Development Windows
MemorySQL
Keep AI coding conversations searchable on your PC and pass project context between supported agents with MemorySQL, a free, open-source desktop knowledge base.
Version 0.7.1 · Direct publisher file
Link checked · Download guidance
What is MemorySQL?
MemorySQL collects conversations from supported AI coding assistants into a local SQLite database. It is designed for developers who switch between tools such as Codex, Claude Code, Cursor and OpenCode and want to recover earlier decisions without manually assembling a new handoff. Its desktop interface groups sessions by project and agent, while full-text search helps locate previous messages.
A local Model Context Protocol (MCP) server lets connected agents request memories, project state, recent sessions and handoff context. Optional semantic search uses local inference after its runtime and model have been downloaded. Project summaries can use local rules or a configured LLM provider; rolling project summaries are off by default.
Local storage is not encrypted storage. The publisher explicitly says that conversations and memories are saved in plaintext and that the loopback MCP endpoint has no authentication. Processes on the same computer can read the exposed context. Configuring an external LLM provider can send conversation excerpts to that provider, so review these settings before capturing sensitive work.
This entry links to the publisher’s Windows 0.7.1 installer. Softonic’s observed listing still identifies 0.5.6; the version here follows the publisher release rather than that older listing. The screenshot is the publisher’s session-library view, documented as refreshed for version 0.6.0, not an independently captured 0.7.1 installation.
What it offers
- Automatically capture conversations from supported AI coding agents, including Codex, Claude Code, Cursor, OpenCode and DeepSeek Harness.
- Browse sessions grouped by project and agent, and search locally stored conversation text.
- Expose memories, project context and session history to connected agents through a local MCP server.
- Enable local semantic search with an on-demand inference runtime and model.
- Create project handoff briefs and optional rolling summaries using local rules or a configured LLM provider.
Good to know
- Conversation history, memories and project state are stored in plaintext on the computer.
- The loopback MCP endpoint has no authentication; local processes can access the context it exposes.
- External LLM features can send conversation excerpts to the configured provider; pattern-based redaction does not guarantee removal of all sensitive information.
- Semantic search requires an initial runtime and model download before local inference is available.
Product images

Sources
References used for this listing. Retrieval dates do not indicate hands-on testing.
- MemorySQL — publisher README ↗Retrieved
- Release MemorySQL v0.7.1 — Logic647/MemorySQL ↗Retrieved
- MemorySQL v0.7.1 — publisher release assets ↗Retrieved
- MemorySQL — Privacy & Security disclosure ↗Retrieved
- MemorySQL — MIT License ↗Retrieved
- MemorySQL — publisher screenshot inventory ↗Retrieved
- MemorySQL - Download ↗Retrieved