Entity resolution for applications and AI agents

From a name
to the right entity.

A name is only the beginning. SuperER connects references to your catalog, keeps track of corrections, and asks when the next action needs a clearer answer.

Python library · v0.10

Explore entity resolution

One name. Different decisions.

Change the task. See what needs to be resolved.

The request

“Notify Alex.”

Directory candidates2 records

Loading the recorded example…

The decisionClarification needed

Which Alex should be notified?

Both names match. Sending a notification requires one specific person.

notify_person

Requires a person ID and email.

No tool parameters proposed.

Evidence: “Alex” matches two directory aliases.

Why resolution gets complicated

The words change.
The reference has to hold.

Finding a name in a sentence is extraction. Connecting it to the right record is entity resolution. A conversation adds another challenge: knowing which parts of the previous request still apply.

Understand the concepts
A request

“Send it to Alex.”

Which person does the name refer to?
A correction

“Actually, Alex Rivera.”

Replace the identity; preserve unrelated constraints.
A change of scope

“Route it to their team instead.”

The tool now needs a team, rather than a person.

Inside the library

Ground the meaning.
Carry the evidence.

01

Interpret

Find mentions, roles, constraints, and requested changes.

02

Resolve

Retrieve catalog candidates and select supported identities.

03

Reconcile

Apply updates with provenance and explicit uncertainty.

04

Bind

Validate the parameters required by the next tool.

Deterministic rules, retrieval, and optional model adapters share typed evidence. Your application authorizes and executes actions.

Clarification that carries state

Return a structured question. Accept the user’s choice without losing the rest of the request.

Memory with boundaries

Track scope, validity, and provenance. Invalidate dependent decisions when their evidence changes.

Recovery with limits

Retain independently valid fields. Block contracts that depend on unresolved information.

Measured, with context

Inspect the results.

Comparisons include simpler baselines, failed repairs, and the limits of each study. Current results do not establish a general accuracy advantage.

Explore evaluations

Build with the concepts

Your catalog. Your tools.
One explicit interpretation.

Explore the Python API