Skip to content
Back to search

roampal

Memory that learns what works.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/roampal-ai--roampal/badge.svg)](/listings/roampal-ai--roampal)

HTML

<a href="/listings/roampal-ai--roampal"><img src="/api/listings/roampal-ai--roampal/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

1. **[Download from roampal.ai](https://roampal.ai)** and extract 2. Install [Ollama](https://ollama.com) or [LM Studio](https://lmstudio.ai) 3. Right-click `Roampal.exe` → **Run as administrator** 4. Download a model in the UI → Start chatting!

Your AI starts learning about you immediately.

---

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

README.md · README · 196 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. # Roampal
  2. [![Status](https://img.shields.io/badge/status-alpha-orange)](https://roampal.ai)
  3. [![Python 3.10+](https://img.shields.io/badge/python-3.10+-blue.svg)](https://www.python.org/downloads/)
  4. [![Built with Tauri](https://img.shields.io/badge/Built%20with-Tauri-FFC131?logo=tauri)](https://tauri.app/)
  5. [![Multi-Provider](https://img.shields.io/badge/LLM-Ollama%20%7C%20LM%20Studio-blue)](https://roampal.ai)
  6. [![License](https://img.shields.io/badge/license-Apache%202.0-green.svg)](LICENSE)
  7. **Memory that learns what works. So you can do more of it.**
  8. *Say it worked. Say it didn't. The AI remembers.*
  9. Stop re-explaining yourself every conversation. Roampal remembers outcomes, learns from feedback, and gets smarter over time—all 100% private and local.
  10. <p align="center">
  11. <img src="screenshots/whatwebeenupto.png" alt="Roampal - AI Chat with Persistent Memory" width="800">
  12. </p>
  13. > **85.8% non-adversarial on LoCoMo (1,986 questions). +23 pts over raw ingestion. Absorbs 1,135 poison memories losing only 4 pts.** [(Paper)](https://github.com/roampal-ai/roampal-labs)
  14. <p align="center">
  15. <a href="https://github.com/roampal-ai/roampal">
  16. <img src="https://img.shields.io/github/stars/roampal-ai/roampal?style=social" alt="GitHub Stars">
  17. </a>
  18. </p>
  19. ---
  20. ## Benchmark Results
  21. *LoCoMo dataset (1,986 questions, 5 categories, corrected ground truths). Evaluated with [roampal-labs](https://github.com/roampal-ai/roampal-labs). Dual-graded by local 20B + MiniMax M2.7.*
  22. | Metric | Result |
  23. |--------|--------|
  24. | **Non-adversarial accuracy (MiniMax-regraded)** | **85.8%** |
  25. | **Overall (all 5 categories)** | **76.6%** |
  26. | **vs raw ingestion baseline** | **+23 pts** (76.6% vs 53.0%, p<0.0001) |
  27. | **Poison resilience** | **-4.2 pts** after 1,135 adversarial memories |
  28. | **No-memory baseline** | 6.0% (model has zero LoCoMo knowledge) |
  29. | **Architecture vs model** | Architecture: +23 pts. Model swap (GPT-4o-mini): 1.5-2.5 pts |
  30. - System learns through natural conversation, not transcript ingestion
  31. - Absorbs 1,135 poison memories with spoofed trust signals, retaining 72.4% accuracy
  32. - Wilson scoring hurts retrieval at every stage (p<0.001) — removed from ranking
  33. <details>
  34. <summary>Component-level retrieval ablation</summary>
  35. | Config | Hit@1 Clean | Hit@1 Poison | p-value |
  36. |--------|-------------|--------------|---------|
  37. | **TagCascade + cosine** | **27.3%** | **29.0%** | **baseline** |
  38. | Overlap + cosine | 25.8% | 28.0% | p=0.0003 |
  39. | Pure CE | 25.4% | 28.4% | — |
  40. | TagCascade + Wilson | 23.0% | 25.0% | p<0.0001 |
  41. - Cross-encoder: +17.8 Hit@1 over cosine (p<0.0001)
  42. - Tag routing (two-lane): +6.1 Hit@1 clean, +7.5 poison (p<0.0001)
  43. - Wilson: -4.3 Hit@1 in every configuration
  44. - Nursery slot: zero benefit (p=1.0)
  45. Full methodology in [roampal-labs](https://github.com/roampal-ai/roampal-labs)
  46. </details>
  47. ---
  48. ## Quick Start
  49. 1. **[Download from roampal.ai](https://roampal.ai)** and extract
  50. 2. Install [Ollama](https://ollama.com) or [LM Studio](https://lmstudio.ai)
  51. 3. Right-click `Roampal.exe` → **Run as administrator**
  52. 4. Download a model in the UI → Start chatting!
  53. Your AI starts learning about you immediately.
