Core concept

Content pipeline

Each pipeline run walks through five steps. The full timeline is streamed over Server-Sent Events on /agents/<runId> so you can watch the agents work.

The five steps

1. Audit. The audit agent calls DataForSEO to pull SERP competitors for the topic and summarises gaps and overlap with your brand pillars.
2. Plan. The reasoning model builds a structured plan — angle, sections, supporting facts, target keyword targets — using the brand profile as a constraint.
3. Generate. The generation model writes the draft. For image-bearing content types (instagram_post, pinterest_pin) it also invokes the configured image provider.
4. Verify. The verifier scores SEO and GEO (Generative Engine Optimisation). The gate is SEO ≥ 80 + GEO ≥ 70. Failure triggers a regen up to the regen cap.
5. Publish. Approved items can be published via the connected social adapter for the requested type, or the operator can publish manually.

Events on the stream

• step_start — the named step begins.
• step_end — the named step succeeded; includes any structured output.
• regen — the verifier rejected and a regeneration loop has started (counted against the cap).
• done — the run finished; a content item now exists in the kanban.
• error — the run failed; includes a fault classification.

Kanban statuses

Drafts land in draft by default. The operator (you) advances them through in_review → approved → published. Verifier escalations land in requires_human_review; the dropdown menu on each card carries the allowed transitions.

Triggering a run

From /agents/new: pick a content type, write a topic, confirm pillars and audience. The run id is returned synchronously; the streamed events follow. You can navigate away and come back to the same run id — the snapshot endpoint GET /v1/agents/runs/<id> stays accessible for 30 minutes after completion.