My AI bots and Claude Code read and write the same memory bank. What one learns, the others know next time, so I stopped explaining the same setup, decisions and fixes over and over.
The setup
Hindsight: open-source agent memory from Vectorize, running on my Mac. https://github.com/vectorize-io/hindsight
One bank: every bot and Claude Code point at the same one.
Tags: each memory records which bot wrote it, so I can see where a fact came from.
How things get in
My bots (Hermes Agent): a memory plugin saves each conversation as it happens. Each bot needs its own small config pointing at the shared bank.
Claude Code: a plugin saves each session when it ends.
My notes: a script syncs my project notes into the bank every hour, and strips secrets first.
How I tested it
I asked a bot about something only Claude Code knew. It answered.
I stored a code word through a bot, then asked Claude for it. It came back.
Test it both ways. One direction working doesn't prove the other.
Three things that broke it quietly
One stray settings line pointed the memory at the wrong model, and hundreds of saves failed.
Missing per-bot configs: those bots had no shared memory at all, and said so only in a log.
A save timeout: saves from the bots timed out until I patched it.
If recall looks empty, check the bank's stats for pending and failed operations first.