Roampal

How is this different from ChatGPT memory, CLAUDE.md, or agent-managed markdown?

ChatGPT Memory stores content on OpenAI's servers and only works inside ChatGPT clients — it's not available via the OpenAI API, so it does nothing for you in Claude Code, Cursor, or OpenCode. CLAUDE.md is a static Markdown file you write and commit by hand; its full contents get injected every session, it doesn't update itself, and it doesn't track which parts actually helped. Agent-managed markdown memory (OpenClaw, Hermes Agent) uses MD files too, but the agent reads and writes them as part of its own workflow — memory maintenance becomes part of the agent's job, spending tokens every turn.

What models does it support?

roampal-core works with whatever model your AI tool already uses (Claude, GPT, etc.) - memory and scoring are model-agnostic. Roampal Desktop supports local models via Ollama or LM Studio including Llama, Qwen, and Mixtral.

Does it work offline?

roampal-core stores and retrieves memory fully locally. Your AI tool (Claude Code, OpenCode) still connects to its own API provider as usual. Roampal Desktop works fully offline after downloading models.

What are the system requirements?

roampal-core requires Python 3.10+ and minimal disk space - runs on anything that runs Python. Roampal Desktop requires Windows 10/11 (64-bit) with 8GB RAM minimum (16GB recommended).

How do I export my data?

roampal-core stores all data locally in your data directory (e.g. %APPDATA%/Roampal/data on Windows) - copy the folder to backup. Roampal Desktop has a Settings → Export option to download a zip file.

Why charge if it's open source?

roampal-core is 100% free - install with pip install roampal and use it forever. Roampal Desktop ($19.99 one-time) gives you a pre-packaged executable with bundled dependencies, tested build, and lifetime updates. You can also support Core development with an optional subscription.

What are the MCP tools and what do they do?

roampal-core (Claude Code and OpenCode) provides 6 MCP tools: search_memory, add_to_memory_bank, update_memory, delete_memory, score_memories, and record_response. Roampal Desktop provides 7 MCP tools: get_context_insights, search_memory, add_to_memory_bank, update_memory, archive_memory, record_response, and score_memories. Desktop includes get_context_insights for manual context retrieval and archive_memory for soft-deletes, while Core automates context injection via hooks and uses delete_memory for hard removal. In Core, score_memories is enforced automatically by hooks; in Desktop, the too

How does Roampal know when something worked?

Just talk naturally. The AI reads your response, determines if you were satisfied, and scores the memory in Roampal. For example: if you say thanks, that fixed it - the memory gets scored higher. If you correct the AI - that memory gets scored lower. No buttons to click, no ratings to give. It learns from the conversation itself.