mem-bridge

Extract personal professional knowledge from your workday — privately, portably, automatically.

Install

npm install -g mem-bridge

Two Commands, Two Machines

COMPANY MACHINE                         YOUR MACHINE
───────────────                         ────────────
                                        
  Claude sessions ─┐                    inbox/
  Claude memory    ─┤                     memory-bridge-2026-04-08.zip
  Claude plans     ─┤                        │
  Claude skills    ─┤                        ▼
  Codex rollouts   ─┼→ mem-bridge harvest  mem-bridge ingest once
  Cursor rules     ─┤      │                 │
  Gemini context   ─┤      ▼                 ▼
  Git commits      ─┤  Redact patterns    daily/2026-04-08.md
  Modified docs    ─┘  Extract via LLM       │
                       Review (optional)     ▼
                           │             MEMORY.md
                           ▼
                   ~/Desktop/memory-bridge-2026-04-08.zip
                           │
                     ── air gap ──
                    (AirDrop, USB,
                     personal sync)

Sanitization happens on the company machine. Proprietary data never crosses the gap.

Daily Workflow

End of day (company machine)

mem-bridge harvest           # Interactive: review each learning
mem-bridge harvest --auto    # Trust the sanitization, skip review
mem-bridge harvest --dry-run # See what would be scanned

The harvester scans:

| Source | What it finds | |--------|--------------| | ~/.claude/projects/*/memory/ | Auto-memory, auto-dream consolidated knowledge | | ~/.claude/projects/*/sessions/ | Session transcripts (JSONL) | | ~/.claude/projects/*/plans/ | Generated execution plans | | ~/.claude/projects/*/skills/ | Generated skills | | ~/.claude/todos/ | Todo items | | ~/.claude/settings.json | Global settings with memory entries | | CLAUDE.md in git repos | Project-level instructions | | .claude/plans/, .claude/skills/ in repos | Project-level plans and skills | | ~/.codex/ | Codex memory, rollouts, skills | | AGENTS.md in git repos | Codex project instructions | | .cursorrules, .cursor/ | Cursor context files | | ~/.gemini/, GEMINI.md | Gemini CLI context | | Git repos | Today's commit messages | | ~/Documents, ~/projects, etc. | Markdown files modified today |

Output: ~/Desktop/memory-bridge-2026-04-08.zip (2-4 KB, fully sanitized)

Transfer (30 seconds)

Move the .zip to your personal machine. AirDrop, USB, personal email, Signal — whatever you prefer. The file is already clean.

Ingestion (personal machine, automatic)

# Watch mode (run once, keeps watching)
mem-bridge ingest watch

# Or process inbox manually
mem-bridge ingest once

# Or ingest a specific file
mem-bridge ingest file ~/Downloads/memory-bridge-2026-04-08.zip

Drop .zip files into ~/.memory-bridge/inbox/. The watcher picks them up automatically.

Weekly consolidation

mem-bridge ingest consolidate           # Last 7 days
mem-bridge ingest consolidate --days 14 # Last 14 days

Uses your Anthropic API key to consolidate daily notes into MEMORY.md.

Check status

mem-bridge ingest status

Sanitization Pipeline

Every piece of knowledge passes through three stages:

1. Pattern redaction — Regex strips emails, internal URLs, API keys, Slack tokens, ticket IDs, UUIDs, IP addresses, file paths, dollar amounts. Add company-specific patterns via company_patterns.yaml.

2. LLM generalization — Tries Anthropic API (if ANTHROPIC_API_KEY is set), then Ollama (if running), then heuristic fallback. The LLM converts specifics to general patterns: "Fixed the auth service rate limiter" → "Learned to debug rate limiting in auth middleware."

3. Human review — Each extracted learning shown for approval. Skip with --auto once you trust the pipeline.

Company Patterns

Copy company_patterns.example.yaml to company_patterns.yaml in your working directory:

- pattern: "\\bAcmeCorp\\b"
  replacement: "[COMPANY]"
- pattern: "\\bProject\\s+Phoenix\\b"
  replacement: "[PROJECT]"

These run before the LLM sees anything. Update when you change jobs.

Memory Structure

~/.memory-bridge/
├── inbox/                         # Drop .zip files here
├── archive/                       # Processed .zip files
└── memory/                        # Git-versioned memory store
    ├── .git/
    ├── MEMORY.md                  # Long-term consolidated knowledge
    ├── daily/
    │   ├── 2026-04-01.md
    │   ├── 2026-04-02.md
    │   └── ...
    └── skills/                    # (Future) technique library

All markdown. All git-versioned. Portable to any future tool.

Environment Variables

| Variable | Used where | Purpose | |----------|-----------|---------| | ANTHROPIC_API_KEY | Both machines | LLM extraction and consolidation | | MEMORY_BRIDGE_HOME | Personal machine | Override ~/.memory-bridge path |

Requirements

• Node.js 20+
• Git (for memory versioning)
• Optional: Ollama (if no Anthropic API key on company machine)

License

MIT