ToolUp.Facts.Presentation.Server 0.38.1

ToolUp.Facts.Presentation.Server

The render_chart, render_table and evaluate_pipeline AI tools in the answering form. The model writes a presentation request (the ToolUp.Facts.Presentation vocabulary: a kind, a fact table, its metric(s), the subjects and a period) and never a row. The server reads the series the request names through the fact browse's series read, under the disclosure gate at the tool-result door. It draws the block with the presentation builders and answers with a text summary the model reads and the block the assistant message carries.

What it gives you

• PresentationCompose.withRenderTools composes both tools onto the AI composition. It is idempotent, and it does nothing under NoFactStore.

• RenderTools.answerWith is the executor over explicit dependencies: the fact browse's dependencies, the caller's resolved scope and principal. RenderTools.answer is the HttpContext adapter the composition registers.

• Refused by name. A malformed request, an unknown table, metric or subject level, and a kind the other tool draws are each refused with a sentence naming what to repair.

• The model reads counts, never values. The summary names the kind, the metric, the table, the canonical period, the title, how many points were read and how many were withheld under which policy. A withheld point is a count there and is never drawn. A caller the policy denies every point gets a refusal, not an empty chart.

• PresentationCompose.withPipelineTool composes evaluate_pipeline in the same answering form and mounts IPipelineBlockApi. The model reads the executor's summary, and the message carries a pipeline block that persists the request and never a row. A reader's panel sends that request to the endpoint, which evaluates it again for the reader under the disclosure gate at the browse door.

• PresentationCompose.withDerivedAssertions lets evaluate_pipeline assert one numeric column of its result as facts of a registered derived metric (the assert argument), each naming the facts it was computed from, so a figure the assistant quotes verifies and a narrative may cite it. Without it the argument is absent and refused.

open ToolUp.AI.AICompose
open ToolUp.Facts.Presentation

AIServerApp.createFrom factory providerProfile app
|> PresentationCompose.withRenderTools
|> PresentationCompose.withPipelineTool
|> PresentationCompose.withDerivedAssertions // optional
|> composeAI

Dependencies

ToolUp.Facts.Presentation, ToolUp.Facts.Server and ToolUp.AI.Server. The published Apache-2.0 Fuaran.UI packages arrive through ToolUp.Facts.Presentation, so a deployment that does not reference this package carries none of them.

Apache-2.0.