Skip to main content
Kviria turns a user’s spoken or typed reference into exactly one place or address and returns its verified coordinate with an evidence trail. Data only, never phrased sentences: your agent phrases the reply, Kviria guarantees the facts. Your wallet is your account. One endpoint: POST /v1/agent.

​
Why agents need it

LLM agents guess locations. They hallucinate coordinates, invent distances, and describe crosswalks they have never seen. Kviria replaces the guess with a deterministic narrowing loop:
  • Never a guessed coordinate. Every coordinate-out goes through a verification gate the user confirms (or the caller explicitly opts out of).
  • An evidence trail on every answer. Each response carries the tests that produced the decision - names, distances, and reason codes exactly as computed.
  • Honest failure. When nothing matches, Kviria says notFound with the absent category - it never invents a plausible-sounding place.

​
The loop in one paragraph

You send the user’s words (raw transcript) or a structured intent, plus their current anchor {lat, lng}. Kviria answers with one of five decisions: a resolved target, a confirmation gate, a candidate set, a not-found, or a clarifying question. When the answer is a candidate set, you ask the user which one; their pick becomes the next call. Each turn narrows until exactly one place remains - then you hold a verified coordinate with an evidence trail.

​
Status

Kviria is in beta. Current coverage: San Francisco (approximately lat 37.70-37.84, lng -122.54 to -122.36). Outside the covered area, queries return honest notFound answers rather than guesses. Get your first verified coordinate in under a minute (with your wallet’s sign-in — see the Quickstart):
Continue to the Playground to try it live, the Quickstart for the full walkthrough, or connect the MCP server. Your agent can also load the skill directly.