Know what the product
can prove today.

A next-generation Agor agent workspace running separately from app.agor.me. It explores a better interface and release model without changing the current production product.

No. The current app remains the production system. This MVP becomes a replacement candidate only after its functional, data, security, and migration gates are proven.

The workspace supports customer-facing agents for focused jobs such as reception, support, onboarding, account recovery, and scheduling. The best first agent owns one measurable job.

A narrow job creates a testable boundary. It tells the agent what completion means, tells the supervisor what to watch, and tells the operator when to hand off.

The MVP runs 100 structured policy and readiness cases against the saved configuration. It reports first-attempt and repeated-attempt reliability. A production release will add model-based persona simulations and independent judging.

Not in this MVP. The latest simulation must score at least 90 with no hard failure. That constraint is intentional: a status badge should mean something.

It independently checks policy boundaries, unsupported claims, looping, frustration, and missed handoffs while the customer-facing agent remains focused on completing the job.

Structured agent configuration, simulations, and releases use platform-backed database storage. Uploaded knowledge files use object storage. The public demo uses sample data and does not write to your workspace.

They can be selected and versioned as target channels, but a channel is not represented as production-connected until its provider integration is verified. The existing app.agor.me remains the live channel system during this build.

This MVP stores the materials you explicitly upload to the signed-in workspace. It does not claim to train a foundation model on that data. Production retention, deletion, and export controls remain release requirements.

The unit of value is a finished job, not a message, seat, or token. The interface therefore keeps the outcome definition visible beside the agent and its release evidence.

A human resolution should become a proposed knowledge update. After review, the changed behavior is simulated again and shipped as a new immutable version.

Trust requires receipts. Operators need to know which context, rule, tool, simulation, and release produced a behavior. Invisible prompt edits are not a release process.