-
Memoir v0.2.1: search, a terminal UI, and a sturdier store
Last month's vector-search preview is now shipping commands. v0.2.1 adds file and folder ingestion with semantic search, a read-only terminal UI, a branch switcher in the web UI, a new default storage backend, and per-call branch routing for the CLI.
Read essay -
Two trees, one commit: vector search lands in Memoir
A feature coming to Memoir: vector recall over the same versioned store. The path tells you where a memory lives; the vector tells you what it says — so Memoir can hold long-form content without losing the ability to find the sentence inside it. Includes an interactive deck.
Read essay -
Keeping the logs, losing the lesson
A popular fix for agent amnesia is to bump cleanupPeriodDays to 3650 — ten years of session JSONL on disk. It buys a bigger archive, not a better memory. Here's why retention and memory are different problems, and what actually carries forward.
Read essay -
Memoir's storage substrate: prollytree
A Merkle search tree — what prollytree is, and what it enables for memoir: key-level merge, subtree proofs, dedup-friendly federation, and vector search that lives inside the data.
Read essay -
Two lessons from one apology
A human developer caught the agent running release builds against a debug .so. The agent investigated, apologized, and explained the JAR-vs-loose-classes pitfall. Memoir auto-classified both lessons into the right taxonomy paths — no future session will repeat the mistake.
Read essay -
Start and stop the Memoir UI from Claude Code
No flags, no ports to remember. Just say "start memoir ui" and "stop memoir ui" — Claude Code launches and tears down the browser-based memory inspector for you.
Read essay -
When two branches remember differently
Why Memoir's merge story matters, what we have today, and what we know we still have to build.
Read essay -
The 5-second answer that was wrong
An auto-memory rule turned a careful 103-second answer into a confident 5-second one — with the opposite recommendation. The speedup came from skipping the verification that mattered.
Read essay -
Stop guessing what your agent cost
Memoir tracks turns, tool calls, errors, and latency per git branch — automatically, with zero LLM in the loop. Here's what changes when you can finally read that.
Read essay -
Why every coding session should start with /memoir:onboard
Every Claude Code session starts from zero. /memoir:onboard makes the first twenty minutes obsolete — a curated codebase snapshot, incrementally refreshed from commit history.
Read essay -
Your second memory file: Claude Code's notebook meets Memoir's repo
Claude Code's auto-memory is the agent's private notebook. Memoir is the project's branchable, queryable archive. Here's how to use both — and why parallel work breaks the first one.
Read essay -
Your coding agent has amnesia, and you've been the unpaid memory layer
Memory is becoming the codebase for AI agents. Here is what version control for it looks like.
Read essay