What is the virtual lab and how does it simulate experiments?

The virtual lab is a physics and chemistry-aware simulation environment. AI agents run your experimental protocol step-by-step, validating thermodynamic constraints, checking kinetic feasibility, and flagging solubility or compatibility issues before you go to the bench. Every step is grounded in real physical and chemical laws.

How does Vicena find and extract protocols from scientific sources?

The AI agent searches PubMed, arXiv, and the open web for papers matching your experiment. It then extracts every variable -- temperatures, concentrations, durations, reagents -- independently from each source, cross-references them, and assigns confidence scores so you know which values are well-supported and which need verification.

Can the AI agents actually carry out experiments in a virtual world?

Yes. The simulation agent takes your complete protocol and executes each step in a virtual environment that models real-world physics and chemistry. It checks energy balances, reaction rates, and reagent compatibility at every stage, producing a feasibility report that tells you whether the experiment will work and why.

What scientific databases does Vicena search?

Vicena searches PubMed, arXiv, Google Scholar, Google Patents, PubChem, and the open web in parallel. It uses adaptive query refinement to ensure comprehensive coverage of your research topic and can cross-reference results across all sources.

Scientific Tools for AI Research

Computational Modeling & Simulation

CFD & Aerodynamics

Geometry repair, body-fitted meshing, steady and transient flow, and aerodynamic verification.

4 packages

OpenFOAM

13

Primary CFD environment

Builds and checks volume meshes, solves steady and transient flow, and extracts pressure, velocity, and aerodynamic forces.
Official project
snappyHexMesh

OpenFOAM 13

Body-fitted meshing

Turns validated surface geometry into a three-dimensional computational mesh with local surface and wake refinement.
Official project
trimesh

4.12

Geometry audit and repair

Inspects connected bodies, boundaries, normals, scale, triangle quality, and surface deviations before CFD meshing.
Official project
SU2

8.5

Aerodynamic verification

Provides an independent open-source route for airfoil, external-flow, multiphysics, and design-optimization studies.
Official project

Open application page

Computational Modeling & Simulation

Computational Chemistry & Molecular Dynamics

Quantum chemistry, molecular simulation, protein cofolding, docking, free-energy, and geometry workflows.

15 packages

Open application page

1 agent tools

Workspace

Persistent files and folders for saved research documents, uploads, generated outputs, and reusable scientific work.

🗂️

Workspace Files

Persistent research file system

Create, read, edit, organize, and search the persistent files shown in the Vicena Files sidebar. Use this for durable research documents, saved protocol audits, uploaded PDFs/images/CSVs, markdown notes, generated plots, and project folders. Do not use this for temporary Jupyter or Linux computer files; use Compute Shell for quick execution work and Workspace Files for files you want to keep.

View methodology and examples

How it works

The agent treats Workspace Files like a small private research drive. It resolves or creates normal forward-slash paths, reads existing files when needed, writes durable Markdown or data files, updates the Files sidebar, and returns clickable workspace links. When a task needs computation, the agent should use the Science Computer separately and must not claim files moved between the two systems; a direct bridge is not available yet.

Workspace File Flow

The agent uses Workspace Files as the durable source of truth for user-facing documents. It can start from an empty workspace, create folders, save research outputs, or edit existing files without involving the compute sandbox.

1 Understand the file goal Decide whether the user wants to create, read, edit, organize, or search durable Workspace files.

2 Resolve the path Use the given path when exact. If the user names a file or folder loosely, search or list before choosing.

3 Perform the file action Create folders, write a new document, read a file, patch text, move/rename, or delete only when explicitly requested.

Empty workspace

Start path

Useful when the user has no files yet and wants Vicena to create the first project folder or document.

1 Create a project folder Choose a clear path such as /Projects/Battery Literature or /Notes.
2 Write the first file Create a README, literature-review note, checklist, protocol draft, or audit certificate.
3 Return a workspace link Show the saved file path so the user can open it in the right panel.

Existing workspace

Maintenance path

Used when the user already has uploaded files, saved audits, or project folders.

1 Search or list files Find candidate files by path, name, or folder before reading or editing.
2 Read before editing Inspect the current file content before patching targeted sections.
3 Update only the requested file Keep edits scoped and preserve unrelated workspace files.

Computer boundary

Separate volume

Used when a task involves both durable Workspace files and temporary Science Computer execution files.

1 Use the right volume Workspace Files are durable documents; Computer files are for execution and notebooks.
2 Do not claim transfer Direct Workspace-to-Computer file transfer is not available yet; the agent should say that clearly instead of implying a copy happened.

