TotalRecalls Library Format

TotalRecalls stores every conversation in a clean, predictable, and vendor-neutral directory hierarchy on your local drive. By default, files are placed in %USERPROFILE%\TotalRecalls-download\Library\, but you can select any internal drive, external SSD, or network-attached storage volume.

Directory Structure Overview

The library organizes chats into dedicated subfolders by provider, with a central metadata tracking directory:

%USERPROFILE%\TotalRecalls-download\Library\
├── chatgpt\
│   ├── 2026-08-09-how-to-refactor-react.md
│   └── 2026-08-09-how-to-refactor-react.json
├── claude\
│   ├── 2026-08-10-distributed-consensus.md
│   └── 2026-08-10-distributed-consensus.json
├── perplexity\
├── gemini\
├── grok\
├── deepseek\
├── mistral\
├── qwen\
├── metadata\
│   └── export-manifest.json
└── README.txt

The Dual-File Architecture (.md + .json)

Every single conversation is saved as an atomic pair of files sharing an identical base filename formatted as YYYY-MM-DD-sanitized-title:

  • Human-Readable Markdown (.md): Structured plain text formatted with standard CommonMark syntax. Each file begins with a YAML frontmatter block recording the original conversation title, source provider, model name, creation timestamp, and unique conversation identifier. Message turns are rendered with clear role headers (### User, ### Assistant), code blocks preserve their original syntax highlighting tags, and mathematical LaTeX expressions stay intact.
  • Lossless Raw JSON (.json): A complete, uncompressed JSON object containing the exact original message tree, full token metadata, tool calls, thinking/reasoning blocks (such as DeepSeek-R1 traces), and web search citation URLs. This structure guarantees that future tools, custom parsers, and vector indexers have access to the raw payload without information loss.

Metadata Manifest (metadata/export-manifest.json)

TotalRecalls maintains an incremental manifest file to enable fast synchronization and deduplication:

{
  "version": "1.0.0",
  "last_export": "2026-11-12T14:30:00Z",
  "providers": {
    "chatgpt": { "conversation_count": 106, "file_pairs": 106 },
    "claude": { "conversation_count": 84, "file_pairs": 84 }
  }
}

Third-Party Integration & Longevity

Because TotalRecalls writes standard text files rather than a proprietary database blob:

  • Obsidian: Open the Library\ folder directly as an Obsidian Vault to explore your chats with bi-directional wikilinks and graph views.
  • Notion: Use Notion's native Markdown import utility to bring research threads into shared workspaces.
  • Windows Search & ripgrep: Instant full-text search across thousands of files directly from Windows File Explorer or developer command-line tools.
  • Local Backups: Compatible with Windows File History, OneDrive, Backblaze, and Git version control out of the box.