The Social Media Agent Preset

Social is the most public thing an agent can do, so this preset pairs broad reach with the platform's most conservative defaults. Out of the box it drafts rather than posts: replies, posts, and engagement all land in a pending queue for your one-click approval, and you loosen that only when and where you choose to.

Seven Platforms, One Agent

The preset speaks Bluesky, X, LinkedIn, Mastodon, Threads, Facebook Pages, and Instagram through one consistent tool layer that handles posting, monitoring, analytics, and the pending queue across all of them. One agent, one set of instructions, and one brand voice covers your whole presence, the same announcement goes everywhere appropriate, each reply happens on the platform it arrived on, and the engagement data flows back into one place. You connect whichever subset you use, configuration lives in config/socialmedia.json with credentials per platform, edited in the Config tab's Social Media panel.

What a Tick Does

Three kinds of work fill each tick. Mentions and replies come first: the agent reads what people have said to you and about you across every connected platform and drafts responses in your voice. Publishing follows a paced schedule: posts go out at a natural rhythm rather than in bursts, drawn from proposals the research pipeline produces, which is a quietly powerful pipeline, research verifies something interesting in your space, and it becomes a sourced, accurate post proposal rather than filler. Discovery rounds it out: a light pass that finds conversations and accounts worth engaging in your space.

Review Mode and the Always-Queue Rule

The default posture is review mode: every outgoing item, post, reply, or engagement, queues as pending, and you approve, edit, or discard from the queue. Approval is fast by design, a queue of drafted items takes a couple of minutes to clear, and every edit you make teaches the system your taste through the memory bank.

One rule stays on even when you relax review mode: replies to large accounts always queue. Talking to an audience of millions is precisely where a human pause earns its cost, so high-visibility interactions wait for your eyes no matter how much autonomy the rest of the preset has earned. Anything ambiguous beyond that becomes a flag, the same human-attention mechanism the whole platform uses.

Voice and Direction

Two files steer the content. memory/system/socialmedia.md holds your social direction: themes, what to engage with, what to stay away from, and platform-specific notes. The shared brand voice file in memory/system keeps social consistent with every other channel your agents write. Both are plain instruction files, editable from the UI, applied on the next tick. With Levity connected, posts can carry generated images; the agents learn your running instance and use it where a visual helps.

Analytics That Close the Loop

Posting without measurement is guessing, so the platform measures on its own schedule. The schedule node pulls social analytics hourly alongside Google Analytics and Search Console, and cross-references the evidence, what was posted, what was engaged with, what drove traffic, so the question "is social working" has data behind it rather than vibes. Those results feed the same learning loop as everything else: engagement patterns land in the memory bank, the learning layer extracts what works into rewards, and the agent's drafts drift measurably toward what your audience actually responds to. Ask the master agent for a social summary any time, it reads the same numbers.

Running It

The standard controls apply: tick and model in the Models panel, individual pause, Kill Now, and the global pause. The natural progression most owners follow: start in full review mode, approve for a few weeks, and watch the edit rate fall as the agent learns your taste. By the time you consider loosening review for routine replies, you have a queue history that tells you exactly how often you actually change anything, and the large-account rule stays as the permanent backstop either way.