Durable result

The user gets a saved file, an updated folder, a rendered preview, or a clear report that no matching file was found.

Input

Natural language instructions such as “make a project folder,” “save this as /Notes/summary.md,” “do a literature review and write it to a file,” “read the PDF I uploaded,” or “move this audit into the project folder.”

Output

Directory listings, file contents, rendered previews, saved documents, or confirmations with clickable workspace paths that open in the right panel.

Limitations

Workspace Files are separate from the Science Computer/Jupyter filesystem. It does not run code by itself and it does not automatically reorganize or delete files without a clear request. Direct copying, importing, exporting, or moving files between Workspace Files and the Science Computer is not available yet. Free accounts have 0 GB storage; Plus has 2 GB and Ultra has 10 GB.

Tips

Use this whenever the result should survive beyond the chat: saved notes, literature summaries, protocol drafts, audit certificates, uploaded files, and final plots. Use Compute Shell or Jupyter only for execution, analysis, and temporary computer files.

Example prompts

› What can the Workspace Files tool do, and how should I use it?
› Create /Projects/Graphene Membranes/literature-review.md with a short literature review on graphene oxide membranes for water purification
› Make a /Notes folder and write a starter README explaining how I should organize my research files

6 agent tools

Research & Discovery

Search scientific literature, patents, and the web. The AI agent iterates with refined queries across multiple databases until it has comprehensive, cited results.

Literature Review

Indexed paper review with freshness supplement

Searches Vicena’s broad scientific paper index across major journals, preprints, biomedical records, abstracts, and available full text. It extracts compact evidence packets with stable paper/excerpt pointers. When the question needs newer publications than the index may contain, it explicitly announces a scientific-only freshness check before using Web Research.

View methodology and examples

How it works

The tool extracts constraints from the request, generates targeted queries over the indexed scientific corpus, selects relevant papers, and returns compact evidence packets labeled P1, P2 and P1-E1 style excerpt pointers. For current, recent, state-of-the-art, or empty-index questions, it shows a Freshness Check explaining why an extra pass is needed, then adds a Web Research supplement restricted to scientific literature sources.

Literature Review Flow

2 modes

The tool first searches the indexed paper corpus. When freshness is needed, it explicitly explains the reason, then fills current-literature gaps with a scientific-only web supplement.

1 Understand request Capture the topic, date limits, authors, methods, materials, exclusions, and paper budget.

2 Extract constraints Use constraints to keep generated queries and selected papers from drifting off topic.

3 Define requirements State the paper evidence needed, coverage targets, and whether a freshness supplement is expected.

Indexed paper pass

Fast path

Used for the primary literature review over Vicena’s indexed scientific corpus.

1 Generate queries Create targeted scientific keyword searches for the indexed corpus.
2 Select papers Rank index hits against the user request and constraints before returning evidence.
3 Extract pointers Return compact paper packets with P1/P2 labels and P1-E1 excerpt pointers.
4 Evidence boundary Report selected paper count, evidence pointer count, freshness status, and answering rules.

Scientific freshness supplement

Contract path

Used for current, recent, state-of-the-art, or empty-index requests.

1 Explain freshness check Tell the user why an extra pass is running: the request is current/recent/state-of-the-art or the index returned no papers.
2 Scientific source gate Restrict Web Research to papers, preprints, publisher pages, PubMed/PMC/arXiv/Crossref-style records, and academic pages.
3 Reject general web Exclude news, marketing, vendor, consulting, patent analytics, product pages, and general blogs.
4 Separate evidence Treat supplement evidence as current-literature coverage distinct from indexed paper evidence.

Grounded synthesis

The chat agent answers from paper evidence pointers and accepted scientific supplement evidence only.

indexed_papersfreshness_supplement

Input

A natural language research question or topic. Be as specific as you would in a database search. You can include constraints like date ranges, organisms, or techniques.

Output

A bounded literature context with selected papers, source URLs, compact excerpts, evidence pointers, a paper evidence boundary, and a visible Freshness Check when a scientific web supplement is needed.

Limitations

The local paper index may lag very recent papers. The automatic supplement is limited to scientific literature sources; paywalled full text and newly posted papers may still be unavailable. Date filters are honored only when the request explicitly asks for a date range or recent/current literature; inferred filters are reported and ignored.

Tips

Use this for broad research questions and trend analysis. For operational protocol drafts, use Protocol Builder instead. For patents, use Patent Research.

