Roampal - AI That Grows With You

I built this because I wanted a personal AI assistant that could learn and grow with me. I needed something private and local.

THE PROBLEM

Every conversation with cloud AI starts from scratch. You explain your preferences, your workflow, what you're working on. Next day? You're explaining it all over again.

Even worse: they don't learn from outcomes. If their advice fails, they'll suggest it again tomorrow. If something works perfectly, they won't remember to prioritize it next time.

I wanted an AI that actually gets smarter the longer I use it. And I wanted my conversations to stay on my machine.

The AI compounds in intelligence. Week 10 is way smarter than Week 1.

HOW IT WORKS

5-tier memory system:

1. Memory Bank - Stores who YOU are. Identity, preferences, tech stack. Permanent. "I use Python 3.11, prefer FastAPI, hate OOP" → Remembers forever, never suggests OOP
2. Outcome tracking - Learns what works:
   • Success = +0.2
   • Failure = -0.3
   • Score ≥0.7 → Moves to long-term memory
   • Score <0.2 → Gets deleted (stops bad advice from persisting)
3. Knowledge Graph - Visual map of what the AI knows:
   • Green nodes = proven patterns (70%+ success)
   • Red nodes = approaches that failed
   • Larger nodes = more usage + higher confidence
4. Knowledge Base - Upload your docs (.txt, .md). AI references YOUR conventions instead of generic advice.
5. Cross-conversation memory - Patterns learned in Week 1 still apply in Week 10.

The AI has autonomy to manage its own understanding. It decides what's worth remembering, how to score things, what patterns matter. You can override, but it learns organically—not following rigid rules I hardcoded.

WHAT YOU GET

⚠️ THIS IS ALPHA - Core features work, rough edges exist, active development.

Included: ✅ Roampal.exe for Windows 10/11 ✅ Embedded Python (zero setup hassle) ✅ 5-tier memory system (working) ✅ Outcome tracking + learning (working) ✅ Knowledge graph visualization (working) ✅ Document upload (working) ✅ Works with any Ollama model (Llama, Mistral, Qwen, etc.) ✅ Lifetime updates

Alpha means:

• Core features fully functional
• More features incoming
• Occasional bugs (GitHub issues are public)
• Your feedback directly shapes priorities

💰 $9.99 - Early adopter pricing. Lock it in forever.

WHY I BUILT IT THIS WAY

I wanted a personal AI assistant that could learn and grow with me. Not reset every conversation. Not store my data in someone's cloud.

Started building it with ChatGPT's help. Hit walls, switched to Gemini for refactoring. Used Grok for code the others couldn't handle. Eventually found Cursor, then Claude Code. Piece by piece, it came together.

I overengineered it massively at first. Psychology background, wanted to model human memory perfectly. Built belief trees, sentiment learning, rigid scoring systems. Went way overboard.

Stripping that down revealed what actually mattered: giving the LLM autonomy to learn, not forcing it into my framework.

The result: ✗ No subscriptions (one payment, yours forever) ✗ No cloud (100% local via Ollama) ✗ No telemetry (your conversations never leave your machine) ✗ No bait-and-switch (open source from day one)

✓ MIT License (verify everything yourself) ✓ True ownership (export/delete data anytime) ✓ Transparent development (GitHub issues public) ✓ Built by one person who wanted better

SETUP

Requirements:

• Windows 10/11 (Mac/Linux in development)
• 8GB RAM minimum (16GB for larger models)
• 10GB disk space + model storage (4-40GB depending on model size)
• Ollama (free from ollama.ai)

Steps:

1. Download & extract zip
2. Install Ollama
3. Run Roampal.exe (Don't forget to run as administrator!)
4. Download your first model in the UI interface.
5. Start chatting

Takes about 5 minutes.

WHO THIS IS FOR

✓ People who want AI that improves over time, not resets daily ✓ Anyone tired of re-explaining themselves every conversation ✓ Privacy-conscious people (local > cloud) ✓ Tinkerers running local LLMs ✓ People allergic to subscription software

WHO THIS ISN'T FOR

✗ Wanting plug-and-play ChatGPT (requires Ollama setup) ✗ Uncomfortable with alpha software (expect occasional bugs) ✗ Mac/Linux users (Windows only right now) ✗ Needing enterprise support (solo dev, email/Discord support)

TRANSPARENCY

⚠️ AI Safety: LLMs can hallucinate or generate incorrect information. Always verify critical info. Not for medical, legal, or financial advice.

🔒 Privacy: Everything runs locally. Zero telemetry. No analytics. No phone-home. Source code is public—verify it yourself.

📧 Support:

• Discord: https://discord.gg/qM4yEXfF
• Email: roampal@protonmail.com
• GitHub: github.com/roampal-ai/roampal

Solo developer. I actually respond.

Built by one person who wanted a personal AI assistant that could learn and grow with me—and stay private.

Get in early. Help shape what this becomes!