Smara Memory API

Smara is a persistent memory service for AI tools. It is meant to address the problem of sessions starting from zero and AI assistants forgetting earlier context, with shared memory that persists across tools such as Claude Code, Cursor, Codex, Windsurf, and VS Code.

The service stores context automatically as work proceeds and loads that context in later sessions. It describes a shared pool of memories that follows the user across tools, with memories ranked by relevance and freshness and a decay model that lets older or less relevant items fade over time. The page also shows that it can capture facts such as technology choices, deployment targets, and API details, and that it supports visibility settings including private and team memories.

Smara is delivered as a REST API and an MCP server. The page says it works with any MCP client, and it gives an installation command using npx @smara/mcp-server --init, alongside an example of adding it to an MCP configuration. It also says there are three API calls for storing, searching, and retrieving user context, and that it works from any language without SDKs or schemas. A hosted API is available, and the service can also be self-hosted with Docker.

It is open source under the MIT License. The page offers a free API key and says no credit card is required. It also states that Smara has 92.2% accuracy on the LoCoMo benchmark, with 1,420 of 1,540 questions correct.