Persistia Memory System

Persistia Memory System

Skill by ara.so — Claude Code Skills collection.

Persistia gives Claude Code persistent, self-updating memory of your entire project. It scans your codebase, architecture, docs, and configs on first run, then monitors changes via git diff to stay current. Memory is stored in _brain/ as plain text files with git versioning, optionally backed up to a private GitHub repository.

Installation

Run from your project root:

curl -fsSL https://raw.githubusercontent.com/bernardohcrocha/persistia-for-claude-code/main/setup.sh | bash

Requirements:

• git
• Node.js 18+
• Optional: GitHub CLI (gh) for automatic cloud backup

The setup process will:

1. Scan your project structure
2. Create _brain/ directory with indexed memory
3. Ask only for information it can't find automatically
4. Install schedulers (launchd on macOS, systemd on Linux)
5. Create a private git repository for the brain
6. Set up automatic daily updates

Directory Structure

_brain/
├── .git/                    # Isolated brain repository
├── index.md                 # Main navigation file (read first every session)
├── dashboard.html           # Live dashboard (auto-refreshes every 5 min)
├── core/                    # Product, brand, ICP definitions
│   ├── product.md
│   ├── brand.md
│   └── icp.md
├── operations/              # Metrics and customer data (auto-updated daily)
│   ├── metrics.md
│   └── customers.md
├── skills/                  # Permanent rules and capabilities
│   └── [skill-name].md
└── tasks/                   # Scheduled task management
    ├── queue.json
    ├── completed/
    └── failed/

Core Concepts

Memory Index

The _brain/index.md file is always read first. It contains:

• Project overview and architecture
• Links to all memory files
• Recent changes summary
• Navigation structure

Self-Updating Memory

Persistia runs git diff daily to detect changes:

• Only modified files are re-indexed
• Token-efficient: 1,000 files with 3 changes = reads 3 files
• Automatic commit and push to brain repository

Skills

Skills are permanent capabilities stored in _brain/skills/. Once written, they're loaded in every session.

Creating a skill:

---
name: deployment-checklist
created: 2026-06-04
---

# Deployment Checklist

Before every deployment:
1. Run full test suite
2. Check environment variables match production
3. Verify database migrations are reversible
4. Update CHANGELOG.md
5. Tag release in git

Skills are created automatically when you say things like:

• "Remember to always run tests before deploying"
• "Add a skill for handling customer data exports"
• "Create a permanent rule about error handling"

Scheduled Tasks

Tasks are defined in _brain/tasks/queue.json:

{
  "tasks": [
    {
      "id": "weekly-metrics",
      "name": "Weekly Metrics Report",
      "schedule": "every Monday at 9am",
      "description": "Pull last week's metrics from Stripe, flag anomalies, queue follow-ups for accounts below quota",
      "batch_size": 50,
      "timeout_minutes": 10,
      "enabled": true
    },
    {
      "id": "inactive-users",
      "name": "Inactive User Outreach",
      "schedule": "every Tuesday",
      "description": "Find signups from last 30 days with no activity. Research each company and draft personalized re-engagement emails",
      "batch_size": 20,
      "timeout_minutes": 15,
      "enabled": true
    }
  ]
}

Task properties:

• batch_size: Maximum items per run (prevents timeouts)
• timeout_minutes: Default is 5, increase for complex tasks
• enabled: Set to false to pause without deleting

Proactive Scans

Every 3 days when idle, Persistia automatically scans:

• Metrics trends
• Customer behavior patterns
• Support channels
• Code changes

It flags observations and suggests actions but never acts unilaterally.

Common Usage Patterns

Initial Setup Flow

After installation, provide context through conversation:

You: "This is a SaaS analytics platform built with Next.js and PostgreSQL"
Claude: *Updates _brain/core/product.md with architecture details*

You: "We target B2B marketing teams at companies with 50-500 employees"
Claude: *Updates _brain/core/icp.md with target customer profile*

Reading Environment Variables

Persistia scans .env files automatically. Reference credentials:

// Good - uses existing environment variables
const stripe = require('stripe')(process.env.STRIPE_SECRET_KEY);
const db = new Database({
  host: process.env.DB_HOST,
  password: process.env.DB_PASSWORD
});

Adding Operational Skills

You: "Always check Sentry for errors before marking a task complete"
Claude: *Creates _brain/skills/pre-completion-checklist.md*

You: "When analyzing churn, segment by signup cohort and usage tier"
Claude: *Creates _brain/skills/churn-analysis-methodology.md*

Scheduling Recurring Tasks

You: "Every Friday afternoon, summarize this week's closed issues and PRs into a changelog draft"
Claude: *Adds task to _brain/tasks/queue.json*

You: "Daily at 8am: check for new Stripe subscriptions and add them to the customer database"
Claude: *Adds task with appropriate batch_size and timeout*

Cross-Tool Queries

You: "Which customers dropped usage 30%+ this month? Cross-check their support tickets and draft personalized follow-ups"
Claude: *Queries metrics, support system, generates drafts*

You: "Find all users who signed up from our latest blog post and haven't activated yet"
Claude: *Checks analytics, CRM, prepares outreach list*

Dashboard

Open _brain/dashboard.html in your browser for a live view:

• Recent memory updates
• Scheduled task status
• Proactive scan results
• Quick navigation to all brain files

Auto-refreshes every 5 minutes.

