OpenMemory

OpenMemory is a local memory layer for LLM conversations, with a GUI-first workflow.

It lets you keep long-term context in a file store, browse commits/history, and move sessions across models from one interface.

---

Setup

1. Prerequisites

• Python 3.11+
• pip
• (Optional) gemini-cli for in-GUI Gemini chat
• (Optional) ollama for local models in GUI
• (Optional) Claude Desktop (launch from GUI link)

2. Install

git clone <your-repo-url>
cd openmem
python -m venv .venv
source .venv/bin/activate
pip install -e .

3. Configure .env

Create .env in project root:

MEMORY_STORE_PATH=/home/<you>/path/to/llm-memory
GEMINI_API_KEY=your_gemini_api_key

Notes:

• MEMORY_STORE_PATH is where OpenMemory stores sessions/commits.
• GEMINI_API_KEY is required when using gemini-cli from GUI.

4. Start GUI

memory-gui

Open in browser:

http://127.0.0.1:8765

---

First-Time GUI Flow

1. In Settings, choose or browse your memory store path.
2. Create a new session (for example: French Revolution Chat).
3. Link a model to that session:

• gemini-cli (pinned to gemini-2.5-flash)
• ollama:<model> (auto-discovered from ollama list)
• claude-desktop (GUI opens Claude Desktop)

4. Chat in GUI (Gemini/Ollama), or in Claude Desktop if Claude is linked.
5. Commit staged context (auto commit metadata is generated by default).
6. View timeline in History and per-commit data in Commit Details.

---

What the GUI Includes

• Memory store path management
• Session creation/switching
• One-model-per-session linking
• Chat panel + staged context flow
• Auto-generated commit message/summary
• Commit history and detail browsing
• Tag/retrieval/search support through backend memory engine

---

Troubleshooting (GUI)

Browse button does nothing on WSL

• Restart memory-gui.
• Ensure WSL can call Windows tools (powershell.exe).
• If needed, set MEMORY_GUI_PICKER_CMD manually.

Gemini linked but chat fails

• Confirm GEMINI_API_KEY is present in .env.
• Restart memory-gui after editing .env.
• Confirm gemini command works in your shell.

No Ollama models in dropdown

• Ensure ollama is installed and running.
• Run ollama list in terminal and verify models exist.