Swarms launches AgentHQ: hire Claude and Codex agents in a retro 2D pixel office
KyeGomezB · x · 2026-08-30
AgentHQ, a multi-agent orchestration tool from the Swarms team, is now in pre-beta with early access via waitlist at swarms.ai/agenthq. It renders multi-agent collaboration as a top-down retro pixel-art office: each agent is an "employee" with a desk — you hire it, name it, pick a model, walk up and hand it a task in plain language, and watch speech bubbles stream its real tool calls and output live.
Under the pixels, these are real coding agents: each employee runs on the Claude Agent SDK or OpenAI Codex SDK with real file and shell access, working inside its own directory on disk.
The core problem it targets is orchestration invisibility — once a run starts, you get walls of JSON in a terminal and only learn afterward which agent was stuck or idle. AgentHQ makes status spatial: an agent walking to its desk is thinking, one typing is working.
Supported collaboration modes: direct assignment, broadcasts that parallelize one task across the team, managers that decompose goals into subtasks, and sequential pipelines with handoffs. Each agent keeps its own persistent conversation, workspace, model, role and state, with isolated context sharing only explicitly passed results.
Related event: Swarms launches AgentHQ: watch AI agents collaborate in a pixel-art office(3 posts)→
More from coding & agent
- Hermes Agent running GLM-5.2 via OpenRouter on a 2GB VPS — granawkins · 2026-08-31
- Plan ordering decides success in 9-week codegen experiment — Ok_Astronomer_526 · 2026-08-31
- Optimizing Claude subagents to reduce high token costs — Nearby_Life9503 · 2026-08-31
- CREAO Launches Indie SaaS Creator to Build Full-Stack Apps via Natural Language — Aiden_Tech_Ai · 2026-08-31
- ITER: Interaction-aware retrieval improves deep-research agents — _reachsumit · 2026-08-31
- ChatGPT long threads now ~10x faster, performance boost available in open source Codex — gabrielchua · 2026-08-31