Skip to main content

AI agents

Agents are the heart of post.buzz's agent-first model. Instead of a generic "generate" button, an agent is a bounded worker assigned to one or more accounts that can plan, draft, schedule, and publish on your behalf — within limits you define, and with every decision recorded.

Find them under Automate → Agents. Each agent is a runbook: mission, sources, boundaries, and review policy in one configuration.

What an agent is​

An agent is bound to:

  • A primary account, plus optional additional channels — everything the agent produces is cross-posted to all of them as one unit (same caption, media, and time; one status).
  • A mode — how autonomous it is. See Modes.
  • Boundaries — how often it posts, what tone and length it writes, which post types it may produce, and how much may queue up for review.
  • Context — a brand profile, active goals, approved media collections, and the account's insights.

The runbook configuration​

Four sections, every field bounded:

Essentials — primary account, additional channels, and the autonomy mode.

Content & voice — the approved collections the agent may pull from (its media pool; an agent with no collections posts text), the when-media-runs-out strategy (post text, or generate a brand-matched image per post — generations draw from your image allowance), a default tone (professional, casual, witty, inspirational, educational, bold — or let the model decide), a default caption length, and the allowed post types (only types every targeted channel supports are offered).

Limits & schedule — posts per day (1–10), weekly limit (1–50), check interval (5–120 min), how far ahead it plans (1–7 days), and — for review modes — the max pending proposals (1–5).

Intelligence — the analysis window (7–90 days) and how many historical posts it studies (10–200), an audit preview of exactly what it may do, and the escalation triggers it watches for (low-confidence captions, complaints, publish failures, unusual posting times, rate-limit risk, large strategy changes).

What an agent does​

On each run (by its check interval, or immediately when poked):

  1. Checks its fences — capacity, weekly limit, pending-proposal buffer, quota. Skipping for a benign reason (nothing due, proposal buffer full, monthly AI quota exhausted) is a healthy no-op, not a failure.
  2. Plans a slot using best-time data, the account's posting gap, and per-day caps.
  3. Drafts a caption grounded in your brand profile, the account's personality, active goals, and its own top performers — then selects media from its collections (or falls back per your exhaustion strategy).
  4. Commits — publishes (autonomous modes) or queues a proposal in ideation (review modes).

Two guards keep output varied: near-duplicate captions are dropped, and an angle used three times in the last eight proposals forces a different direction. Cross-post captions respect the strictest character limit among the channels.

The audit log​

Every action an agent takes is written to a decision audit log that records the action (post proposed / published / failed / auto-published / rescheduled, media selected, schedule computed), the decision reason, and the insight basis it was grounded in. This is non-negotiable: agents are visible. The agent's Logs tab merges decisions and proposals into one timeline with filters, and lets you approve or reject pending proposals inline.

Agent health​

An agent's status is one of:

  • Active — running normally.
  • No content — no collections assigned (text-only fallback applies).
  • Error paused — hit its failure threshold (3 consecutive real failures) and deactivated itself; fix the root cause and re-enable.
  • Inactive — paused by you.

Skipping never counts as a failure; real errors do. The dashboard surfaces all of this at a glance, and the agent's Performance tab shows the proposal funnel (proposed → approved → published) plus engagement totals over 7/30/90 days.

Plan limits​

Your plan caps how many agents you can run (3 on Starter up to 50 on Agency). Each run that reaches generation consumes 1 AI action from your monthly allowance.

Start small​

The recommended path: begin in approval mode so the agent proposes and you review. Once you trust its judgment, raise autonomy step by step. See Modes.

In this section​