Personal AI, powered by markdown.
A personal AI assistant that actually knows you. No app, no server, no Docker. Just markdown files + Claude Code.
Copana gives your AI persistent memory, personality, and proactive behavior — all stored as plain text files you can read, edit, and own.
copana.ai · ~14k tokens · 7% of context window
I built a personal AI assistant that runs in Claude Code, stores everything in markdown files, and actually knows me — my goals, my patterns, my open tasks, my preferences. After a month of use it felt like having a co-founder who never forgets anything.
The problem: it was tightly coupled to my setup. Hardcoded names, personal scripts, specific integrations.
So I extracted the framework. Copana is the generalized version — packaged so anyone can have the same experience in 30 seconds.
No Docker, no server, no build step. Just git clone and go.
git clone https://github.com/PayRequest/copana.git
cd copana
claudeThen run /setup inside Claude Code. It asks your name, AI name, timezone, and vibe — then handles everything including your first commit.
Alternative: Run ./setup.sh directly if you prefer the shell.
For the full walkthrough, see docs/getting-started.md.
Most AI tools forget you the moment you close the tab. Copana doesn't.
- It remembers. Your goals, preferences, and open tasks persist across sessions.
- It learns. Every conversation teaches it something about you — silently captured in markdown.
- It follows up. Open loops don't just sit there. Your AI checks back.
- It thinks ahead. Proactive nudges, daily reflections, and pattern recognition.
- You own it. Plain markdown files, version-controlled with git. No lock-in.
Copana is a template repository for Claude Code. When you run claude in this directory, Claude reads CLAUDE.md automatically — your AI's instructions, startup ritual, and behavior rules.
copana/
├── CLAUDE.md # AI instructions (auto-read by Claude Code)
├── soul.md # AI identity & values
├── personality.md # How the AI talks
├── user.md # About you (AI fills this in over time)
├── preferences.md # Your likes/dislikes
├── memory.md # Long-term memory & session log
├── tasks.md # Active tasks
├── insights.md # Strategy & follow-ups
├── routines.md # Your daily patterns
├── contacts.md # Important people
├── projects.md # Bigger initiatives
├── journal.md # Daily reflections
├── .claude/
│ ├── settings.json # Status line config
│ ├── agents/ # Custom subagents (memory-manager, research, daily-review)
│ ├── output-styles/ # Personality as output style
│ └── skills/ # Setup, customize, add-telegram, etc.
├── scripts/ # Optional automations
└── docs/ # Full documentation
Your AI maintains structured context across sessions:
| File | What it stores | Updated by |
|---|---|---|
memory.md |
Core facts, open loops, decisions, session log | AI (every session) |
user.md |
Who you are — background, goals, personality | AI (learns over time) |
preferences.md |
Quick-lookup likes, dislikes, work style | AI (captures silently) |
tasks.md |
Active tasks, completed items | You + AI |
insights.md |
Advice given, lessons learned, follow-ups | AI |
routines.md |
Your daily/weekly patterns | AI (observes over time) |
contacts.md |
People in your life with context | AI (as they come up) |
Every file is plain markdown. Read it, edit it, version it, delete it. You're always in control.
Every session, your AI automatically:
- Reads
memory.md,tasks.md,insights.mdfor context - Checks
git statusfor changes since last session - Gets the current time
- Starts working — no "how can I help?" filler
This is what makes Copana different from a fresh AI conversation. Your AI picks up exactly where you left off.
Your AI silently captures information about you during conversations:
- You mention you hate phone calls → saved to
preferences.md - You talk about a friend → saved to
contacts.md - You describe your morning routine → saved to
routines.md - You share a life goal → saved to
user.md
No prompting needed. It just listens and learns.
Copana uses Claude Code skills for extensibility. Instead of bloated features, skills teach Claude how to transform your setup.
| Skill | What it does |
|---|---|
/setup |
Interactive first-time setup |
/customize |
Guided customization of any aspect |
/add-telegram |
Set up Telegram bot for mobile access |
/add-fitness |
Add workout tracking and gym accountability |
/add-budget |
Add personal budget tracking |
/add-agent |
Create a custom subagent |
Want to add Slack support? Don't create a PR that adds Slack. Create a skill file that teaches Claude how to add Slack to any Copana installation. Users run the skill and get clean, tailored code.
See CONTRIBUTING.md for details.
Copana ships with specialized Claude Code subagents that handle background work:
| Agent | What it does | Model |
|---|---|---|
memory-manager |
Consolidates memory files, removes duplicates, archives old sessions | Haiku (fast) |
research |
Web research with structured output, saved to your files | Default |
daily-review |
Reviews all files, writes daily summary to journal.md | Haiku (fast) |
Subagents have restricted tool access (e.g., memory-manager can't browse the web) and persistent memory across sessions.
Create your own with /add-agent or drop a markdown file in .claude/agents/.
