See it in action
Real CLI. Real output. Runs every night automatically.
▶ Auto-plays · Click to pause · Loops automatically
Stop re-explaining context. Stop losing decisions. Stop watching tokens burn on data your agent already saw last week.
*Demo-skill benchmark. Results vary by workflow.
And you've been quietly paying for it in tokens, time, and repeated work.
You explain the same context again. The agent burns tokens re-learning what it knew yesterday. Nothing carries over.
You grep through memory files manually. Or re-ask the agent. Or just accept that the decision is lost somewhere in a compacted session.
Raw API responses flood context with 5-50KB of data you needed 2% of. The context window fills up. Compaction destroys the rest.
Real CLI. Real output. Runs every night automatically.
▶ Auto-plays · Click to pause · Loops automatically
Not another vector database. Not another cloud service. Just fast, local, honest memory.
Every search result carries a depth label: D0 (minutes ago) through D4 (months ago). Your agent can judge whether to trust it or re-fetch.
Session logs, decisions, research notes — all indexed. Ask "what did we decide about X?" and get an answer with a timestamp, not silence.
At midnight: reflect on yesterday's sessions, update long-term memory, index new files, push to Obsidian, update mind maps. All automatic.
FTS5 is SQLite's built-in search. No embeddings model. No GPU. No external process. Nightly cost on 96 unchanged files: literally milliseconds.
One SQLite file. Your memory never touches a cloud server. No API keys for storage. No vendor. Full control over your agent's history.
gm sync, gm search, gm status — the entire memory system through a single CLI. No juggling 5 separate scripts.
"I asked my agent about a product decision from two weeks ago.
It found it in under a second. With a timestamp."
Private beta is open. Free while it lasts. Request access and I'll walk you through setup.
📧 I'll reach out within 24–48h · Zero spam, ever