[{"data":1,"prerenderedAt":4},["ShallowReactive",2],{"readme:headlong":3},"\u003Cpre>\u003Ccode>██  ██ █████  ████  █████  ██     ████  ██  ██  ████               ██\n██  ██ ██    ██  ██ ██  ██ ██    ██  ██ ███ ██ ██           ██   ██\n██▀▀██ ████  ██████ ██  ██ ██    ██  ██ ██▀███ ██ ███     ██   ██\n██  ██ ██    ██  ██ ██  ██ ██    ██  ██ ██ ▀██ ██  ██   ██   ██\n██  ██ █████ ██  ██ █████  █████  ████  ██  ██  █████      ██\n\u003C\u002Fcode>\u003C\u002Fpre>\n\u003Cp>\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Flaude-institute\u002Fheadlong\u002Factions\u002Fworkflows\u002Fci.yml\" rel=\"nofollow ugc noopener\">\u003Cimg src=\"https:\u002F\u002Fgithub.com\u002Flaude-institute\u002Fheadlong\u002Factions\u002Fworkflows\u002Fci.yml\u002Fbadge.svg\" alt=\"CI\" \u002F>\u003C\u002Fa>\u003C\u002Fp>\n\u003Cp>\u003Ca href=\"https:\u002F\u002Fwww.laude.org\u002Fupdates\u002Fheadlong-a-microharness-for-persistent-agents\" rel=\"nofollow ugc noopener\">\u003Cstrong>Headlong\u003C\u002Fstrong>\u003C\u002Fa>\nis an open source agent microharness, a complete agent harness with a core of\nabout 10K lines of Bash. Headlong's defining feature is \u003Cstrong>persistent\nagency\u003C\u002Fstrong>. Your agent keeps thinking between external interactions in a\nself-guided loop inspired by human inner monologue. A message from a human\ndoesn't start a session. It lands in the agent's thought stream as one more\nobservation, and the agent decides if and when to respond. You give your agent\na name and a personality, and it sets its own interests and priorities, starts\nits own projects, and pings you when it has something to say.\u003C\u002Fp>\n\u003Cp>A Headlong agent is also built to be shared. A whole team can talk to one\nagent over Slack, Telegram, and a chat app, and every conversation lands\nin the agent's single stream of thoughts. The agent follows what different\npeople are working on, connects them, and pings whoever seems most\nrelevant. Sharing one agent is fun, because it behaves more like a person\nthan a service.\u003C\u002Fp>\n\u003Cp>At the heart of Headlong is \u003Ccode>shellm\u003C\u002Fcode>, a Bash implementation of a\n\u003Ca href=\"https:\u002F\u002Falexzhang13.github.io\u002Fblog\u002F2025\u002Frlm\u002F\" rel=\"nofollow ugc noopener\">recursive language model (RLM)\u003C\u002Fa>.\nThe agent thinks by writing shell commands, running them, and reading the\noutput. No tool system besides Bash is needed.\u003C\u002Fp>\n\u003Ch2>Get started\u003C\u002Fh2>\n\u003Cp>One line installs everything, interviews you to bring a Headlong agent to\nlife, and opens a dashboard where you can watch its mind run:\u003C\u002Fp>\n\u003Cpre>\u003Ccode class=\"language-bash\">curl -fsSL https:\u002F\u002Fheadlong.ai\u002Finstall.sh | bash\n\u003C\u002Fcode>\u003C\u002Fpre>\n\u003Cp>You'll need bash 3.2+, git, curl, jq, and an LLM API key (Anthropic,\nOpenAI, Gemini, or OpenRouter); the dashboard also needs\n\u003Ca href=\"https:\u002F\u002Fdocs.astral.sh\u002Fuv\u002F\" rel=\"nofollow ugc noopener\">uv\u003C\u002Fa> and bun or node, and the installer offers\nto fetch those. \u003C\u002Fp>\n\u003Cp>Headlong is alpha research software. Use a dedicated, spend-capped key, because\nyour agent runs real shell commands and thinks around the clock. With Docker\nrunning, the installer offers to keep the whole agent in a container, or to\ninstall on your machine with the agent's commands sandboxed in a container\n(an unsandboxed host install exists too, behind an explicit yes, and is not\nrecommended).\nWithout Docker the commands would run directly on your machine as you, so the\ninstaller stops and asks for an explicit yes before setting that up. How much\nthe background thinking costs depends on how quickly\nthe agent loops and which model backs it. The rate of thinking backs off\nexponentially when nobody is talking to the agent and resets the moment a\nmessage arrives.  At the settings we run our agent with, it comes to $1 to $2\nan hour.\u003C\u002Fp>\n\u003Cp>The agent's name becomes a command:\u003C\u002Fp>\n\u003Cpre>\u003Ccode class=\"language-bash\">ada hello            # one message, wait for the reply\nada                  # chat\nada stop \u002F ada start # pause \u002F resume its mind\nada dash             # open the dashboard\nada bugreport        # bundle logs + trajectory (keys scrubbed) for a bug report\n\u003C\u002Fcode>\u003C\u002Fpre>\n\u003Cp>\u003Ccode>headlong-killall\u003C\u002Fcode> stops every Headlong process on the machine if you need a\npanic button. \u003Ccode>curl -fsSL https:\u002F\u002Fheadlong.ai\u002Fstatus.sh | bash\u003C\u002Fcode> shows what is\ninstalled and running; \u003Ccode>curl -fsSL https:\u002F\u002Fheadlong.ai\u002Funinstall.sh | bash\u003C\u002Fcode>\nremoves it all (details in\n\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Flaude-institute\u002Fheadlong\u002Fblob\u002FHEAD\u002Fdocs\u002Finstall.md#stopping-and-uninstalling\" rel=\"nofollow ugc noopener\">docs\u002Finstall.md\u003C\u002Fa>).\u003C\u002Fp>\n\u003Cp>The container flow the installer offers is this, and you can also run it\nyourself:\u003C\u002Fp>\n\u003Cpre>\u003Ccode class=\"language-bash\">docker run -it --name headlong --restart unless-stopped -p 8080:8080 buildpack-deps:curl \\\n  bash -c 'curl -fsSL https:\u002F\u002Fheadlong.ai\u002Finstall.sh | bash; exec bash'\n\u003C\u002Fcode>\u003C\u002Fpre>\n\u003Cp>Details, no\u003C\u002Fp>\n",1787958255013]