Ai Memory
Give any MCP agent durable SQLite-backed memory that loads only on recall so long projects do not burn tokens re-sending history every turn.
Overview
ai-memory is a MCP server for the Build phase that provides persistent SQLite memory for agents with tiered recall and minimal prompt tokens until data is fetched.
What is this MCP server?
- Docker OCI package ghcr.io/alphaonedev/ai-memory:0.5.1 with stdio mcp entrypoint
- Feature tiers via --tier: keyword, semantic, smart, autonomous
- SQLite persistence controlled by AI_MEMORY_DB path
- Positioned as zero token cost until explicit recall
- Version 0.5.1 with documentation at alphaonedev.github.io/ai-memory-mcp
- 4 documented --tier values: keyword, semantic, smart, autonomous
- 1 optional AI_MEMORY_DB environment variable for database path
Community signal: 24 GitHub stars.
What problem does it solve?
Multi-session agent work wastes tokens and loses continuity when every chat restarts without a dedicated memory store.
Who is it for?
Indie agent builders who want self-hosted, Docker-run memory with selectable keyword-to-autonomous tiers.
Skip if: Teams that need hosted multi-user memory with enterprise ACLs, or builders who only need ephemeral in-chat context.
What do I get? / Deliverables
Your agent stores and retrieves memories through MCP with tiered search while you keep a single SQLite database under your control.
- Tiered memory store shared across agent sessions
- Explicit recall tools instead of always-on context injection
- Local database file you can back up and migrate
Recommended MCP Servers
Journey fit
Persistent agent memory is foundational build-phase tooling for multi-session products, support bots, and personal coding copilots. Agent-tooling covers memory layers that sit beside the model rather than inside a single app feature.
How it compares
Persistent recall MCP layer, not a vector SaaS database product or a one-shot summarization skill.
Common Questions / FAQ
Who is ai-memory for?
It is for developers running custom agents via MCP who want durable, recall-on-demand memory without paying to resend full history each request.
When should I use ai-memory?
Use it while building agent-powered apps or long coding sessions where decisions, preferences, and facts must survive restarts.
How do I add ai-memory to my agent?
Run the ghcr.io/alphaonedev/ai-memory OCI image with mcp and your chosen --tier, set AI_MEMORY_DB, and register the container stdio server in your MCP client.