Personal Context Agent

Extract personal context from AI chat transcripts and structure into themed files. Designed to work with Claude Haiku for cost-effective batch processing.

Workflow

Chat Exports → Process → Extract → Chunk → Integrate

Reference Files

File	Purpose
references/transcript-processor.md	Convert JSON/text exports to clean transcripts
references/context-extractor.md	Extract personal facts from conversations
references/context-chunker.md	Split context into themed markdown files
references/batch-processing.md	Handle multiple transcripts, merge context

Quick Start

Single Transcript

1. Process: Convert export to clean text (see transcript-processor.md)
2. Extract: Pull out personal context (see context-extractor.md)
3. Chunk: Split into themed files (see context-chunker.md)

Multiple Transcripts

See batch-processing.md for:

• Sequential processing (Haiku-friendly)
• Merging extracted context
• Incremental updates

Source Formats

• Claude.ai JSON exports
• Claude Code session logs
• ChatGPT JSON exports
• Plain text/markdown chat logs

Output Files

Use kebab-case: work-context.md, technical-stack.md

Standard categories:

• personal-background.md - Name, location, family
• work-context.md - Job, role, company
• technical-stack.md - Languages, tools, frameworks
• current-projects.md - Active work with details
• preferences.md - Likes, dislikes, opinions
• constraints.md - Limitations, requirements
• goals.md - Objectives
• communication-style.md - Interaction preferences

Integration

Platform	Limit	Method
Claude Projects	None	Project Knowledge
Custom GPTs	10 files	Knowledge upload
Gems	10 files	Context files
Claude Code	None	CLAUDE.md or docs/

Haiku Optimization

These prompts use:

• Short, direct instructions
• Explicit output formats
• Concrete examples
• Step-by-step structure
• No ambiguity

For best results with Haiku:

• Process one transcript at a time
• Keep input under 50K tokens
• Use the exact prompts provided
• Review and merge outputs manually if needed