Evaluate API · Free accounts · 100 evaluations / monthRelease status
DeltaX Evaluate
Menu

WORKFLOWS YOU CAN INSTALL

Agent skills

Installable recipes that put a decision checkpoint in the middle of an agent workflow. Evaluate selects one step from the eligible candidates or refuses, with an auditable trace. Your application keeps execution.

How a skill works

  1. Build candidates
  2. Set support
  3. Call Evaluate
  4. Reconcile
  5. Act and record

A selection is never permission or authority.

GROK BOT TEMPLATE

A Grok Bot template that routes decisions through Evaluate

The recipes below are installable skills. If you want a ready-made Grok Bot that routes next-step decisions through Evaluate, use the public Evaluate Decision Gate template.

Evaluate selects a review step from your application's proposed options, or refuses, and returns a trace with SHA-256 receipt references. Your application keeps execution. The public template, Evaluate Decision Gate, routes your agent's next-step decisions through DeltaX Evaluate (free tier) so a hash-consistent receipt can sit with the sourced data pinned alongside. A selection is never permission or authority. Evaluate does not verify caller-reported support or carry out the selected step. Free accounts include 100 evaluations per month, no card required, resetting on the first of each month at 00:00 UTC.

Measured data first

Content decision loop

Measure, score, evaluate, act, log: set support scores from your own measured analytics (or labeled judgment), let Evaluate select one internal next step, and keep the receipt.

  • analyze
  • compare
  • draft
  • validate
  • external (shadow only)
View skill

Labeled judgment first

Pre-publish check

Before anything goes out, have Evaluate select one review step (validate, compare, draft, analyze) and keep publishing as the owner's own decision.

  • analyze
  • compare
  • draft
  • validate
  • external (shadow only)
View skill

Labeled judgment first

Support triage step

Pick one internal triage step for a new ticket, with replies to customers left to a person.

  • analyze
  • compare
  • draft
  • validate
  • external (shadow only)
View skill

Labeled judgment first

Research next-step pick

Choose one bounded research step, such as opening sources or checking claims, and keep publishing or sharing findings as a separate decision.

  • analyze
  • compare
  • draft
  • validate
  • external (shadow only)
View skill

Plans

Free accounts include 100 evaluations per month, no card required, resetting on the first of each month at 00:00 UTC. Paid subscriptions are planned and paid signup is not open yet. Current details are on /pricing.

Create a Free account

Scenarios are illustrative, and their support values are operator judgment. Sample receipts are runs by DeltaX on those scenarios, not customer evidence.