Write in one place
The three-pane workspace keeps cluster sources, your document, and the agent session together. Shape the title, lead, and sections without losing sight of the material behind them.
AI Editor
The agent proposes every change. You decide what becomes part of your text.
Write in a focused newsroom workspace with the full source cluster beside you and an agent ready to help. Every suggested edit arrives as a reviewable hunk, so the text remains yours from first line to delivery.
Start with a blank draft or open material built from a source cluster.
You approve every change
- The editor opens a draft built from three articles in a trade-news cluster.
- You ask the agent to tighten the lead and support the deadline with another trade source.
- The agent reads the cluster and searches your collected clusters for relevant coverage.
- It attaches a fourth article that gives the January 2027 timetable.
- The agent checks how the revised claim relates to the attached source material.
- It proposes a replacement hunk, and you choose whether to accept or reject it.
The agent gathers relevant material, proposes a precise revision, and waits for your decision.
Sources, document, and agent stay in view.
Keep the reporting close
Each cluster source shows its publication and date. The agent can read every article in the cluster, so its suggestions begin with the same material you are reviewing.
Retain the final say
The agent cannot write directly into your text. It can only propose changes, and nothing enters the document until you accept it.
Begin with a blank page or a reported event.
The same editorial workspace supports original writing and cluster-led coverage.
Start from your own angle
Create an empty draft from the app's top bar and enter fullscreen immediately. The agent is available from your first sentence, while you remain the author.
- An empty document opens straight away
- Agent support is available from the first line
- Autosave runs from the start
Develop a cluster into a story
Open material assembled from all articles covering the same event. The agent reads the cluster, searches your own collected clusters, and can attach a useful article to strengthen the working material.
- Every source includes its publication and date
- Semantic search stays within your collected clusters
- Attached articles become part of the source material
Proposal, not control
Work with an agent that cannot overwrite you.
Ask for a rewrite, a new section, a deletion, a reordered list, or a field change. The agent can propose each operation, but only you can put it into the document.
- Every edit is a proposal
- Text replacements, item edits, insertions, deletions, reordering, and field changes all follow the same rule: the agent proposes and waits.
- Review one hunk at a time
- Each hunk shows the removed fragment beside the proposed addition. Accept or reject it on its own without committing the rest of the agent's work.
- See how the work happened
- The agent conversation forms a session. You can follow what it did and see which tool it used at each step.
Attached knowledge
Bring specialist references into reach when they belong there.
Define a knowledge base on your account using your own JSON API or PubMed, then attach it where the work needs it. Access is explicit, not account-wide by default.
- Grant access by attaching
- Attach a base to a goal, a workflow node, or an agent. Only then can that agent search it for material with its knowledge-search tool.
- Begin with useful defaults
- Every new account includes PubMed and Polish Wikipedia as default knowledge bases. You decide where either one is available.
- Keep reference material in its lane
- Knowledge bases do not create articles or clusters. Their material never enters the Stream or passes through a goal filter; it supports the writer inside the work.
Source grounding
Research further, then check every claim against the material.
The agent can find recent public-web sources and extract article text from a page. Its claim check evaluates only whether your given sources support the wording in the document.
- Extend the reporting
- Use web search to find fresh public sources and page retrieval to bring an article's text into the research process.
- Read a clear source verdict
- Each assessed claim is marked as supported, contradicted, or unsupported according to the source material you provided.
- Know what the check means
- This check measures the relationship between a claim and the supplied sources. It does not judge whether a source itself is reliable, and a failed run leaves the claim not checked.
- Backed by sources
- Contradicted by sources
- No source backing
Section contract
Write for a defined destination, not a loose template.
Each goal carries a section contract that tells the agent which sections a result requires and how each one is shaped. Metadata stays separate from the article body.
- Follow the receiving format
- The agent reads the goal's current contract and writes to that structure, keeping required sections and metadata in their proper places.
- Spot contract drift
- When the contract changes, an older result shows what was added, what dropped out, and what the receiving end no longer renders natively. You can then choose to regenerate it.
- Preserve delivered work
- A result that has already been delivered cannot be regenerated. This keeps your copy aligned with what the recipient already has.
Stay with the text while the workspace handles the edges.
Use the whole screen
Open the editor in fullscreen and hide the detail column when you want a quieter document view.
Let drafts save as you work
The status moves from ‘Changes will be saved automatically’ to ‘Saved automatically’ once the latest edits are stored.
Move without breaking focus
Use the left and right arrow keys to open the previous or next result without leaving fullscreen.
Send approved work where publication begins.
Create a WordPress draft
The SemanticHub plugin sends the approved result to WordPress as a draft post, ready for the CMS stage of your workflow.
Choose another route
Deliver to your own webhook endpoint or send the result by plain email to the newsroom.
Put approval before delivery
Delivery begins after an editor approves the result. Automatic delivery is a separate setting for each goal and remains off by default.
What editors usually want to know.
- Does the agent write the article for me?
- It can draft and propose changes, but it cannot place them directly into your document. You review each hunk and decide what becomes part of the article.
- Is the finished text mine?
- Yes. You direct the work and make every acceptance or rejection, so the finished document reflects your editorial decisions. The agent remains a visible collaborator, not an unseen author.
- Where does the agent get its claims?
- It works from the articles in your cluster, sources you attach from your own collected clusters, and public-web research you bring into the material. Claim checks report whether the wording is backed by, contradicted by, or unsupported by those sources; they do not assess the reliability of the sources themselves.
- Can I attach my own knowledge base?
- Yes. Add your own JSON API or PubMed as a knowledge base, then attach it to a goal, workflow node, or agent. The agent can search it only where you have granted that access.
- Can I work without a cluster?
- Yes. Choose New post in the top bar to open an empty draft in fullscreen, with the agent and autosave available from the start. You can write the piece from scratch.
- How does a result reach WordPress?
- The SemanticHub plugin creates a draft post in your WordPress CMS after editorial approval. Automatic delivery is configured separately for each goal and is off by default.
Keep the sources close and every decision yours.
Open a writing workspace where the agent proposes and you remain in control of the text.