Docs/Memory

Memory

Your AI assistant remembers context, preferences, and history across sessions — so you never repeat yourself.

How Memory Works

Unlike chatbots that forget everything after you close the window, Pinchr's agent builds persistent memory over time:

First conversation:
"I prefer dark mode and minimal logging. Also, I usually work in ~/Projects/acme-app."
Two days later:
"Can you check the latest git status?"
Agent:
Let me check ~/Projects/acme-app (your main project)...

Your agent doesn't just remember facts — it learns your workflow patterns, preferences, and context to work faster over time.

MEMORY.md — Long-Term Storage

Your agent maintains a MEMORY.md file with key information it should always remember:

📝
~/Library/Application Support/Pinchr/MEMORY.md

A markdown file your agent reads at the start of every conversation.

# User Preferences
- Name: Alex
- Primary project: ~/Projects/acme-app
- Prefers TypeScript over JavaScript
- Uses Vitest for testing
# Workflow
- Commit messages should follow Conventional Commits
- Always run tests before committing
- Deploy to staging via `npm run deploy:staging`

Your agent updates MEMORY.md automatically as it learns about you. You can also edit it manually from Settings → Memory.

Daily Context Files

Pinchr creates a daily context file for each day you use it. These files capture conversations, decisions, and work done that day:

📅
~/Library/Application Support/Pinchr/daily/2026-02-12.md

A running log of the day's work, tasks, and conversations.

# 2026-02-12
## Morning
- Fixed auth bug in login flow (commit: a3f8d2c)
- Refactored API client to use fetch instead of axios
## Afternoon
- Started work on payment integration
- User mentioned they prefer Stripe over PayPal
- Created task: "Integrate Stripe API"

Daily files help your agent recall recent context without searching through every past conversation. They're automatically archived after 30 days (configurable).

Memory Explorer

Browse and search your agent's memory from the Memory Explorer page:

🔍
Semantic Search

Search memory using natural language. "What did I decide about the database migration?"

🗓️
Timeline View

Browse memory by date. See what you and your agent worked on each day.

🏷️
Tags & Topics

Your agent automatically tags memories by topic: projects, people, decisions, bugs.

✏️
Edit or Delete

Remove outdated or incorrect memories. Your agent won't reference them again.

Open Memory Explorer from the sidebar or press⌘⇧M.

What Gets Remembered

Your agent decides what's worth remembering based on importance and repetition:

Automatically Remembered
Your name, role, and preferences
Project names, paths, and structures
Important decisions and their reasoning
Recurring patterns (commands, workflows, file locations)
Tasks, automations, and goals
Integrations and API credentials (encrypted)
Not Remembered (Unless Important)
One-off questions or casual chat
Temporary debugging output
File contents (unless explicitly saved)
Exploratory searches or experiments

You can also manually add memories by telling your agent: "Remember that I prefer Tailwind CSS" or "Save this for later: [note]."

Memory Across Sessions

Every time you start a new conversation, your agent loads:

  • MEMORY.md — Your core preferences and long-term context
  • Recent daily files — The last 7 days of work and conversations
  • Active tasks and projects — Current goals and priorities

This means your agent doesn't start from scratch every time. It knows where you left off and what you're working on.

💡

Pro tip: If your agent seems to have forgotten something, check Memory Explorer to see if it was recorded. You can manually add it if needed.

Managing Memory

You're in full control of what your agent remembers:

Edit MEMORY.md

Open Settings → Memory → Edit MEMORY.md to manually update long-term context. Add or remove preferences, correct mistakes, or clarify information.

Clear Specific Memories

From Memory Explorer, search for a memory and delete it. Your agent won't reference it again.

Reset All Memory

In Settings → Memory → Advanced, you can completely wipe memory and start fresh. Use this if you're handing off Pinchr or want a clean slate.

Export Memory

Export your memory as a ZIP archive for backup or migration to another device.

Memory & Privacy

All memory is stored locally on your Mac. Here's what that means:

MEMORY.md and daily files are never sent to Pinchr servers.
Your agent sends only relevant context to LLM APIs (Anthropic/OpenAI) for reasoning.
You can exclude sensitive topics from memory using filters in Settings.
Memory files are encrypted at rest using macOS FileVault (if enabled).

For more details, see the Security & Privacy documentation.

Questions about memory?

Reach out if you need help managing or understanding your agent's memory.