himanshu748/mnemo

Mnemo is a Slack app built to give a chatbot actual long term memory instead of forgetting everything between conversations. It was built for a Slack hackathon focused on the platform's AI Assistant features. The core idea is that Mnemo treats memory like a limited budget rather than an ever growing pile: it recalls only the small set of facts that matter for a given question, tracks how many tokens of context that recall costs, and periodically runs a background pass that merges related memories into more durable facts while letting unimportant details fade away over time. Inside Slack, people can talk to Mnemo directly through the AI Assistant pane, save a fact explicitly with a slash command, check what it would recall about a topic, trigger the memory cleanup pass manually, or save any message to memory from a message's menu. Mnemo can also be added to a channel, where its memory is shared across everyone in that channel rather than kept private, unlike direct messages, which stay personal to each user. One notable feature is how Mnemo handles memories that contradict each other, such as two people stating different meeting times. Rather than quietly averaging or picking one, it flags the conflict to the channel and asks the team to confirm which fact is correct. The same memory store can also be reached outside Slack. Mnemo runs an MCP server, a standard way for AI coding tools like Claude Desktop or Claude Code to connect to external data, so a user's personal memory can be reused inside their coding assistant after generating a one time access token inside Slack. Under the hood, the memory engine is written in pure Python with no dependency on Slack itself, while a thin Slack app layer and the MCP server sit on top of it. It uses a pluggable language model, defaulting to Google's free Gemini tier, and includes an offline fallback mode so the whole system can run and be tested without any API key at all.