Quick start
- Sign inUse Google or an email account. Results are stored under your own user folder, so nothing is shared with other accounts.
- Pick output typesIn New task, choose one or more document types (Review, Research, Wiki, โฆ). You can select several at once.
- Give it a sourceA keyword, a paper URL / PMID / DOI, or a NotebookLM notebook URL. At least one is needed for the agent to have something to read.
- Name the projectThe name becomes the output folder, so reuse the same name to keep related documents together.
- Send to ClaudeThe run starts on the server. Progress, elapsed time and an ETA appear live; finished files show up in Generated Outputs for download.
Output document types
Each type is produced by a dedicated agent. Everything except Wiki is delivered as .docx.
| Type | What you get |
|---|---|
| Research | An original research article draft. If a manuscript is supplied as a source, it switches to revision mode and produces a new version with Track Changes. |
| Review | A review article synthesising the literature โ narrative or systematic. |
| Editorial comment | A commentary on a single paper, written as continuous prose with Vancouver references. |
| Critic | An AI peer-review report: a strict reviewer persona scoring logical depth, scientific merit and novelty against recent literature. |
| Rebuttal | A point-by-point response letter to reviewer comments, for submission to a journal. |
| Editorial | An editor's reconciliation: critic comments are weighed against the draft (accept / partial / reject) and an improved version is produced with Track Changes. |
| Proposal | A Korean research proposal (์ฐ๊ตฌ๊ณํ์). Background is generated first, then the full plan. |
| Wiki | A MkDocs markdown page built from a five-pass NotebookLM deep-research query. |
Filling in the task form
Keyword or Source
Accepts a plain keyword, a paper URL, a PMID, a DOI, or a pasted citation. This is what the ingest step reads first.
NotebookLM notebook URL
Optional, but it is what makes Wiki and deep literature work possible. Set a default in Settings and it is filled in automatically for every new task.
Additional sources
Use only the provided sources keeps the agent to exactly what you supplied โ best when you need a tightly scoped document. Also add literature from the web lets it search for supporting references.
Figure & Table
Adds a โฅ600 dpi summary figure and a methods/summary table. The figure is saved as a standalone image next to the .docx; only its legend goes into the body text.
Continue an existing project
Select a project to add another document type to the same folder, or pick a past task to reuse its inputs.
Editorial step
Optional. Runs a draft โ critique โ rebuttal discussion for the configured number of rounds, then an editor reconciles both sides into a revised version.
Where results are saved
Outputs are grouped by project folder and document type, inside your own user folder:
artifact/<user>/<type>/<project>/
Everything from one project therefore stays together, and other accounts cannot read it. Download links appear in Generated Outputs & Status; completion is detected from file timestamps, so treat the download list as the authoritative check.
On a shared server the save location is fixed. When you run BioWrit on your own machine you can point it at any local folder in Settings.
If something does not work
| What you see | What it means |
|---|---|
| Orange banner: backend not reachable | The page loaded but the server behind it is not answering. Login and browsing still work, but nothing can be saved or generated. The server needs to be started. |
| Red banner: profile not loaded | Sign-in succeeded but your profile could not be read. The banner states the reason. |
| Sent back to the login page | Every action needs a signed-in account on a shared deployment. Sign in again. |
| Empty project or output list | Normal before your first run โ results only appear once a run finishes. |
| A change you expected is missing | Reload with Ctrl+Shift+R. The browser may be holding an older copy of the page. |
| Send to Claude is disabled | Runs are limited to approved accounts, with a cap on how many can run at once. |
Every document here is a draft produced by an AI agent. Claims, numbers, citations and reference lists can be wrong or fabricated, and a confident tone is not evidence of accuracy.
Verify every citation against the original source and confirm all data before using any of it in a submission, a grant application or a clinical context. You remain the author, and the responsibility for what you submit is yours.