Memoryport

Memoryport is permanent-memory software for AI conversations, built for persistent context across sessions with any LLM. It is local-first and open source, and its core promise is that each session can continue where the last one left off.

Installation is described as a one-command setup for macOS and Linux, followed by an interactive wizard for configuring integrations. Once in use, conversations are indexed locally and context is injected only when relevant. The feature set includes a local LanceDB vector index with zero cloud dependency, optional Arweave storage for permanent backup, cross-device sync, and rebuilds from the chain. It also uses AES-256-GCM encryption with Argon2id key derivation.

The tool includes what it calls intelligent retrieval, using a three-gate system with query expansion so the AI receives relevant context rather than everything at once. It also supports multi-tool memory across Claude Code, Cursor, Ollama, and OpenAI, and lists integrations for Claude Code MCP Server, Cursor MCP Server, Open WebUI, an Ollama proxy, an Ollama native API proxy, a ChatGPT OpenAI proxy, and any MCP client through a standard protocol. The page also reports benchmark results, including 97.9% session recall on LongMemEval, 500 million tokens tested, under 100 ms p95 retrieval, and 0.82 recall@10.

Memoryport is licensed under Apache 2.0 and is presented with a full Rust codebase that can be read, self-hosted, or contributed to. The pricing section lists a free plan with unlimited local storage, Claude Code integration via MCP, Cursor integration, Ollama and Open WebUI support, knowledge graph visualization, session analytics, and encrypted-at-rest storage. A Pro plan is also offered at $9.99 per month and adds permanent memory backup, memories available forever after cancellation, about 250 million tokens per month, cross-device sync, and cold start rebuilds from the chain.