AGENT MEMORY · MCP
Experience that carries over to the next task.
ctxMem gives AI agents a shared memory for mistakes, tools, patterns and strategies. Experiences are stored as readable Markdown files and searched through a SQLite index, so lessons can be reused within and across projects.
Local Markdown memories · SQLite index · MCP server and web interface
FEATURES & USE CASES
Each memory is a Markdown file with metadata for type, project, tags and reliability. The files are understandable outside the service too.
Review why an agent recommends a particular pattern and correct the lesson if needed.
Find experiences with full-text or semantic search. Results include relevance scores and excerpts; embeddings can run locally or through a configured provider.
Find an earlier debugging strategy even when the new task uses different words.
MCP tools save, search, list and retrieve memories. Suggestions connect relevant strategies, mistakes and anti-patterns to the task.
Let an agent retrieve known pitfalls before it starts changing a project.
Collect mistakes that have not yet been analysed and turn them into strategies. Reflection can be performed by the connected agent or through a configured model API.
Turn recurring test problems into a reusable working routine.
Memories are organised by project or globally, with separate types for tools, patterns, mistakes and strategies.
Keep project-specific details local and share useful practices across codebases.
The web interface provides an overview of memories and projects, plus ZIP backup export and restore. API keys can protect access.
Move the memory archive to another computer and import missing memories.