Building Agent Applications,
Fast and Efficient.

Modular, Event-Driven Multi-Agent Collaboration Framework written in Go. Design complex business logic with highly concurrent and unified abstraction layers.

bash
go get github.com/openmodu/modu

Core Infrastructure

Everything you need to orchestrate complex agent behaviors and integrations.

Agent Engine

Generic, stateful Agent core with powerful tool calling and event streaming capabilities.

Mailbox Teams

Complete communication layer for multi-agent collaboration: message passing, task registry and status tracking.

Coding Agent

Advanced programming Agent building on core with session management, skill loading, and context compression.

Moms Telegram Bot

Go/Telegram port of the pi-mono mom Slack bot. Supports bash execution, file operations, and cross-session memory.

Unified Providers

Multi-provider streaming LLM interface abstracting away the differences between OpenAI, Anthropic, DeepSeek, etc.

Business Repos

Includes NotebookLM unofficial SDK, Gen Image Repos, and a powerful Web Scraper out of the box.

Agent Teams

Hub-and-spoke multi-agent coordination with asynchronous mailboxes, task registries, and real-time dashboards.

Mailbox Hub Task Registry · Event Bus SQLite Store Planner Agent inbox / outbox Worker Agent inbox / outbox Coding Agent inbox / outbox Reviewer Agent inbox / outbox
Independent Agents with typed Mailboxes
Hub routes messages, tracks tasks & events
SQLite-backed persistence, zero external dependencies

Coding Agent

AI-native tools designed around how LLMs think — not wrappers around CLI commands.

pkg/coding_agent

Every tool is designed around how a large language model thinks and reads: structured context windows, LLM-friendly output truncation, and semantic memory — not raw shell output.

  • Persistent sessions with context compaction
  • Dynamic skill injection from SKILL.md files
  • Spawn sub-agents for parallel task decomposition
  • Human-in-the-loop approval & slash commands
read_file
Chunked reads with line numbers — LLM sees exactly what it needs
edit_file
Surgical line-range edits — no full rewrites, preserves token budget
grep_search
Ripgrep-powered semantic search returning structured JSON results
memory_write
Typed key-value memory store persisted across sessions
bash
Sandboxed execution with smart output truncation built-in
spawn_subagent
Delegate subtasks to child agents — parallel, isolated, composable

One Interface. Unlimited Models.

Switch seamlessly across ecosystem leaders with zero codebase changes.

Anthropic (Claude)
OpenAI (GPT / o-series)
DeepSeek
Ollama (Local)
LM Studio