memoryblock — local AI workspaces
Installation
# Install globally with npm
npm install -g memoryblock
# Or use with bun
bun install -g memoryblock
Run on a $5 VPS as comfortably as a MacBook Pro. Single binary, no Docker, no Python runtime, no cloud account.
What memoryblock gives you
- Blocks — isolated workspaces with their own memory, identity, and tools.
- Any model — Ollama, AWS Bedrock, OpenAI, Anthropic, Google Gemini, plus 100+ providers via the bundled
aiplugruntime. No per-provider code to maintain. - Multi-agent — your block can spawn short-lived sub-agents in parallel for fan-out work.
- Channels — chat from the CLI, the local web dashboard, or Telegram.
- Stable API — a single
POST /v1/chatendpoint for external apps and tools. - No lock-in — your data is plain Markdown + JSON in
~/.memoryblock. Copy a folder, move a monitor to another machine.
Quick start
1. Initialise
mblk init
The interactive setup wizard walks you through:
- LLM provider — pick from any registered aiplug provider (Bedrock, OpenAI, Anthropic, Gemini, Ollama, plus 100+ more).
- Channel — CLI, local web dashboard, or Telegram.
- First block — creates your initial AI workspace with sensible defaults.
All credentials are stored in auth.json at the workspace level. Move the folder, your auth travels with it.
2. Start your monitor
# CLI mode (interactive terminal)
mblk start home
# Telegram mode
mblk start home --channel telegram
# Web dashboard as a background daemon
mblk start home --channel web -d
The web dashboard binds to 127.0.0.1 by default — it’s a local control panel, not a public service. The server prints its bearer token on first start; copy it into the dashboard to log in.
3. Check status and stop
mblk status # list blocks + their state
mblk stop home # stop a specific block
mblk stop # stop all blocks
Directory structure
~/.memoryblock/ws/ ← your workspace (portable, plain text)
├── config.json ← workspace defaults
├── auth.json ← credentials (API keys) — gitignored
├── founder.md ← your profile (shared across blocks)
└── blocks/
├── home/
│ ├── config.json ← block settings, adapter, permissions
│ ├── monitor.md ← monitor identity and personality
│ ├── memory.md ← persistent context
│ ├── session.json ← crash-recovery state
│ ├── pulse.json ← autonomic background tasks
│ ├── agents/ ← ephemeral sub-agent workspaces
│ └── logs/ ← conversation history
└── work/
└── ...
Next steps
- Architecture — blocks, monitors, the engine, and the conversation loop.
- Tools reference — the full list of built-in and plugin capabilities.
- CLI commands — every
mblkcommand with options. - Configuration — the memoryblock filesystem and settings, end to end.
- Aiplug — the transport layer underneath memoryblock. Read this if you want to embed
aiplugin another project, or add your own provider.