Skip to Content (Press Enter)
Community › Forge › Assets list › AgenticChunkingHelpers
Login to follow
AgenticChunkingHelpers

AgenticChunkingHelpers (ODC)

Stable version 0.1.3 (Compatible with ODC)
Uploaded on 22 Jun by Michael Guzman
AgenticChunkingHelpers

AgenticChunkingHelpers (ODC)

Details
Detailed Description

A stateless C# External Logic library that handles the text processing work on either side of your AI Gateway calls for Level 5 Agentic Chunking workflows inside ODC.


Three actions are exposed. PreChunkForExtraction splits source text into token-safe batches before the extraction call, respecting paragraph and sentence boundaries. ParsePropositions parses the raw AI Gateway response into a clean list of proposition strings, handling code fences and escaped quotes. NormaliseAgenticOutput maps the grouping response to typed AgenticChunk structures with ChunkId, ThematicCategory, MergedContent, PropositionCount, CharacterCount, TokenEstimate, and SHA-256 hash.


The library does not call the AI Gateway. It does not store state. It handles the C# work ODC cannot do natively: JSON parsing, SHA-256 hashing, and token-safe text batching.


Designed to pair with an ODC App that owns the AI Gateway calls and the AgenticChunkingODC orchestration logic. ODC only. Not compatible with OutSystems 11.

Limitations
  • ParsePropositions and NormaliseAgenticOutput return empty results on parse failure rather than throwing. Callers must check IsSuccess and handle the empty case explicitly.
  • PreChunkForExtraction batches by token estimate only (1 token = 4 characters). It does not call an embedding model or tokeniser. Actual token counts may vary by model.
  • The library expects the AI Gateway response to be a plain JSON array (extraction) or a plain JSON array of objects (grouping). Responses wrapped in additional structure outside those formats will fail to parse.
  • ODC only. Not compatible with OutSystems 11.