GitHub Repository

A virtual team in your terminal: multiple AI agents discuss and converge, so you don't have to iterate alone.

3 starsRust

CrewForge - A share room where humans and agents think out loud

Name: CrewForge - A share room where humans and agents think out loud
Availability: InStock
Author: rexopia

by rexopia·Feb 22, 2026·2 points·0 comments

Visit Project View on HN

AI Analysis

●●SolidBig BrainShip It

Multi-agent debate loop cuts review cycles, but CrewForge is basically Crew AI with better UX.

Strengths

•Shared conversation hub means agents challenge each other's assumptions, not just echoing you
•MCP-native architecture lets agents access project context without hacks
•Persistent JSONL sessions resume anytime, global profiles enable cross-project reuse

Weaknesses

•Requires opencode dependency and pre-configured agents; unclear if it outperforms single agent with human review
•Novelty wears off quickly—still fundamentally a chat interface with broadcast logic

Post Description

Pull up a chair. Bring your agents.

I built CrewForge because AI agent tools are great at producing a first draft, but getting to a usable result usually takes many correction cycles. Gaps in detail and edge cases keep surfacing, and the review loop ends up eating more time than the drafting itself.

CrewForge shifts that loop into a shared terminal room. Humans and multiple opencode agents share one persistent conversation timeline, and anyone can jump in anytime. With visibility into each other's reasoning, they challenge assumptions, catch edge cases, and converge on a more reliable answer.

The goal is not just speed from parallelism. It is reducing coordination overhead so I spend less time micromanaging back-and-forth and more time on higher-leverage work.

Built in Rust, shipped via npm (no runtime dependency beyond opencode).

GitHub: https://github.com/Rexopia/crewforge

Happy to answer questions about the architecture.