Software Agent Types

Created: 2026-05-12 · 9 concepts

https://ontology.global.church/core#SoftwareAgentTypeScheme

Runtime-shape classification for gc:SoftwareAgent: LLM Agent, MCP Server, ETL Bot, Scheduled Job, Webhook Handler, Crawler, Conversational Agent, RAG Pipeline, Inference Worker. Flat scheme — agents that span shapes are typed by their primary surface and linked to additional capabilities via gc:agentInvokesService.

Code	Label	Definition
`SAT-CONV`	Conversational Agent	An agent whose primary interface is multi-turn dialogue with a human or another agent. Distinct from an LLM Agent — a Conversational Agent emphasizes the dialogue surface (chat UI, voice channel) while an LLM Agent emphasizes tool-using autonomy.
`SAT-CRAWL`	Crawler / Scraper	An agent that discovers and harvests content by following links across the public web or a federated set of endpoints. The KNIME pipeline's church-discovery and church-scrape stages are canonical examples; the planned federation indexer is another.
`SAT-CRON`	Scheduled Job	A time-triggered background process (cron, scheduled GitHub Action, Supabase pg_cron, launchd timer). Distinct from an ETL Bot by its trigger (clock) rather than its work shape (transform). Use SAT-ETL for the transform-shape, SAT-CRON when the time-trigger semantics dominate.
`SAT-ETL`	ETL Bot / Bridge	An extract-transform-load worker that pulls from one system and writes to another. The GC-Core bridges (gc-jp-bridge, gc-his-bridge, gc-org-bridge) are the canonical examples — each is an ETL bot operating on Joshua Project, HIS, and Supabase respectively.
`SAT-INF`	Inference Worker	A worker that performs LLM or other ML inference as its core function — classification, extraction, summarization, embedding. The KNIME enrichment worker (church record classification) is a canonical example. Distinct from an LLM Agent in that inference workers operate against batch queues rather than tool-using dialogue loops.
`SAT-LLM`	LLM Agent	A large-language-model-backed agent that decides which tools to call and what payloads to send. Distinct from a Conversational Agent (which centers on dialogue) and a RAG Pipeline (which centers on retrieval): an LLM Agent has tool-using autonomy. Model identifier (e.g., claude-sonnet-4-6) goes in gc:agentModelIdentifier.
`SAT-MCP`	MCP Server	A Model Context Protocol server process exposing tools, resources, and prompts to LLM agents. The agent IS the server (in PROV-O terms, an active prov:SoftwareAgent); the corresponding OfferedService instance (with gc:hasServiceInterfaceType gc:SIT-MCP) describes the wire contract.
`SAT-RAG`	RAG Pipeline	A retrieval-augmented-generation pipeline: retrieves from a vector store or knowledge graph, conditions an LLM call on the retrieved context, returns a grounded answer. The engage AI agent's gap-analysis pipeline is the canonical example in this ecosystem.
`SAT-WHK`	Webhook Handler	A long-running process that consumes inbound webhook deliveries from upstream systems. The GC-Core sync Edge Functions (sync-need-to-graphdb, sync-commitment-to-graphdb, sync-org-to-graphdb) are canonical examples — they consume pg_net deliveries and write to GraphDB.