Agent-as-LLM: The Pattern That Makes Persistent MCP Memory Free to Run

Most AI dev tools bill you per file indexed because they run their own LLM on their infra. The agent-as-LLM pattern flips this — the calling agent does the semantic work in its own context, so the MCP server is pure persistence.

TL;DR: Traditional AI dev tools pay LLM costs on their own infrastructure, so they bill per file or per token. The agent-as-LLM pattern delegates all semantic work (purpose tagging, embeddings, summaries) to the calling agent's LLM context. The MCP server is pure persistence with zero LLM spend. Tentra is free during beta because indexing literally costs us nothing.