Skip to content

Agen V2.0.0 - Swarm AI CLI

Latest
Latest

Choose a tag to compare

View all tags
@AmanBanik AmanBanik released this 12 Aug 15:48
· 43 commits to v2-stable since this release
v2.0.0
f06154d

🚀 Agen V2.0.0 - Swarm AI Architecture Release

Welcome to the official 2.0 release of Agen! This release completely rebuilds the core engine from the ground up, transforming the CLI from a simple text interface into a fully autonomous, multi-agent AI software engineer.

✨ Major Features

  • Multi-Agent Swarm Orchestration (backend.swarm): The main agent can now spawn independent, asynchronous clone subagents in the background using the invoke_subagent tool to tackle multi-threaded software tasks concurrently.
  • Model Context Protocol (MCP): Dynamic integration with the MCP registry. The agent can instantly boot up servers like Tavily, Puppeteer, GitHub, or Postgres and execute external tools natively.
  • RAG Semantic Memory: Replaced simple file-attachments with a local ChromaDB vector database. Running /index allows the agent to semantically search and recall relevant snippets across your entire codebase.
  • Interactive REPL UI (prompt_toolkit & rich): Enjoy beautiful gradient rendering, LaTeX math parsing, native / commands, and intelligent @ file/folder attachment parsing without relying on shell commands.
  • Multi-Provider Support: Switch between Cloud mode (Google Gemini) and Local hardware execution (Ollama Llama3/Gemma) instantly via /gemini or /ollama.
  • Workspace Security Boundaries (New!): Agents are now strictly sandboxed to your active workspace folder. Any attempt to read, write, or list directories outside this boundary triggers an interactive terminal prompt, granting you manual override control.
  • Embedded Autonomous Skills (.agent/): The engine now natively ships with an embedded framework of agent "skills" (secure-scripting, ci-cd, etc.), ensuring all background subagents adhere to strict engineering standards.

📦 Installation

You do not need to install Python to use Agen. The entire engine (including RAG databases and MCP protocols) has been heavily optimized and compiled down to a single ~170MB standalone binary.

The Terminal Way (Recommended):

  • Windows: irm https://raw.githubusercontent.com/AmanBanik/Agen/v2-stable/setup.ps1 | iex
  • macOS / Linux: curl -sSL https://raw.githubusercontent.com/AmanBanik/Agen/v2-stable/install.sh | bash

The Manual Way:
Simply download the agen.exe binary attached below, drop it in your project folder, and start building!

For contributors, you can clone the repository and run pip install -e . to install it directly into your Python environment.

🛠️ Internal Refinements

  • Removed legacy V1 polling limits in favor of Server-Sent Events (SSE) streaming.
  • Wiped all hardcoded API keys and implemented secure .env placeholders.
  • Integrated a full GitHub Actions CI/CD pipeline for open-source contributions.
  • Included a standard CODE_OF_CONDUCT.md for community building.
  • Aggressively stripped heavy data-science dependencies (pandas, scipy, cv2) to shrink the Windows binary from 300MB down to 170MB.
Assets 3
Loading