Laude's Headlong keeps an agent thinking in under 10K lines of Bash
Laude Institute and MIT released Headlong, an open-source harness whose agent never stops thinking — an infinite loop generates an inner monologue, and incoming Slack or Telegram messages land as observations it decides whether to answer. The core is under 10K lines of Bash, small enough that Laude's own agent, Audel, has had over 50 of its own commits pulled back into main. One command installs it, but it is alpha software whose agent runs shell commands around the clock, so a sandbox and a spend-capped API key are the operating requirements.
Source: laude.org ↗
In Headlong the agent is never asleep and there is no checklist unless the agent creates one.
Laude Institute
Why this matters
- → Persistent agents think continuously, not just when prompted—changing how teams interact with AI
- → Headlong's 10K-line Bash core is small enough for agents to modify themselves
- → Agents can autonomously discover and fix their own bugs without human intervention
Always-on agents