AI agents for R&D, every claim sourced before your review.
Hand Duale AI a research question, a candidate or a case file. You get back one result in the format you set. One Task reads the papers, patents and files in focused branches and runs in the background.Literature synthesis, what the evidence supports and what it leaves open
Your team sees how strongly the sources back each claim, with the quoted passage beside it.You hand over: “Map the evidence on recycled content in packaging film for Monday’s project review. How does strength change, what offsets the loss, and below which share does it hold?”
How the Task splits it: one branch per sub-question searches and reads the published papers with the Web research built into Tools. A review branch then checks every claim against the passage it quotes. You can watch the same fact-check in the 3-minute live demo on the use cases page.
You get back: one evidence map. Each claim comes with its quoted passage, its source and date, and how well the sources support it. The gaps that no source answers come last.
Illustrative example: an evidence map on recycled content in packaging film, ready for Monday’s project review. The papers are fictional.
- Claim
- Tear strength falls as recycled content rises
- Quoted passage
- “Tear resistance fell by 18% at 50% recycled content.”
- Source
- Paper A (2025), section 3.2
- Status
- Supported by two papers
- Claim
- An additive offsets most of that loss
- Quoted passage
- “With 5% additive, the loss fell to 4%.”
- Source
- Paper B (2024), table 4
- Status
- One study; to confirm
- Claim
- Below 20% recycled content, strength holds
- Quoted passage
- No passage found
- Source
- None
- Status
- Gap: no study tests it
Your team decides: the research direction, and which gap to test first.
Candidate compound dossiers, checkable line by line
Each candidate reaches your portfolio review with a dossier that separates calculated values from sourced ones.You hand over: “Prepare the dossier on this candidate additive for marine coatings: its properties, the prior art and our earlier tests, for the portfolio review.”
How the Task splits it: separate branches read the published literature, the patents and your earlier test reports, kept in a Library. Code execution calculates the properties you ask for. The dossier marks which values were calculated and which come from a source.
You get back: one dossier with the calculated properties, the prior art with each source dated, your earlier results that bear on the candidate, and the open questions.
Illustrative example: a dossier on a candidate additive for marine coatings. The compound and the patents are fictional.
- Section
- Calculated properties
- What comes back
- Molecular weight of 312 g/mol; low estimated water solubility
- Source
- Calculated with code execution from the structure you provided
- Section
- Prior art
- What comes back
- Two patents claim close structures for marine coatings; one expired in 2024
- Source
- Patent A and Patent B, each with its date
- Section
- Your earlier tests
- What comes back
- A close analogue failed your adhesion test in 2023
- Source
- Your March 2023 test report, page 6
- Section
- Open question
- What comes back
- Does the candidate pass the adhesion test that its analogue failed?
- Source
- To test in your lab
Your team decides: whether to advance the candidate, test it further, or drop it.
Case pre-reads, each deviation sourced and set against your guidelines
Your review board opens each case with the questions to settle already listed.You hand over: “Prepare the pre-read on the returned heat gun case for tomorrow’s product safety board.”
How the Task splits it: one branch per section of the case file reads it against the guideline sections that apply. The Task gathers what they find into one pre-read.
You get back: a summary of the case, the quoted guideline passages that apply, each deviation with its source in the file, and the questions for the board.
Illustrative example: a returned heat gun, a fictional product, before tomorrow’s product safety board.
- Case file section
- Incident reports
- Guideline passage
- “Three reports of the same failure within 30 days open a safety review.” (section 4.1)
- Finding
- Four overheating reports within 22 days
- For the board
- Review opened as required
- Case file section
- Test results
- Guideline passage
- “Surface temperature stays below 70 °C after 10 minutes of use.” (section 6.3)
- Finding
- One lab test recorded 78 °C after 10 minutes
- For the board
- Retest the current batch?
- Case file section
- Design changes
- Guideline passage
- “A new supplier for a heating part requires a new test.” (section 7.2)
- Finding
- The heating element supplier changed in May; no new test is on file
- For the board
- Deviation: ask engineering for the test
Your board decides: the outcome of the case and the actions that follow.
The capabilities behind each research result
Each one is set up once on the Platform and serves your other use cases too.Libraries
Keep your test reports and guidelines in Libraries. Each result cites the document and the page it used.
Tools
Web research reads published papers and patent records, each with its date, and code execution calculates the values you ask for.
Model routing
Each branch runs on a model you enabled that suits its work: long papers on one, calculations and synthesis on another.
Deployment
Your unpublished research and your case files can stay in your own infrastructure.
Before your first synthesis
The facts your security, finance and procurement colleagues check.One fixed price per Task
A standard Task costs €0.10 and a fast Task €0.20, excluding VAT, whatever the number of branches or the duration. See pricing
Your providers, your contracts
Your model providers bill you directly for inference.
Your choice of hosting
Use the managed service, hosted in the European Union, or run the Platform in your own infrastructure.
Encrypted storage
All stored data is encrypted, from Library files to database backups.
Unlimited agents
Add agents as your teams and use cases grow.
Your content remains yours
Duale AI does not reuse your content to train a model. See subprocessors
R&D questions
Yes. The evidence map sets each finding from your internal reports beside the published claims it supports or contradicts. See how Libraries keep your reports, then see how Web research in Tools brings in public sources.
Yes. The dossier’s calculated values come from the code your team already trusts. See how your own code becomes a Tool.
Yes. The pre-read applies both, and the board sees whether each passage is an internal rule or a published standard. See how Libraries keep your guidelines.
Yes. Keep each finished evidence map in a Library, and the next synthesis starts from it, cites it and adds what was published since. See how Libraries keep your past work.