Hey PH 👋 I built Elephant because I got tired of starting every Claude Code session with the same 200-token preamble: "we use Bun not Node, auth moved to packages/core, don't touch the stripe webhook it's mid-migration..." Elephant is three files doing boring work well: • ELEPHANT.md — repo-local, committed, team-shared memory. Every entry tagged with author. • ~/.claude/elephant/memory.md — personal cross-repo memory. • CHANGELOG.md — auto-generated from git log with semver-aware bumps. A SessionStart hook loads memory automatically. A Stop hook nudges Claude to save important decisions when topics shift. Caveman-compressed entries keep a week of history at ~150 tokens. Bonus: /elephant takeover reads your last 60 commits and seeds memory on a fresh clone — new teammates get briefed before their first session. MIT. No cloud. Works today on Claude Code. Would love feedback! What's the most painful context you lose between AI coding sessions?

Elephant

Persistent memory + auto changelogs for Claude Code

Persistent memory + auto changelogs for Claude Code