Example prompts

› What can the Literature Review tool do, and how should I use it?
› Find recent papers on CRISPR delivery mechanisms in solid tumors
› Review the literature on metal-organic frameworks for CO2 capture

Patent Research

Indexed patent search with freshness supplement

Searches Vicena’s indexed patent corpus for patent records, assignees, inventors, publication dates, and legal technical context. When the question needs newer patent activity than the index may contain, it announces a patent-only freshness check before using Web Research.

View methodology and examples

How it works

The tool extracts constraints from the request, generates targeted queries over the indexed patent corpus, applies date and inventor filters when explicitly requested, requires indexed assignee/applicant/owner evidence for assignee-specific requests, selects relevant patent records, and returns compact evidence packets labeled with source-stable publication-number pointers such as US10932928B2-E1. For current, recent, state-of-the-art, empty-index, or unverifiable-assignee patent requests, it shows a Freshness Check and adds a Web Research supplement restricted to patent sources.

Patent Research Flow

2 modes

The tool first searches the indexed patent corpus. When freshness is needed, it explains the reason, then fills current patent-record gaps with a patent-only web supplement.

1 Understand request Capture the technology, materials, assignees, inventors, jurisdictions, patent classes, date limits, and patent budget.

2 Extract constraints Use constraints to keep generated patent queries and selected records from drifting off topic.

3 Define requirements State the patent evidence needed, coverage targets, active filters, and whether a freshness supplement is expected.

Indexed patent pass

Fast path

Used for the primary patent search over Vicena’s indexed patent corpus.

1 Generate queries Create targeted patent keyword searches for the indexed corpus.
2 Apply filters Honor explicit publication-date and inventor filters, require matching assignee metadata for assignee-specific requests, and report ignored inferred date filters.
3 Extract pointers Return compact patent packets with source-stable publication-number labels and excerpt pointers such as US10932928B2-E1.
4 Evidence boundary Report selected patent count, evidence pointer count, freshness status, and answering rules.

Patent freshness supplement

Contract path

Used for current, recent, state-of-the-art, or empty-index patent requests.

1 Explain freshness check Tell the user why an extra pass is running: the request is current/recent/state-of-the-art or the index returned no patents.
2 Patent source gate Restrict Web Research to patent-office records, Google Patents-style records, patent families, and assignee patent pages.
3 Reject general web Exclude scientific papers, blogs, news, marketing, vendor pages, consulting pages, product pages, and general web pages.
4 Separate evidence Treat supplement evidence as current patent-record coverage distinct from indexed patent evidence.

Grounded synthesis

The chat agent answers from patent evidence pointers and accepted patent supplement evidence only.

indexed_patentsfreshness_supplement

Input

A natural language description of the invention or technology. You can include specific assignees, date ranges, or patent classification codes.

Output

A bounded patent context with selected patent records, source URLs, compact excerpts, source-stable evidence pointers, date/inventor/assignee filter reporting, a patent evidence boundary, and a visible Freshness Check when a patent-only web supplement is needed.

Limitations

The local patent index may lag recently published applications and may not expose assignee metadata for every indexed patent. Assignee-specific answers require positive assignee evidence from indexed metadata or the patent-only supplement. The automatic supplement is limited to patent sources such as patent-office records, Google Patents-style pages, patent families, and assignee patent pages. It does not provide legal opinions on patentability or freedom to operate.

Tips

Use this for patent landscape analysis and patent-record discovery. For scientific papers, use Literature Review. For broad non-patent prior art, use Web Research.

Example prompts

› What can the Patent Research tool do, and how should I use it?
› Find patents on electrochemical CO2 reduction catalysts filed after 2020
› Search patents for a biodegradable polymer stent with drug elution

Protocol Builder

Evidence-grounded operational protocol drafts

Turns a scientific objective into a structured lab protocol draft: checklist, reagents, equipment, chronological steps, data analysis, troubleshooting, assumptions, gaps, and an AI-led sanity review. Helper chemistry and thermo tools provide context when the draft has enough structured inputs.

View methodology and examples

How it works

Searches indexed papers and scientific web evidence, drafts a protocol, records source-backed details separately from assumptions, and runs a sanity review. Existing chemistry helpers are used as advisory context, not automatic verdicts; the AI decides whether a tool result applies and preserves unresolved concerns.

Protocol Builder Flow

3 modes

The tool turns a scientific objective into a Nature/protocols.io-style operational draft. It separates source-backed details from assumptions and then runs an AI-led sanity review before presenting the final protocol.