  54. ---
  55. ## Table of Contents
  56. - [Benchmark Results](#benchmark-results)
  57. - [Key Features](#key-features)
  58. - [MCP Integration](#mcp-integration)
  59. - [Architecture](#architecture)
  60. - [Supported Models](#supported-models)
  61. - [Documentation](#documentation)
  62. - [Pricing](#pricing)
  63. ---
  64. ## Key Features
  65. **Memory That Learns**
  66. - Outcome tracking: Scores every result (+0.2 worked, -0.3 failed)
  67. - Smart promotion: Good advice becomes permanent, bad advice auto-deletes
  68. - Cross-conversation: Recalls from ALL past chats
  69. **Your Knowledge Base**
  70. - Memory Bank: Permanent storage of preferences, identity, goals
  71. - Books: Upload .txt/.md docs as searchable reference
  72. - Pattern recognition: Detects what works across conversations
  73. **Privacy First**
  74. - 100% local: All data on your machine
  75. - Works offline: No internet after model download
  76. - No telemetry: Your data never leaves your computer
  77. ---
  78. ## MCP Integration
  79. Connect Roampal to **Claude Desktop, Cursor**, and other MCP-compatible tools.
  80. ```
  81. Settings → Integrations → Connect → Restart your tool
  82. ```
  83. **7 tools available**: `search_memory`, `add_to_memory_bank`, `update_memory`, `archive_memory`, `get_context_insights`, `record_response`, `score_memories`
  84. [Full MCP documentation →](dev/docs/architecture.md#mcp-integration)
  85. ---
  86. ## Architecture
  87. ```
  88. ┌─────────────────────────────────────────────────────────┐
  89. │ 5-TIER MEMORY │
  90. ├─────────────┬─────────────┬─────────────┬──────────────┤
  91. │ Books │ Working │ History │ Patterns │
  92. │ (permanent) │ (24h) │ (30 days) │ (permanent) │
  93. ├─────────────┴─────────────┴─────────────┴──────────────┤
  94. │ Memory Bank │
  95. │ (permanent user identity/prefs) │
  96. └─────────────────────────────────────────────────────────┘
  97. ```
  98. **Core Technology:**
  99. - TagCascade Retrieval: Tag-routed search + cross-encoder reranking (ONNX)
  100. - Outcome-Based Learning: Memories adapt based on feedback
  101. - Sidecar LLM: Background model summarizes exchanges, extracts facts and tags
  102. [Architecture deep-dive →](dev/docs/architecture.md)
  103. ---
  104. ## Supported Models
  105. Works with any tool-calling model via Ollama or LM Studio:
  106. | Model | Provider | Parameters |
  107. |-------|----------|------------|
  108. | Llama 3.x | Meta | 3B - 70B |
  109. | Qwen 2.5 | Alibaba | 3B - 72B |
  110. | Mistral/Mixtral | Mistral AI | 7B - 8x22B |
  111. | GPT-OSS | OpenAI (Apache 2.0) | 20B - 120B |
  112. ---
  113. ## Documentation
  114. | Document | Description |
  115. |----------|-------------|
  116. | [Architecture](dev/docs/architecture.md) | 5-tier memory, retrieval pipeline, technical deep-dive |
  117. | [Benchmarks](dev/docs/releases/v0.3.1/RELEASE_NOTES.md#benchmark-evidence) | LoCoMo evaluation, TagCascade results |
  118. | [Release Notes](dev/docs/releases/v0.3.3/RELEASE_NOTES.md) | Latest (v0.3.3): multimodal image input, dynamic capability + context detection, ChromaDB phantom-handling closing issue #8, atomic config writes, Harmony token cleanup |
  119. ---
  120. ## Important Notices
  121. **AI Safety**: LLMs may generate incorrect information. Always verify critical information. Don't rely on AI for medical, legal, or financial advice.
  122. **Model Licenses**: Downloaded models (Llama, Qwen, etc.) have their own licenses. Review before commercial use.
  123. ---
  124. ## Support
  125. - **Discord**: https://discord.gg/F87za86R3v
  126. - **Email**: roampal@protonmail.com
  127. - **GitHub**: https://github.com/roampal-ai/roampal/issues
  128. - **Author**: [Logan Teague](https://www.linkedin.com/in/logan-teague-6909901a5/)
  129. ---
  130. ## Pricing
  131. **Free & open-source** (Apache 2.0 License)
  132. - Build from source → completely free
  133. - Pre-built executable: **$19.99 one-time** (saves hours of setup)
  134. - Zero telemetry, full data ownership
  135. ---
  136. **Made with love for people who want AI that actually remembers**

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.