicit deployment capability.

Questions teams ask before they begin

Direct answers based on the current Taught product and deployment boundaries.

Is Taught the same project as ImageTagger?

Taught is the active product brand. ImageTagger is the historical repository and development codename.

Who is Taught for?

It is designed for ML teams governing reviewed training evidence and for contributors completing clear, project-scoped labeling missions.

Is every described capability live on taught.dev?

No. The public site describes product contracts and marks deployment-gated capabilities. Access and service availability must be verified for the target deployment.

Bring the decisions, data, and model version into one record.

Request access for a labeling project, a custom-model workflow, or an integration with a promoted model.

Request early access