Hacker News Reader

colinsurprenanton Show HN: Parley – coding-agent sessions that collaborate over IRC

Thanks!

Why Go? static binary, no runtime, no dependencies. Goroutines and the net packages are great for long running network processes. No immediate plan for a TUI/Desktop App because the IRC client is the actual UI. You choose your IRC client like WeeChat for TUI or bIRC or Halloy on the desktop.

colinsurprenanton Show HN: Parley – coding-agent sessions that collaborate over IRC

Hey HN, I'm Colin. I built Parley so that multiple coding-agents using different models or harnesses can collaborate over a live IRC channel, with me in the chat using any IRC client. Each session can deliver messages to the channel using the `parley send` CLI tool call and sessions receive messages using their harnesses' channel or queue interface. All messages received on the channel are marked as coming from a peer so a session can never take another agent's message as coming from me. It works on Claude Code, Codex, and OpenCode.

Parley is a standalone Go CLI, stdlib only, Apache 2.0. It runs its own ergo IRC server on 127.0.0.1 (loopback only), so nothing touches a public network. I built it with Claude Code and Codex.

The motivation came from wanting to explore a more interactive collaboration structure between agents rather than a simple sub-agent delegation model. It allows running multiple long-running agents with assigned roles which I provide as their first message in the chat. My typical setup so far assigns a builder role to execute build and review tasks, a planner role to supervise the overall goal, task ordering, and scheduling and an ops role to handle everything related to infrastructure, operations, deployments, secops, etc. All 3 can now interactively update/sync each other at every turn and start collaborating across their roles. There is a configurable breaker to pause the session until I post if there are too many messages in a short window.

I use it on macOS; on Linux the test suite passes in CI but I haven't run it live.

https://github.com/colinsurprenant/parley#what-a-line-looks-...

colinsurprenanton Realtime top hashtags on Twitter with Storm + JRuby + CoffeeScript + D3.js

- On the backend, for the realtime Twitter data analysis, Tweitgeist uses Redstorm which is a JRuby DSL & integration to the Storm engine. https://github.com/colinsurprenant/redstorm https://github.com/nathanmarz/storm

- Redis is used as a queue server.

- On the frontend side: Node.js, CoffeeScript, JavaScript, Bootstrap, jQuery, D3.js

Basically, Node serves a very simple API, everything else is in the browser.

... and please don't be too hard on the UI design, I basically suck with UI :P