1 Understand objective Capture the method, sample or substrate, target readout, constraints, scale, equipment assumptions, and safety-sensitive requirements.

2 Search evidence Read indexed papers and scientific web evidence for method details, reagent choices, operating conditions, controls, and troubleshooting signals.

3 Set draft boundary Declare that the result is an evidence-backed draft for expert review, not a validated executable protocol.

Build draft protocol

Fast path

Used to assemble the first structured protocol from the evidence and explicit assumptions.

1 Create protocol sections Produce objective, background, scope, safety, materials, equipment, before-you-begin notes, procedure, timing, QC, expected results, data analysis, troubleshooting, and evidence.
2 Label assumptions Record inferred values as assumptions instead of presenting them as extracted facts.
3 Track missing details Keep unresolved variables visible when the evidence does not support a concrete choice.
4 Reduce critical gaps Run one bounded V1 gap search for the most important missing detail, then preserve anything still unresolved.

Sanity review path

Contract path

Used to check whether the draft is internally coherent before the final answer is shown.

1 Extract check targets Identify explicit solvent/temperature, reagent-vessel, reagent-reagent, yield, stoichiometry, and biological constraint targets when the draft provides enough structure.
2 Add helper context Call chemistry or thermo helpers only when required inputs are present; otherwise record the missing input as a gap.
3 AI scientific review Use evidence, assumptions, and helper context to flag blockers, warnings, gaps, or notes.
4 Preserve concerns Surface unresolved issues in the final sanity review instead of letting the draft hide them.

Operational draft for review

The chat agent presents a scientist-facing protocol in the format: Brief, Materials, Equipment, Safety, Procedure, Timing, QC, Expected Results, Troubleshooting, and Evidence.

builder_passgap_passsanity_review

Input

A scientific objective or experimental goal. The more specific you are about the target material, organism, technique, scale, constraints, and available equipment, the better the draft.

Output

An operational protocol draft with checklist, reagents, equipment, steps, analysis, troubleshooting, source-backed details, assumptions, missing critical details, and sanity findings labeled as blockers, warnings, gaps, or notes.

Limitations

This is a draft builder, not a validated executable protocol. Helper checks have limited coverage and may miss context. A scientist must review unresolved gaps, assumptions, safety, and feasibility before any lab use.

Tips

Use this when you need to build, draft, create, write, design, or assemble a first operational protocol draft. Use Protocol Audit only after a protocol exists.

Example prompts

› What can the Protocol Builder tool do, and how should I use it?
› Build a protocol for synthesizing gold nanoparticles via citrate reduction
› Draft an operational Suzuki coupling protocol for aryl bromides

Web Research

Verified web resources and sourced facts

Searches the open web for practical lab information: technical specifications, manuals, chemical pricing, supplier pages, safety data sheets, and other sources outside papers and patents. It verifies source text before the agent answers.

View methodology and examples

How it works

Routes each request to the smallest workflow that can answer it. Resource lookups build explicit required resources, verify candidate URLs against source text, and return only verified resources. Comparisons, pricing, specifications, and broader questions use a deeper evidence contract that tracks coverage and missing fields.

Adaptive Web Research Flow

6 modes

The tool first decides whether the user needs verified resource URLs or researched facts, then runs only the path needed for that request.

1 Understand request Capture the user objective, requested resource, entities, values, and page/query budget.

2 Extract constraints Pull out vendors, models, CAS numbers, purity, date limits, deliverable type, and exclusion rules.

3 Choose path Route to a resource lookup for direct documents/URLs, or to an evidence contract for comparisons, aggregations, diagnostics, and broad research.

Resource lookup path

Fast path

Used for manuals, SDS documents, datasheets, and other requests where the answer should be verified URLs.

1 Required resources Create one required resource per requested vendor or entity, such as Sigma-Aldrich SDS for DMSO, CAS 67-68-5.
2 Search candidates Run targeted queries and read likely product, SDS, PDF, or manual pages.
3 Verify identity Accept a candidate only when source text supports the requested resource, source, subject, and CAS or model identity when requested.
4 Lookup boundary Report Required resources, Verified resources, Missing resources, and Lookup complete: Yes or No.
5 Verified URLs only The agent may list only verified resource URLs. Related but unverified pages remain discovery metadata.

Evidence research path

Contract path

Used when the answer needs facts, tables, ranges, comparisons, or coverage across several entities and fields.

