Why not built-in memory?
Claude Code and Codex remember things on their own. Use that first. It needs no setup, and Claude Code loads its memory into every session, while rcllr answers only when the agent asks. If you work on one project on one machine, built-in memory is enough.
| Built-in | rcllr | |
|---|---|---|
| Claude Code auto memory | One folder per Git repository in ~/.claude/projects/, on this machine only |
One brain for every project, on every machine with a synced brain |
| Codex memories | Off by default, kept in ~/.codex/memories/ |
The same brain for Codex and Claude Code |
CLAUDE.md, AGENTS.md |
Rules loaded into every session | Notes with a key fact, aliases and a search that cites the file |
| Notion memory (MCP) | Hosted by Notion, needs a Notion account | Markdown in your own Git repository, local or synced |
| Private and work | Separated by project at most, no hidden notes | Scopes, and sensitive notes agents open only when you name the topic |
Sources: Claude Code memory, Codex memories, as of October 2026.