Amnesic
Add durable semantic memory on top of Postgres, MySQL, MSSQL, or SQLite so your coding agent recalls facts across sessions without a separate vector SaaS.
Overview
Amnesic is a MCP server for the Build phase that provides persistent semantic memory for AI agents using Postgres, MySQL, MSSQL, or SQLite.
What is this MCP server?
- Persistent semantic memory stored in SQL databases you already run
- Driver extras via uvx: amnesic[postgres], [mysql], [mssql], [sqlite], or [all]
- PyPI package amnesic v0.2.0 with stdio MCP transport and runtimeHint uvx
- Supports Postgres, MySQL, MSSQL, and SQLite from one server design
- Open-source implementation on GitHub SurajKGoyal/amnesic
- Version 0.2.0 on PyPI
- 4 SQL dialects: Postgres, MySQL, MSSQL, SQLite
- stdio transport with uvx runtimeHint
What problem does it solve?
Agents forget everything between chats unless you build custom tables, embeddings, and recall glue for every database you use.
Who is it for?
Solo builders who want self-hosted agent memory on SQL they control and already deploy for their app.
Skip if: No-SQL-only stacks, teams that want a managed hosted memory API with zero DB ops, or production without testing driver-specific setup.
What do I get? / Deliverables
After you run amnesic via uvx with the right SQL driver extra and stdio config, your agent can read and write semantic memory in your existing database.
- Cross-session semantic memory read/write tools backed by your SQL instance
- Multi-dialect support chosen via amnesic[postgres|mysql|mssql|sqlite|all] install flag
Recommended MCP Servers
Journey fit
Persistent agent memory is implemented while building the product backend and data layer, before you are tuning launch SEO or production error dashboards. SQL-backed memory is a backend concern—schemas, drivers, and retention—so the canonical shelf is build → backend.
How it compares
SQL-native semantic memory MCP, not a hosted Pinecone connector or ephemeral chat context skill.
Common Questions / FAQ
Who is amnesic for?
Developers building agent-powered apps who prefer Postgres, MySQL, MSSQL, or SQLite for durable memory instead of a separate vector-only cloud.
When should I use amnesic?
Use it during Build when wiring backend persistence and cross-session recall; pair it with your existing DB provisioning and migrations.
How do I add amnesic to my agent?
Configure MCP stdio with runtimeHint uvx, install amnesic from PyPI using --from amnesic[postgres] (or mysql, mssql, sqlite, all), and point connection settings at your database.