Memex
Give your agent durable memory across sessions with semantic plus recency search so long-running solo projects do not reset context every chat.
Overview
Memex is a MCP server for the Build phase that provides persistent semantic and recency-based memory retrieval for AI agents via self-hosted Docker.
What is this MCP server?
- Persistent memory store for AI agents with semantic and recency-ranked search
- ONNX embeddings for local-friendly vector recall without mandatory cloud vector DB
- Docker Compose deployment with streamable HTTP MCP at localhost:8001/mcp/
- OCI image ghcr.io/ayushagrawal288/memex:v0.1.0
- Early v0.1.0 registry release for self-hosted agent memory
- Version 0.1.0 with OCI identifier ghcr.io/ayushagrawal288/memex:v0.1.0
- MCP endpoint http://localhost:8001/mcp/ (streamable-http)
- Features semantic search, recency search, and ONNX embeddings per server description
What problem does it solve?
Long solo-builder agent sessions lose thread context, so the model repeats questions and forgets decisions you already documented last week.
Who is it for?
Indie developers self-hosting agents who want ONNX-powered local memory without bolting on a separate commercial memory SaaS.
Skip if: Builders who need enterprise compliance-ready hosted memory or who only work in single-shot chats with no continuity.
What do I get? / Deliverables
After you run Memex and register the HTTP MCP endpoint, your agent can write and query durable memories with semantic plus recency search across sessions.
- Cross-session memory storage queryable by the agent
- Semantic plus recency-ranked retrieval over ONNX embeddings
- Self-hosted HTTP MCP memory layer for agent workflows
Recommended MCP Servers
Journey fit
Build is where agent tooling matters most: memory MCP lets you wire persistence before you scale features, support, and ops workflows on the same agent stack. Agent-tooling is canonical because Memex is infrastructure for recall and retrieval—not a one-off integration—using ONNX embeddings and Docker Compose.
How it compares
Self-hosted agent memory MCP, not an in-chat scratchpad skill.
Common Questions / FAQ
Who is Memex for?
Solo builders and agent hackers who run Claude Code or similar clients and need cross-session recall with semantic search on infrastructure they control.
When should I use Memex?
Use it during Build when your project spans many agent sessions and you want stored facts, embeddings search, and recency ranking instead of re-pasting context.
How do I add Memex to my agent?
Deploy with Docker Compose using ghcr.io/ayushagrawal288/Memex:v0.1.0, expose http://localhost:8001/mcp/ as streamable HTTP MCP in your agent config, and restart the client to enable memory tools.