Copana adds a status line to your Claude Code terminal showing:
- Open tasks from tasks.md
- Open loops from memory.md
- Last session timestamp
Always visible, refreshes every 30 seconds. Configured in .claude/settings.json.
Copana includes an output style (.claude/output-styles/copana.md) that shapes how your AI communicates — direct, casual, no filler. This works alongside personality.md to give your AI a consistent voice without bloating CLAUDE.md.
Copana works great with just claude and markdown files. But if you want proactive behavior between sessions, there are optional automation scripts:
| Script | What it does |
|---|---|
| Morning briefing | Tasks, loops, and an AI nudge to start your day |
| Proactive pings | Smart nudges about stale items (birthdays, overdue tasks) |
| Evening check-in | Quick energy/mood capture via Telegram |
| Daily reflection | AI reviews your files overnight, prepares tomorrow's nudge |
| Weekly review | Patterns, wins, and focus areas |
| News briefing | Relevant news from configurable RSS feeds |
| Telegram bot | Chat with your AI on mobile |
Setup: pip install -r scripts/requirements.txt + add API keys to .env.
See docs/automation.md for details.
Copana syncs via git. Work on your Mac with Claude Code, capture thoughts on mobile via the Claude app + GitHub, and everything stays in sync.
See docs/multi-device.md for setup.
| Copana | NanoClaw | OpenClaw | |
|---|---|---|---|
| Source files | ~50 | 15 | 3,680 |
| Lines of code | ~3,600 | ~3,900 | 434,453 |
| Dependencies | 5 | <10 | 70 |
| Config files | 1 | 0 | 53 |
| Time to understand | 8 minutes | 8 minutes | 1–2 weeks |
| Security model | OS container isolation | OS container isolation | Application-level checks |
| Architecture | Single process + isolated containers | Single process + isolated containers | Single process, shared memory |
| Setup | git clone + /setup |
Docker + Node.js | npm install -g + onboard wizard |
| Infrastructure | None | Container runtime | Node ≥22 |
| Memory | Structured markdown (12 files) | Per-group CLAUDE.md | Per-agent sessions |
| Personality | soul.md + output style |
None | None |
| Personal context | user.md + preferences.md |
None | Workspace config |
| Active learning | Captures facts silently | None | None |
| Proactive behavior | Nudges, reflections, briefings | Scheduled tasks | Cron + webhooks |
| Subagents | memory-manager, research, daily-review | None | Multi-agent routing |
| Channels | CLI + mobile (git sync) | CLI | WhatsApp, Telegram, Slack, Discord, Signal, iMessage + 6 more |
| Status line | Open tasks, loops, last session | None | None |
| Extensibility | Skills + agents | Skills | Skills + tools + channels |
| Privacy | Local files, git-versioned | Container | Local gateway (127.0.0.1) |
| Token footprint | ~14k tokens (7% of window) | ~35k tokens (17%) | — |
Do I need Docker?
No. Copana has zero infrastructure requirements. Just git clone and claude.
Does it work with other AI tools besides Claude Code? The markdown files work with anything that can read text. The CLAUDE.md instructions and skills are Claude Code specific, but the memory system is portable.
How much context does it use? ~14k tokens total (~7% of Claude's context window). Core files read every session are only ~2.6k tokens. Your AI has plenty of room to work.
Is my data private? Everything stays on your machine unless you push to GitHub. No telemetry, no analytics, no cloud processing. Use a private repo for backup.
Will it slow down Claude Code? No. Claude reads markdown files extremely fast. The startup ritual adds 2-3 seconds to read your context files — after that, it's normal Claude Code speed.
What if my memory files get too big?
Archive old session logs to memory/archive/. Keep memory.md focused on current facts and recent sessions. The AI can help with this — just ask it to consolidate.
Can I use this with the Claude mobile app? Yes. Push to a private GitHub repo, then reference your files in Claude.ai conversations. Or use the Telegram bot for a more seamless mobile experience.
How is this different from just putting instructions in CLAUDE.md? Copana is a structured system, not just a prompt. It includes a memory architecture (12 interconnected files), a startup ritual, active learning behaviors, proactive follow-ups, and optional automation scripts. You could build this yourself — Copana saves you the work.
- Getting Started — Setup walkthrough and first steps
- Memory System — How the 12-file memory architecture works
- Automation — Morning briefings, nudges, Telegram bot
- Multi-Device — Sync across Mac and mobile
- Philosophy — Why plain text, why markdown, why local
Copana is built on a simple idea: your AI should know you.
Not through a proprietary database or a cloud service — through plain text files that you own, stored in a git repo you control.
The best personal AI isn't the smartest model. It's the one with the most context about you.
Don't add features. Add skills.
See CONTRIBUTING.md.
MIT — do whatever you want with it.
Built by PayRequest. Powered by Claude Code.