Bulk Text Index Agents

AI · Multi-Agent Pattern · Bulk Text Ingestion · September 2026

When dealing with massive amounts of text—like multiple days of raw transcripts, interview dumps, or extensive logs—querying an LLM directly is a nightmare. The context window gets bloated, latency goes through the roof, and the model starts hallucinating or losing the thread.

I needed a way to create dedicated agents in tools like Microsoft 365 Copilot that could read all of it, understand the context, and return fast, grounded answers with great quality. No lag. No dropped context.

The prompts and workflow are open source on GitHub at github.com/soundisspirit/bulk-text-index-agents.

How it works

The solution is a multi-agent architectural pattern. Instead of feeding all raw files to a single agent, you split the workload across specialized agent steps:

Step 1 — Agent 1 (Indexer): Runs independently for each file, extracting verbatim anchor phrases, decisions, and local topic segments without summarizing the whole file into generic blurbs.

Step 2 — Agent 2 (Consolidator): Takes all generated local index files and merges them into a single Master Index, tracking position changes and open items.

Step 3 — Agent 3 (Use Case / Query Agent): Uses the Master Index as a fast lookup map to locate the exact raw files and verbatim anchors needed, reading only those files to complete the task.

Use cases: Multi-day meeting Q&A, backlog creation, timeline reconstruction, and executive report generation from unstructured text collections.