Persistent memory for developers and their AI agents.
Memwyre eliminates agent amnesia. We build the shared context layer that keeps your architecture decisions, project conventions, and debugging history accessible across every session and IDE.
The repetitive context problem.
Anyone who writes code alongside Cursor, Claude Code, Windsurf, or ChatGPT experiences the exact same friction: you spend 20 minutes explaining your database schemas, internal routing patterns, and previous debugging iterations. You make great progress, close the terminal or editor, and the very next morning your AI has complete amnesia.
Standard flat vector databases and naive RAG don't solve this. They dump thousands of raw tokens into your context window, bloat billing costs, and frequently dredge up obsolete patterns from refactors that happened three months ago.
Memwyre was founded in 2026 to solve this at the protocol layer. We treat memory as a living substrate: one that captures atomic facts, decays outdated decisions naturally, and connects natively to your developer environment through the Model Context Protocol (MCP).
Founding Team & Engineering
Memwyre is built and led by Himansh Shivhare, operating from Bengaluru, India with an open-source community worldwide.
Himansh is a software engineer specializing in distributed backend systems, vector retrieval architectures, and context engineering. He founded Memwyre after experiencing firsthand the daily friction of working with AI coding tools that lose context every time a session terminates.
Himansh designed and built Memwyre's core architecture from the ground up: the decoupled memory protocol, the Model Context Protocol (MCP) server for Cursor and Claude Code, and the temporal decay engine that prevents context windows from clogging with outdated information.
As founder and lead engineer, Himansh oversees all systems engineering, open-source development, and product strategy. He builds in the open, reviews community pull requests, and personally works with developers integrating Memwyre into their engineering stacks.
Core Engineering Principles
The design decisions that shape everything we build at Memwyre.
Model Agnostic
Your context belongs to you, not your model provider. When you switch between Claude, OpenAI, or Gemini, your accumulated knowledge, conventions, and project history remain fully intact.
Open by Design
Memory should be inspectable. Our core memory protocols and local MCP connector are open-source under the Apache 2.0 license, allowing you to audit, extend, and self-host.
Natural Context Decay
Hoarding every token leads to bloated prompts and confused agents. Memwyre applies intelligent recency decay so superseded decisions fade while foundational architecture rules persist.
Zero Training on Your Data
We never use your private code, memories, or workspace interactions to train foundation models. All data is encrypted at rest and in transit, with full self-hosting options for air-gapped setups.
Key Information
Essential operational details for developers, partner programs, and verification.
Give your AI tools persistent memory.
Start free with 50 memories, connect your IDE via our open MCP server, or write directly to Himansh with any questions.
