Your first investigation
From sign-in to a checked answer you can take away. You need an account that Essarion has already set up for you.
01Sign in
Open science.essarion.com, choose Sign in, and enter the email address and password for your account. You land in the Workbench.
There is no sign-up page. If you do not have an account, see Access & accounts.
02Ask a question
Write the question the way you would ask a colleague. A short prompt is fine: the agent works out what the answer is for before it starts, and shows you that reading.
Look into ROCK2.
What does rs334 do to the protein, and how common is it?
Which GLP-1 receptor agonist trials stopped early, and why?
What is this sequence?
MQIFVKTLTGKTITLEVEPSDTIENVKAKIQDKEGIPPDQQRLIFAGKQLEDGRTLSDYNIQKESTLHLVLRLRGG
Propose mechanisms worth testing in dry AMD.03Set depth and sources
This step is optional. The Databases & skills button by the composer opens a sheet titled “What this run may use”.
- Depth: Scan, Deep dive or Exhaustive. Deep dive is the default. See what depth changes.
- Categories, Connectors and Skills: switch whole groups of databases, single databases, or computational skills on and off. Everything is on by default, and nothing needs a key.
Connector and skill choices are remembered in your browser and sent with every message. Depth goes back to Deep dive when the page reloads.
04Attach data
Also optional. Use Attach research data, drop files onto the composer, or paste a screenshot. Each file can be up to 64 MB. The agent sees the files attached to the session, reads them with its file tools or in the notebook, and can look at an image when the question depends on what it shows.
05Watch it work
Send the message. A few things appear:
| You see | What it tells you |
|---|---|
| The reading | A collapsed card showing how the agent understood the request, what it will hand over, and the assumptions it made. It appears before the first tool call. |
| The work log | The agent's reasoning, with its tool calls folded into short summary lines you can open. |
| The status tray | Above the composer: what is running now and how long the turn has taken. |
| The workbench panel | Open it from the composer. Notebook holds the code cells, records, charts and 3D models, plus a console into the same kernel. Evidence and Sources hold what the turn found and where from. |
The stop button ends the current turn. Anything that ran before you stopped stays on the page, and the session carries on.
06Read the answer
Read it in this order. It is the order that catches problems fastest.
- The reading. Did it answer the question you meant?
- The reviewer card. A separate reviewer pass checks the answer against the evidence. If it found problems, the answer was rewritten; the first draft is folded away, shown as a diff against the final one.
- The citations. A marker after a sentence opens its support: a source shows that source's own sentence, and a computed result names the tool and the figure it returned. A sentence with no marker is the agent's own reasoning.
- The document. Its header counts the claims and the share that rest on reasoning alone. It also lists the operations that ran, including the ones that failed.
- The artifacts. Open any figure or model to see the code that made it, what that code printed, the files it read and the environment it ran in.
07Keep going
Follow-up messages continue the same session. The kernel, the attached files and the sources carry over, so you can ask for the next step without repeating the context. Each message uses the depth selected when you send it.
To change a chart, open it and click where the change is needed, then leave a note. That sends a message asking the agent to update the figure's code and redraw it. The new version is kept alongside the old one.
08Take it with you
| What | Format |
|---|---|
| The turn's document | Markdown (.md). |
| Tables | CSV of what is on screen, after your filters and sorting. Tables can also be copied. |
| Figures and models | The script that produced them (.py). |
| Bundled datasets | CSV, whole or the current view, from the Data page. |
| Fold and Studio | PNG images with labels; from Studio, also FASTA, PDB and an edit report. |
09Where to go next
- The workbench: how the agent picks its next step, and what the kernel can run.
- Data sources: the 29 databases and the two bundled datasets.
- Verification: what is checked before an answer reaches you.
- Engines & discovery: the tools that compute what no database holds.