Give your AI a memory that learns: the Hindsight setup for Claude Code, Codex and your own app, my 3 real tests, and the 2 mistakes to avoid

The companion to the video. Hindsight is a free, open-source (MIT) memory for AI agents from Vectorize. I ran it myself on September 25, 2026 and tested it three ways. PAGE 1 — what it is in one breath: retain, recall, reflect, the four memory types, and the benchmark with its caveat. PAGE 2 — the three ways to set it up, commands word for word: one command for your coding agents, the Docker server plus three calls for your own app, and the two-line LLM wrapper, plus which models work without an API key. PAGE 3 — my three tests with the real output: the bakery email (42 lines with blanks vs 5 lines to the right supplier), reflect drawing its own conclusion, and Claude Code telling Codex a rule it never heard. PAGE 4 — the two mistakes that cost me an hour, the honest notes, and who it is and isn't for. Independent guide from Hyperautomation Labs — not affiliated with Vectorize.

By submitting, you'll be subscribed to the Hyperautomation AI Report newsletter — three practical AI emails a week. Unsubscribe anytime.

Free. No spam. Unsubscribe anytime.