43. Skip to content

43. GAP Agentic

GAP Agentic is the reasoning layer of the Global Access Platform (GAP). Where GAP v1 curates weather and climate data and serves it through APIs, GAP Agentic turns that data into per-request agronomic advice: it reasons over the current context for a location and returns a recommendation together with the evidence and confidence behind it.

For the underlying data platform, datasets, and public APIs, see the existing developer architecture and API guide.

43.1 What "agentic" means here

Agentic does not mean "a language model writes the advisory". It means the system reasons over context per request rather than firing a fixed rule on a schedule. For a given request it consults the relevant forecast, calibration, and validation signals, evaluates them against a decision engine, and reports three things together:

  • what it recommends,
  • the evidence chain that produced the recommendation, and
  • how confident it is, including where data was missing.

That last point is a design requirement: the platform is expected to be explicit about its own uncertainty rather than presenting a confident guess.

43.2 How it relates to GAP v1

Concern GAP v1 GAP Agentic
Primary job Curate and serve weather/climate data Reason over that data to produce advice
Interaction Data queries and downloads Per-request or scheduled advisory preparation
Output Datasets and measurements An auditable advisory package

GAP Agentic is additive: it builds on the same data foundation and does not replace the data APIs.

43.3 Two commitments

Two principles run through the whole platform and are worth stating up front:

  1. Every advisory is auditable. No recommendation exists without a traceable chain back to the signals that produced it. See Decision trace and confidence.
  2. The system degrades honestly. When data is missing or a source is unavailable, the platform records that in a structured way rather than silently substituting a guess. See Observability and degradation.

43.4 Where to go next