Manual Operations

Force Memory Update

cd _brain
git add .
git commit -m "Manual memory update"
git push

View Memory History

cd _brain
git log --oneline
git diff HEAD~1  # See what changed in last update

Share Memory with Team

# On first machine
cd _brain
git remote -v  # Get repository URL

# On second machine (in project root)
git clone <brain-repo-url> _brain

Disable Automatic Updates

# macOS
launchctl unload ~/Library/LaunchAgents/persistia.brain.plist

# Linux
systemctl --user disable persistia-brain.timer

Re-enable Automatic Updates

# macOS
launchctl load ~/Library/LaunchAgents/persistia.brain.plist

# Linux
systemctl --user enable persistia-brain.timer
systemctl --user start persistia-brain.timer

Troubleshooting

Brain repository not pushing to GitHub

Check if GitHub CLI is authenticated:

gh auth status

If not authenticated:

gh auth login

Memory updates not running automatically

Check scheduler status:

# macOS
launchctl list | grep persistia

# Linux
systemctl --user status persistia-brain.timer

View logs:

# macOS
tail -f ~/Library/Logs/persistia-brain.log

# Linux
journalctl --user -u persistia-brain.service -f

Task timeouts

Increase timeout_minutes in _brain/tasks/queue.json:

{
  "id": "complex-analysis",
  "timeout_minutes": 20,
  "batch_size": 10
}

Memory files growing too large

Split large files by topic. Update _brain/index.md to link to all sections:

## Architecture

- [Backend Services](core/architecture-backend.md)
- [Frontend Components](core/architecture-frontend.md)
- [Database Schema](core/architecture-database.md)

Resetting memory completely

rm -rf _brain
curl -fsSL https://raw.githubusercontent.com/bernardohcrocha/persistia-for-claude-code/main/setup.sh | bash

This will re-scan your project and rebuild all memory files.

Best Practices

1. Be conversational during setup - Persistia learns from natural conversation, not forms
2. Review proactive scans - Check suggestions before acting on them
3. Keep skills focused - One skill per workflow or rule
4. Use batch_size for large datasets - Prevents timeouts and token limits
5. Commit brain changes - Treat _brain/ like important code
6. Back up to GitHub - Install gh CLI for automatic cloud backup
7. Share with team - Clone brain repository across machines for consistent context

Integration Examples

Stripe Metrics

// _brain/skills/stripe-metrics.md references this pattern
const stripe = require('stripe')(process.env.STRIPE_SECRET_KEY);

async function getMonthlyMetrics() {
  const subscriptions = await stripe.subscriptions.list({
    status: 'active',
    limit: 100
  });
  
  const revenue = subscriptions.data.reduce((sum, sub) => {
    return sum + (sub.items.data[0].price.unit_amount / 100);
  }, 0);
  
  return { active_subscriptions: subscriptions.data.length, revenue };
}

GitHub Issue Analysis

// Uses existing gh CLI authentication
const { execSync } = require('child_process');

function getRecentIssues(days = 7) {
  const since = new Date(Date.now() - days * 24 * 60 * 60 * 1000).toISOString();
  const cmd = `gh issue list --state closed --json title,closedAt,labels --search "closed:>=${since}"`;
  const output = execSync(cmd, { encoding: 'utf-8' });
  return JSON.parse(output);
}

Customer Database Updates

// Credentials from environment
const { Pool } = require('pg');
const pool = new Pool({
  host: process.env.DB_HOST,
  database: process.env.DB_NAME,
  user: process.env.DB_USER,
  password: process.env.DB_PASSWORD
});

async function updateCustomerMetrics() {
  const result = await pool.query(`
    SELECT customer_id, last_active, usage_count
    FROM customers
    WHERE last_active < NOW() - INTERVAL '30 days'
  `);
  return result.rows;
}

Architecture

Persistia uses an isolated git repository for _brain/:

• Separate from main project repository
• Pushes to private GitHub remote (if gh CLI available)
• Contains only memory files, no code
• Can be cloned independently across machines
• Full version history of memory evolution

See the project's ARCHITECTURE.md for detailed git workflow.