Local Shared Memory for Ollama Agents

A truly local, queryable shared memory store for multiple AI agents running on your machine

Problem

When using multiple agents like Claude, Codex, and Cursor simultaneously, each one's context is isolated, forcing repeated re-entry of project background and decisions. While generic agent databases exist, many developers want data to stay fully local, independent of third-party services, and to work seamlessly with local models like Ollama.

Solution

A fully local shared memory service optimized for Ollama and local models: it provides an MCP interface so different agents can read and write the same vectorized context, supports project/topic namespaces and natural-language queries, and uses local embedding models for retrieval. Nothing leaves the machine, and it works out of the box.

Users

Privacy-conscious local AI users, developers using Ollama, self-hosting workflow enthusiasts

MVP

One-command local service startup; built-in vectorization (can use local embedding models); MCP server for major agents; project-isolated namespaces; natural-language query; works with Ollama with zero config

Revenue

Open-source core + paid Pro ($5/mo) for multi-device sync, larger storage, and advanced retrieval; enterprise self-hosted support

Why now

Local models and self-hosted AI workflows are rapidly growing, yet there's no local-first lightweight option for shared agent memory; MCP maturity lets indie developers plug into existing agent ecosystems quickly