Glossary
Every SemanticHub concept defined once, in one place.
A concept is defined here and nowhere else. The rest of the documentation uses the terms and points back to this list.
- Goal
The configuration of a single monitored topic: a set of sources, a relevance filter, and clustering thresholds.
New goal · All goals
- Source
A place the system fetches material from — a site, a channel, or a profile added to a goal.
Goal → Sources
- Article
A single piece of material fetched from a source, which the system judges for relevance and groups with similar ones.
Goal → Feed · Before filter / After filter
- Relevance filter
A plain-language instruction the model uses to decide whether fetched material belongs to the goal's topic.
Goal → Filter
- Cluster
A set of materials that the system judged close enough in meaning and carries forward as a single story.
Goal → Clusters
- Candidate
A cluster that has not collected as many articles as its goal requires yet, and is waiting for more or for a human decision.
Goal → Clusters · Candidates
- Similarity threshold
A goal setting that says how close in meaning two materials must be to land in the same cluster.
Settings → Thresholds · Similarity threshold
- Minimum article count
The number of articles a cluster has to collect before it stops waiting and moves on.
Settings → Thresholds · Min. articles
- Workflow
A graph of AI steps attached to a goal that turns the contents of a cluster into a result.
Goal → Workflow
- Workflow step
A single workflow node with its own model, prompt, and skills. The text of the step wired into the Composer becomes the result body.
Goal → Workflow · Add step
- Result
The outcome of processing a cluster, handed over at the end of the goal's work and ready for review or for delivery.
Goal → Post moderation
- Review
The stage at which a person approves or rejects a result before it leaves the system.
Goal → Post moderation · Needs approval
- New post
An empty post created by hand in the current goal, with no cluster and no source articles. It opens straight away in the full-screen view with the agent panel and goes through the same moderation as any other result.
Top bar → New post
- Final result format
The set of elements (sections) a finished result is made of, such as main content, FAQ, or a timeline. You set it in the Composer panel. Main content is mandatory and comes from the step wired into the Composer. The Composer writes the other sections itself.
Goal → Workflow · Composer · Final result format
- Delivery
Handing a result to the recipient configured on the goal.
Action → When to send
- Test mode
A goal state in which deliveries stay inside SemanticHub instead of reaching the recipient's real address.
Action → Delivery mode
- Webhook
A delivery channel that sends a result as an HTTP POST request to a given address. In the app it is the HTTP address channel.
Action → HTTP address · Recipient URL
- API token
A personal secret that supported integrations use to authenticate against the SemanticHub API.
Profile & account → API tokens
- Knowledge base
An external knowledge source — your own JSON API, or a catalogue SemanticHub ships with, such as PubMed — that the model can search while writing. Access comes from attaching the base to a goal, a workflow step, or an agent. A knowledge base never creates articles or clusters; nothing from it enters the Feed.
Library → Knowledge bases
- AI-assisted setup
A way of creating a goal in which you answer an agent's questions by clicking ready-made controls instead of filling in a form. The conversation runs through seven stages — opening, sources, filter, thresholds, workflow, action, wrap-up. The goal exists as a draft after the first stage, so switching to the manual form edits that same goal. A new goal starts in test mode.
New goal · AI-assisted
- Plan
The purchased service tier of an account, which decides what the account may do and how much of it.
Profile & account → Plan & limits
- Limit
The ceiling the plan sets on one account resource. Limits do not share a single period: some are capacity (goals, sources, workflow steps), some are daily windows (filter decisions, agent messages, research runs), and retention is measured in days of history kept.
Plan & limits → Limit usage