Claude memory, agent skills, and persistent memory tools

Posts compare ChatGPT and Claude memory designs and report related tool updates. - ChatGPT memory approach: full injection with lightweight recent-dialogue summaries. - ChatGPT four-layer memory: session metadata, user memory, recent-dialogue summaries, current session window. - ChatGPT precomputes and injects recent summaries into each prompt. - Claude memory approach: on-demand retrieval via tools. - Claude tools: conversation_search and recent_chats for selective retrieval. - Claude injects user memories in XML into system prompt. - Tradeoff: ChatGPT offers seamless continuity; Claude offers precise retrieval of exact past content. - ChatGPT added support for agent skills mechanism similar to Claude's design. - ChatGPT thinking model can package files and includes built-in skills: excel, word, pdf. - Claude-Mem open-source tool provides persistent compressed memory for Claude Code workflows. - Claude-Mem stores compressed summaries locally using SQLite and vector DB (Chroma), saving tokens. - Claude-Mem features: hooks into Claude Code lifecycle, background workers, automatic retrieval injection and a local web UI at localhost:37777.
Key insights
Two distinct memory paradigms: injection vs on-demand retrieval
ChatGPT uses an injection-first memory design, precomputing recent-dialogue summaries and injecting user memory and metadata into each prompt for seamless continuity. Claude uses an on-demand retrieval paradigm where the model invokes tools like conversation_search or recent_chats only when it determines past conversation context is needed. The posts compare advantages: ChatGPT prioritizes flow and speed while Claude prioritizes retrieval precision.
ChatGPT's four-layer memory structure clarified
ChatGPT's memory architecture was described as four layers: session metadata (temporary), user memory (long-term facts with edit tools), recent-dialogue summaries (lightweight summaries of past 10–15 user turns) and the current session window. The system injects these layers into prompts each conversation.
Claude-Mem open-source provides persistent compressed memory for Claude Code
Claude-Mem is an open-source persistent memory compression system for Claude Code that captures tool-call history, compresses semantic summaries, stores entries locally (SQLite + vector DB), and injects relevant history into new sessions. It hooks Claude Code lifecycle events, runs background workers for summary compression, provides a local web UI and claims token and cost savings.
ChatGPT adopts agent skills pattern and ships built-in skills
ChatGPT reportedly added support for an agent skills mechanism similar to Claude's approach. A test prompt using the thinking model shows built-in skills for excel, word and pdf and the ability to package files from a filesystem path, indicating modular, off-context callable skill components.