Skip to main content

Autonomy & trust modes

Every agent runs in one of three modes. The mode is the single biggest lever for how much trust you're placing in the agent — start low and raise it as the agent earns confidence.

ModePublishes?Review?Best for
ApprovalNo — proposesRequiredStarting out, high-stakes accounts
SequentialOptional (autoPublish)Per stepStep-by-step pipelines
AutonomousYesNoneTrusted, hands-off operation

Approval mode​

The agent proposes posts but never publishes them. Proposals queue up for your review in ideation; you approve, edit, or reject each one.

  • A proposal buffer (1–5, default postsPerDay + 1 capped at 5) limits how many pending proposals accumulate, so proposals don't pile up while you're away. When the buffer is full, the agent skips its run — cleanly.
  • Use this when you're calibrating the agent's voice, or for accounts where a bad post is costly.

This is the default mode — and the recommended starting point. Agents created during onboarding start here.

Sequential mode​

The agent works through a sequence of steps — plan → draft → media → schedule — pausing between steps, and posts through its collections in order. The autoPublish switch controls the final step:

  • autoPublish: false (default) — drafts proposals for review (like approval mode, but structured as a pipeline through your collections).
  • autoPublish: true — runs the pipeline to completion and publishes.

Use sequential mode when you want the agent to walk your approved media collection in a defined order.

Autonomous mode​

The agent publishes within its boundaries with no per-post review. It still respects every limit: posts per day, weekly limit, posting gaps, caption tone and length, enabled post types, and diversity guards.

If the agent's collections run dry (every media item used, or none assigned), it doesn't go quiet. Each agent picks a fallback strategy in its settings: Post text (default) publishes a standalone text post on the channels that accept one, and Generate images creates a brand-matched image (1 image generation) to attach to the post. If image generation fails — or none of the agent's channels accept images — the run degrades to the text fallback. Every volume limit still applies. Review modes do not fall back — the slot is skipped and logged so you can top up the collections.

Use autonomous mode once you've reviewed enough proposals to trust the agent's judgment. Everything is still logged to the audit trail, so you can audit after the fact and dial back if needed.

The trust dial​

The intended progression:

  1. Start in approval mode.
  2. Review proposals, adjust the brand profile and tone until the output consistently fits.
  3. Switch to sequential with autoPublish: false to watch it work through your collections.
  4. When you're confident, go autonomous (or sequential with autoPublish: true).

You can change modes or pause an agent at any time.

Failure handling​

If an agent hits its failure threshold (3 consecutive real failures), it pauses itself (Error paused) rather than hammering a broken path — and notifies you. Check the audit log to see what failed, fix the root cause, and re-enable; re-enabling resets the failure counter. Benign skips (capacity, buffer full, quota exhausted) never count toward the threshold.