Nowledge Mem

Nowledge Mem Cloud

Use the connected Nowledge Mem tools when the answer depends on prior decisions, current working context, exact earlier conversations, or durable knowledge that should survive this chat.

Connect

If the Nowledge Mem tools are unavailable, direct the user to the official setup guide at https://mem.nowledge.co/docs/integrations/chatgpt-web. The MCP endpoint must be public HTTPS and end in /mcp. Complete the host's OAuth flow; never ask the user to paste a Nowledge API key into chat.

This public plugin package is skills-only until its production MCP connection is registered with OpenAI. Do not invent an app identifier or imply that installing the skill alone creates a Cloud connection.

Use

1. Start with read_context_bundle when broad current context is useful.
2. Use memory_search for focused recall and get_memory_by_id for exact follow-up.
3. Use thread_search and thread message tools only when exact prior conversation evidence matters.
4. Use Library tools for source-backed material.
5. Before memory_add or memory_update, make the intended Space and durable claim clear. A write may require separately approved mem:write scope.

Respect the tools actually exposed by the connected server. Do not claim access to hidden administrative, identity, scheduler, deletion, or review/governance tools.

Boundaries

• Remote MCP lets ChatGPT or Codex call Mem. It does not let Mem read the host's private transcript.
• Use the browser extension or an official export for ChatGPT conversation capture. Use the dedicated local Codex connector for Codex lifecycle hooks and transcript capture.
• A client name and source_app are provenance, not an AI Identity or Space selector.
• Never create a new Space merely to make a connection succeed.

When a task materially depends on remembered context, cite or summarize the retrieved evidence honestly. Do not describe a retrieval as proof that the memory helped unless the user or downstream work confirms it.