SemanticHub does the work of a large newsroom at a fraction of the cost, 24/7 and without holidays

A goal is one standing brief: a set of sources, a filter that decides what is on topic, and a workflow that turns a cluster into text. We count goals that exist, not runs — deleting a goal frees the slot immediately. Nothing resets, because this is capacity, not a daily window. Trying to create a goal beyond the limit is refused at save time; the goals you already have keep running untouched. Your own key changes nothing here — you are paying for system capacity, not for tokens. One goal is enough while you watch one topic; the second usually comes from a second audience, not a second niche.

A source is a portal, an RSS feed or a profile we pull content from. We count unique portals attached to your account, not assignments to goals: the same portal used in three goals takes one slot, not three. No reset — this is capacity. Once the limit is reached you cannot add another source, and the goal wizard's automatic source discovery stops at the same boundary. Your own key does not apply here: fetching content costs no tokens.

On Pro and Scale you can connect your own AI provider account for article filtering, text generation and conversations with the agent. We store the credentials encrypted and use them only for your work; usage is billed directly by the provider. This connection does not increase the other plan limits.

A node is one step on the canvas: fetching a cluster, a prompt, a branch, publishing. We count nodes within a single workflow, checking the limit when the canvas is saved. No reset. A save carrying too many nodes is rejected whole, before we delete the previous graph — a refusal will not leave you with an empty canvas. The median across current users is six nodes, and the largest workflow actually in use has ten; five on Free is tight, but it is not theoretical.

The agent is a conversation about finished text: you ask for a cut, a change of tone, a number checked. One message is one turn of yours — the agent's reply, however long or multi-step, does not count separately. A correction inside the same conversation is a new message, because it is a new model call. The counter resets at midnight UTC. Once it is spent the agent declines until the next day, while the rest of the system keeps working. Scale has no window. Your own key does not remove it.

Research adds context from outside your own sources to a cluster: it searches, reads and summarises material from the web. We count search queries, not clusters — a single cluster can spend several. Reset at midnight UTC. Once the pool is spent research is skipped and the cluster carries on without enrichment; nothing stops. Why the numbers are so low: we pay an external search provider for every query, per query. This is the only axis in the table that reflects a real invoice rather than the capacity of our system.

We do not delete your data after seven days. The plan states only the maximum of a slider — you set the lifespan of articles and results yourself, per goal, within the plan's bounds. On Free you may keep articles for up to seven days, on Pro up to ninety, on Scale up to a year; if you set less, your value applies, not ours. Once the time is up an article is marked and later swept by housekeeping. Results run on their own, longer clock, because results are the product of the work. After moving to a lower plan, settings above the new maximum are trimmed to it the next time the goal is saved.