Memwyre is an AI memory platform that compiles work done in coding tools, terminal agents, and chat sessions into a persistent, shared memory layer, then recalls it before any connected tool acts. The console runs at app.memwyre.tech, with signup at memwyre.tech/signup.
What is Memwyre?
Memwyre is an AI memory platform that gives every AI tool, agent, and conversation a persistent, shared memory, positioning itself as "the memory that self-compiles." Input arrives from IDE sessions, terminal agents (Claude Code, Codex, Cursor, Grok Bot), documents, and conversations; the output is searchable, retrievable context handed back to those tools before they act. The web console organizes this into a Memories library with filters for All, Code & IDE, Conversations, and Documents, plus Spaces, Inbox, and Ask AI. A Slack demo shows @memwyre answering a multi-part engineering question in 42 ms.
Key Features
- Memories library — a browsable, searchable store of captured technical context showing each entry's name, modified timestamp, and file size; the demo lists 15 records ranging from a 1.8 KB Postgres pooling note to a 14.2 KB retrieval engineering spec.
- Native IDE memory over MCP — connecting an editor through MCP makes past bug fixes, configuration choices, and API schemas searchable inside Cursor, VS Code, or Windsurf.
- Code agent memory — terminal tools such as Claude Code and OpenClaw save debug results and project conventions back into memory, building cumulative operational memory.
- One layer across LLMs — a shared memory layer spanning Claude, ChatGPT, and Gemini, so a code pattern explained in ChatGPT is retrievable by a Cursor agent.
- Project memory spaces — scoped workspaces such as payments-service group memories by service and owner (for example, the architecture owner recorded on the space).
- Retrieval engineering — internal specs describe memory ingestion pipelines, HNSW index partitioning, and client SDK interfaces for IDE integrations.
- Tenant-scoped encryption — per-tenant KMS keys encrypt memory embeddings and entity nodes, with a 90-day automatic purge for inactive sessions.
Who is it for?
- Backend and platform engineers who want a previous Postgres pooling fix, PgBouncer connection ceiling, or Stripe webhook idempotency decision recalled automatically in the next session.
- Teams running multiple coding agents that need Claude Code, Codex, and Cursor to read from one project memory instead of re-explaining context.
- Engineering leads tracking decisions across microservice migrations, RFCs, and monorepo audits so consensus like a REST v1 to gRPC v2 cutover stays visible.
- On-call and ops engineers keeping runbooks and failover checklists retrievable at the moment an incident starts.
What can you do with Memwyre?
- Multi-tool developers: record a hybrid search or schema-validation decision in one assistant and have it surface later in a different editor without pasting context again.
- Agent operators: isolate background subagents by tenant and workspace namespace so autonomous tasks cannot overwrite production architectural prompt templates.
- Team knowledge capture: ask @memwyre in Slack what was decided on retries, billing RPC, PgBouncer pooling, and token rotation, and receive one consolidated answer.
How does Memwyre work?
Memwyre connects to the tools you already use rather than replacing them, so there is no separate chat window to work in. You sign up, add a connector (an IDE through MCP, or a terminal agent), and then keep working normally while memories are captured and indexed. The console then lets you search and inspect what was captured, filtered by Code & IDE, Conversations, or Documents, and connected tools retrieve matching memories before they act.
FAQ
Is Memwyre a chatbot?
No. Memwyre describes itself as a universal context layer that plugs into the models, editors, and terminal tools you already use, not a standalone chatbot you visit to hold conversations.
Which tools and models does Memwyre connect to?
It lists Claude, ChatGPT (OpenAI), Google Gemini, Perplexity, Cursor, Windsurf, and Notion, plus terminal agents such as Claude Code, Codex, and Grok Bot, and IDE connectivity through MCP.
How is stored memory secured?
Memwyre states that memory embeddings and entity nodes are encrypted with per-tenant KMS keys, and that a 90-day automatic purge applies to inactive sessions.
How fast is memory retrieval?
An internal specification on the site targets sub-50 ms predicate-filtered memory retrieval through memory ingestion pipelines and HNSW index partitioning.
