memini
Active·★ 28·AGPL-3.0·Updated 2026-10-19
memini is a shared, persistent memory service designed for AI agents, providing a single location for agents to remember and recall information with continuously improving retrieval quality. It's built as a Go binary, offers zero-configuration startup, and scales from embedded SQLite on a laptop to Postgres in Kubernetes.
#AI Agents#Persistent Memory#Vector Search#Hybrid Retrieval#Go#SQLite#Postgres#Model Context Protocol
01
Features
01Shared, Persistent Memory: Provides a dedicated service for AI agents to store and retrieve memories, designed for long-term persistence.
02Hybrid Retrieval & Re-ranking: Employs a hybrid (vector + keyword) retrieval method, fused with Reciprocal Rank Fusion (RRF), and re-ranks results by relevance, recency, and importance, not just similarity.
03Tiered Memory System: Implements a tiered memory structure (working, episodic, semantic, procedural) with decay mechanisms.
04Pluggable Storage: Supports flexible storage options, including embedded SQLite-vec for local use and Postgres with VectorChord for scalable deployments.
05Optional LLM Consolidation: Features an opt-in LLM consolidation pipeline for background deduplication and contradiction resolution, but can run headless (without an LLM) using marker heuristics.
06Model Context Protocol (MCP) Support: Compatible with MCP-capable agents, offering both Streamable HTTP and stdio transports.
02
Compatibility
Go
Runtime
Verified via docs
SQLite-vec
Embedded Storage
Verified via docs
Postgres
Scalable Storage
Verified via docs
MCP (Model Context Protocol)
Protocol
Verified via docs
Docker
Deployment
Verified via docs
Kubernetes
Deployment
Verified via docs
03
Quick start
1
$ export MEMINI_EMBED_BASE_URL=http://localhost:8081/v1
2
$ export MEMINI_EMBED_MODEL=bge-m3
3
$ export MEMINI_EMBED_DIMS=1024
4
$ mise run run
04
Use cases
↳Enhancing AI Agents with Long-Term Memory: Integrate memini to give AI agents (like Claude Code, opencode, Hermes) the ability to remember past interactions and learn over time.
↳Self-Hosting a Shared Memory Service: Deploy memini for a team or homelab environment to provide a centralized, persistent memory store for multiple AI agents.
↳Improving AI Recall Quality: Tune memini's retrieval parameters to fix bad recall and achieve higher accuracy in retrieving relevant information for AI models.
↳Managing Multi-Agent Namespaces: Lay out distinct namespaces within memini to provide isolation and organized memory management for several different AI agents.
↳Backend for Open WebUI Memory: Utilize memini as a robust memory backend for platforms like Open WebUI, integrating its memory and tool capabilities.
05
Alternatives
okf-agent-memory★ 827
Git-native persistent memory for AI coding agents. Implements Google OKF v0.2 with sub-300µs in-memory BM25 search, embedded MCP server, and progressive disclosure. Slashes token bloat by 80% with zero external databases or dependencies. Built in pure Go.
mnemosyne★ 151
Evidence-backed memory for AI agents: local-first recall, temporal history, tested skills, and gradual migration.
mex★ 1.8k
Persistent project memory for AI coding agents. Structured scaffold + drift detection CLI.
montycat-mcp★ 28
Shared, persistent memory for AI agents. Self-hosted MCP server with semantic search, vector RAG, and live updates. Works with Claude, Cursor, Codex, and any MCP client.
caura★ 504
Caura (formerly MemClaw) — governed shared memory for AI agent fleets. Multi-agent, multi-tenant, MCP-native. Trust tiers, keystone policies, audit trails, knowledge graph, self-improving retrieval. Apache 2.0.
Comments
Log in to leave a comment
No comments yet. Be the first!