JSPM

  • ESM via JSPM
  • ES Module Entrypoint
  • Export Map
  • Keywords
  • License
  • Repository URL
  • TypeScript Types
  • README
  • Created
  • Published
  • Downloads 13
  • Score
    100M100P100Q70410F
  • License MIT

Deploy isolated, multi-agent AI assistants on your local machine with extreme resource efficiency.

Package Exports

  • memoryblock
  • memoryblock/commands

Readme

memoryblock logo

Run AI agents that remember, think, and work - without burning through your budget.

npm version downloads stars license

Website • Documentation • Report Bug


memoryblock preview
memoryblock web client dashboard preview

memoryblock is a lightweight framework for orchestrating isolated AI background workers. Instead of building monolithic chatbots, we provision local workspaces called blocks. Each block acts as a dedicated environment equipped with its own continuous memory, toolset, and execution loop.

Spawn a devops block to watch your infrastructure, a research block to scrape web data, and your primary home block to coordinate tasks. No state pollution, no crossed context boundaries.

  • Absolute Isolation: Every agent lives inside a dedicated filesystem directory (config.json, memory.md, logs/). To move an agent to a new server, you just copy its folder. No centralized databases to manage.
  • Token Pruning: Running background agents is historically expensive. memoryblock mitigates this by lazy-loading tool schemas only when invoked, and proactively truncating long tool outputs once analyzed. This typically halves the cost of sustained sessions.
  • Native & Lean: Zero dependency on heavy Node.js runtimes or Electron ecosystems. The core engine, HTTP server, and WebSocket router rely purely on Bun's native primitives for maximal I/O speed and minimal RAM overhead.
  • Model Agnostic: Provision blocks dynamically utilizing native definitions for OpenAI, Anthropic, AWS Bedrock, or Google Gemini.

Quick Start

memoryblock leverages Bun under the hood for native execution speed. If you use npm, we automatically manage the local Bun environment for you.

Install the framework globally:

bun install -g memoryblock     # Option A: Fastest
npm install -g memoryblock     # Option B: Managed internally

Initialize your environment:

mblk init

This interactive wizard verifies your local environment and configures your initial LLM provider credentials.

Start your default block:

mblk start [block-name]

Your autonomous assistant is now running in the background.

Working with Channels

Your blocks are decoupled from the UI. Communicate with them via the terminal, secure web dashboard, or standard chat clients.

Launch the Web Dashboard:

mblk web

Access real-time stream logs, cost tracking, and memory management at localhost:8420.

Route a block to social channels:

mblk start home --channel telegram

Securely interact with your existing agent state from anywhere without losing active history.

Command Reference

Command Description
mblk init Interactive setup - configure credentials and create your first block
mblk create <name> Create a new block (isolated AI workspace)
mblk start [block] Start a block's monitor loop (or all blocks)
mblk stop [block] Stop a running block monitor (or all blocks)
mblk status Show all blocks and their state
mblk delete <block> Archive a block safely (use --hard to permanently delete)
mblk restore <name> Restore an archived block
mblk reset <block> Reset memory, costs, and session (use --hard to wipe identity)
mblk permissions <block> View or update block permissions
mblk settings [plugin] View or edit plugin settings
mblk add [plugin] Install a plugin (no args lists available)
mblk remove <plugin> Remove an installed plugin
mblk server start Start the web UI and API server
mblk server stop Stop the running server
mblk server status Show server status
mblk server token View or regenerate the API auth token
mblk service install Register memoryblock to start on boot/login
mblk shutdown Stop all blocks and the server
mblk restart Full restart — shutdown then start everything as daemons

Full reference: command docs

Plugins

Blocks come with a core set of tools. Need more? Add them:

mblk add web-search    # Enables high-fidelity SERP querying
mblk add fetch-webpage # Extracts and chunks text from structured URLs
mblk add agents        # Allows blocks to spawn ephemeral sub-agents

Adapters for OpenAI, Anthropic, Google Gemini, and AWS Bedrock are natively supported out-of-the-box. Adding a custom provider adapter requires implementing a single unified payload interface.

Monorepo Architecture

memoryblock is built as a highly modular TypeScript monorepo utilizing a strict one-way Directed Acyclic Graph (DAG) for dependency management. All sub-packages are independently publishable.

Package Responsibility
memoryblock Global executable wrapper, setup tooling, and CLI orchestration.
@memoryblock/core Extracted engine runtime (Gatekeeper, Memory Manager, Monitor loops).
@memoryblock/types Centralized Zod validation schemas and TypeScript interfaces.
@memoryblock/daemon Low-level OS process spawner and background lifecycle manager.
@memoryblock/adapters Unified REST/SDK implementations for LLM providers.
@memoryblock/channels Transport layer for CLI, WebSockets, and messaging platforms.
@memoryblock/tools Core functional schemas (File I/O, OS interactions).
@memoryblock/api Fast, dependency-injected HTTP web server integration.
@memoryblock/locale Formatting tools and centralized translation strings.
@memoryblock/web Standalone UI distribution package.

Community & Support

The memoryblock Ecosystem

memoryblock is a highly modular system. Here are the official packages:

The Core

Integrations & Tooling

Plugins

License

Distributed under the MIT License. See LICENSE for more information.

Website Badge Sponsor Badge