Bare status
memory-hive now opens a one-screen receipt for install path, hive path, silos, recall state, stale learning count, and maintenance status.
How it works
One local folder. Two memory layers. A small CLI.
Memory Hive keeps the public workflow simple: install once, check status, add agents as needed, recall cited context, and update periodically.
~/.memory-hive/
├── memory-hive
├── update.sh
└── hive/
├── index.md
├── agents/
│ └── coder/
│ ├── log.md
│ ├── context.md
│ └── memory.md
├── knowledge/
├── learnings/
│ ├── raw/
│ └── distilled/
└── .hivecode/
v1.2.0 update
What changed in the latest release.
The GitHub repo now points first-time users at the day-one path instead of making them learn every internal command.
Update is the habit
memory-hive update refreshes tool files, preserves private silos, updates recall, and runs maintenance quietly.
Recall is direct
memory-hive recall "task context" retrieves cited local context without forcing users to build the index by hand first.
Two layers
Private silos stop context collisions. The hive compounds shared truth.
Each agent owns its own working notes. The shared hive is curator-governed, so every raw lesson does not automatically become truth.
| Private silo | One agent's log.md, context.md, and memory.md. |
|---|---|
| Shared hive | Curated project truth, durable patterns, decisions, and reusable lessons. |
| HyperRecall | A rebuildable SQLite/FTS index beside Markdown. Markdown stays source of truth. |
| Curator | The reserved main agent promotes verified raw learnings and resolves conflicts. |
Copy-paste path
Use these until you need advanced commands.
The full command catalog still exists for scripts, CI, power users, and curator work. Normal use should stay boring.
curl -fsSL https://hive.neural-forge.io/install.sh | shinstall oncememory-hivecheck current statememory-hive add coder --role coderadd another private silomemory-hive updaterefresh and maintainmemory-hive help --advancedopen the deeper catalog only when needed