Skip to content

Plant Protection Products · Biocides · REACH · Made in EU

Regulatory dossiers built by AI to pass the authority's check

The AI platform for EU regulatory dossiers. Agents review the literature, build the dossier and take it to submission, in a fraction of the time and cost. Your experts sign off the protocol, the screening, the assessment and the report.

  • RerunnableEvery query, date and hit count logged
  • ReasonedA coded reason for every exclusion
  • CitedThe exact quote behind every score

Diagram of the QST 713 pilot, point 8.3 (bees): a signed protocol passes through a search, screening, assessment and report agent, with a reviewer signing off the screening, the assessment and the report; 769 hits give 534 unique records, 72 are screened on title and abstract and 7 go to full text, ending in a search report and study table.

8.3 · BeesProtocolSignedSearch agent769 hits · dedup534Screening agentTitle and abstract72Assessment agentQuote and page7Report agentReport and tableDOCXSigned offSigned offPoint 8.3Search report+ study table534 records72 screened7 to full text

The authorities check what you hand them

ToxAgent helps you build each data point and reduces the error margin.

The search can be rerun
Every database, search string, date and hit count, across the time window of interest.
Criteria are set before screening
Inclusion and exclusion criteria each time adjustable by reviewers.
Every excluded study has a reason
A screening log with a coded reason for each exclusion, at title and abstract and at full text.
Relevant studies are read in full
Relevant papers are analysed by their entire text, never scored from the abstract.
Each study's reliability is justified
A score per criterion with the exact quote behind it, and a proposed Klimisch category.
A named expert stands behind it
Your reviewer signs off the protocol, the screening, the assessment and the report. Every approval is logged with name, time and what was approved.
Creation of assessments and reports
Assessments and reports building, tailored to regulatory needs.

Protocol proposed based on current EU regulations.

A case study: Potential effects of Bacillus amyloliquefaciens QST 713 on honey bees

Every exclusion has a reason. Every score has a quote.

Search hits across logged queries
769
Unique records after dedup
534
Screened on title and abstract
72
Sent to full text
7
ST-001 · R02 · Viable count of test item verified
“Biopesticides were prepared in the feeding solution (2 M sucrose) at the manufacturer's recommended concentrations.”

Agent: noCFU taken from the label, not measured. A reviewer may accept the nominal rate for a commercial product.

Environ Sci Pollut Res Int 2024 · doi:10.1007/s11356-024-32753-9 (opens in a new tab)

A search the rapporteur can rerun A reason for every exclusion A quote behind every score

Five stages Four signatures

Agents do the reading. A named reviewer signs off the protocol, the screening, the assessment and the report; the search runs the queries the signed protocol sets.

  1. 01

    Protocol

    Substance, data point, databases, years and criteria.

    You set and sign the scope

  2. 02

    Search

    Logged queries on open databases, plus your own exports. Duplicates merged.

    Search agent runs the queries you signed

  3. 03

    Screening

    Title and abstract, then full text. Uncertain calls come to you.

    Screening agent proposes, you decide

  4. 04

    Assessment

    Each study scored against your criteria, quote and page for every answer.

    Assessment agent scores, you confirm

  5. 05

    Report

    Search report, screening log and study table. Study summaries in IUCLID format are coming.

    Report agent drafts, you sign off

Questions before a pilot

Which dossiers does ToxAgent cover?

The literature part of EU plant protection dossiers under Regulation (EC) No 1107/2009, following the EFSA 2011 guidance on open literature. Biocides are next.

What do I get for each data point?

A search report with every database, search string, date and hit count, a screening log with a coded reason for each exclusion, and a study table with a reliability score per criterion, the quote behind it and a proposed Klimisch category.

Does the AI make the final call?

No. Agents propose and your expert decides. A named reviewer signs off the protocol, the screening, the assessment and the report, and uncertain screening calls come to them directly.

How is each sign off recorded?

Every approval is logged with the reviewer's name, the time and what was approved. Changing the criteria after the protocol is signed locks the review until it is signed again.

What happens with paywalled papers?

They are listed for you to purchase or upload. A study is never scored from its abstract alone.

Where is our data processed?

In the EU. Study PDFs are processed in memory and not stored. Saved studies stay in an EU database that only your account can read, until you delete them.

How does the pilot work?

Pick one data point the authority has already accepted. We run the review on your protocol and criteria, then you compare it with the accepted version: studies found, studies missed, exclusions and scores.

Bring us one data point

Pick one the authority has already accepted. We run the review, and you compare it with the accepted version.

What happens next

  1. 01Reply in two working days. A short call to pick the substance and data point.
  2. 02We run the review on your protocol and criteria; your expert signs off screening, assessment and the report.
  3. 03Compare with the accepted review: studies found, studies missed, exclusions and scores.