Esme (mente en marcha)

Esme is a local cognitive memory server that lets any AI assistant recall relevant context, store facts, and give feedback so the memory learns.

What you can do

• Recall the most useful stored context for a query, with the ids of what was used, optionally per AI model
• Store a fact, decision or preference and get back its id
• Report whether recalled context was useful so the memory reweights what it returns
• Memory is associative and multi-hop, with adaptive context budgeting and abstention rather than plain text search

When to use it

Useful when you want assistants to remember what you have told them across sessions and clients without pasting markdown memory files, and when you want context selection that adapts per model and improves from feedback.

Requirements

The README describes a local install (pipx or uvx of the daimon-esme package, then daimon init and daimon serve-mcp) with a one-time model download of about 316 MB; the manifest instead lists a remote streamable HTTP endpoint at a configurable host with an Authorization header. No accounts or cloud service are described.