1 Research contract Define coverage targets, minimum evidence, comparison fields, and quality preferences.
2 Coverage-aware search Search and select useful sources across targets before spending reads on duplicates.
3 Extract evidence Read candidates and return exact paragraph IDs, source URLs, and supporting text.
4 Evidence boundary Report covered targets, missing targets, supported evidence, and whether another pass is needed.

Grounded answer

The chat agent answers from accepted evidence only. If required resources or coverage are missing, it runs another focused pass instead of filling gaps.

resource_lookupfact_lookupaggregationcomparisondiagnosticbroad_research

Input

Any question about equipment, pricing, suppliers, technical specifications, or practical lab information that is not covered by scientific papers or patents.

Output

Verified resource links for lookup requests, or a structured answer with extracted facts, prices, or specifications. Every claim is grounded in accepted source text and cited by URL.

Limitations

Results depend on what is publicly available on the web. Paywalled content, internal company documents, and very recent pages may not be accessible.

Tips

Use this for practical lab questions: equipment specs, reagent pricing, supplier comparisons, SDS sheets. For scientific literature, use Literature Review.

Example prompts

› What can the Web Research tool do, and how should I use it?
› What is the price range of 99.9% pure titanium dioxide from lab suppliers?
› Compare specifications of benchtop centrifuges rated for 15,000 RPM

Dataset Search

Public repository dataset discovery

Searches public dataset repositories through official APIs and returns normalized candidates with source links, IDs, DOIs, public status, file counts, and direct file candidates when an API provides them. It separates fuzzy discovery from deterministic download so the agent can ask for clarification when results are ambiguous.

View methodology and examples

How it works

The agent sends a focused query to supported repository adapters such as BioStudies, Zenodo, and Figshare. Each adapter uses the repository API, applies conservative per-host pacing, parses only documented JSON fields, and returns ranked candidates without saving bytes. Follow-up download requires an explicit selected candidate, dataset_resolve result, or direct file URL.

Repository Discovery Flow

3 modes

Search official APIs first, then resolve or download only the selected dataset candidate.

1 Search official APIs Query supported repositories with conservative request pacing and no page scraping.

2 Normalize candidates Return comparable IDs, DOIs, public status, file counts, snippets, and links across repositories.

3 Select deliberately If results are ambiguous, the agent asks the user or uses surrounding evidence before resolving files.

Ready to resolve

The selected result can be passed to Dataset Resolve or Dataset Download depending on whether it is a repository page or direct file URL.

repository_searchaccession_lookupcandidate_review

Input

A dataset query, title, DOI, accession, organism, topic, or repository hint. Optional inputs select repositories, result limit, and BioStudies collection search such as transqst.

Output

A candidate list with repository, title, landing URL, API URL, DOI or accession, release date, public status, file count, snippets, and available direct file links.

Limitations

This is not a crawler and does not scrape arbitrary websites. It does not download files, bypass login walls, or silently choose among ambiguous datasets. Repository failures and rate limits are returned as warnings so the agent can continue with other sources or ask the user for a direct link.

Tips

Use Dataset Search when the user describes a dataset but does not provide a direct file URL. Use Dataset Download only after a specific candidate or file link is selected.

Example prompts

› What can the Dataset Search tool do, and how should I use it?
› Find the S-TQST115 BioStudies dataset
› Search Zenodo for the Clotho dataset

Dataset Download

Direct source files saved to Workspace

Downloads public dataset files, article supplements, repository exports, PDFs, CSVs, ZIP archives, and other HTTP(S) file URLs into Vicena Workspace Files. Supported repository adapters resolve official BioStudies, Zenodo, and Figshare metadata into file candidates or single-file downloads. It keeps provenance with source URL, final redirected URL, MIME type, size, and sha256 checksum.

View methodology and examples

How it works

The agent first uses Dataset Search, Literature Review, Patent Research, Web Research, or a repository page to identify a direct file URL or supported repository record. Repository adapters resolve BioStudies, Zenodo, and Figshare metadata into single-file downloads or multiple candidate files. The downloader then starts a background MCP job with redirects, a clear Vicena user agent, retries for transient failures, size limits, optional sha256 verification, and a durable Workspace save. It refuses relative Workspace paths, path traversal, local hosts, oversize files, and accidental overwrites unless overwrite is explicit.

Dataset Acquisition Flow

3 modes

The tool separates discovery from byte transfer: research tools find and verify the source, then Dataset Download saves exactly one file with provenance.

1 Start from a direct URL Use a verified file link from a repository, article supplement, or prior Web Research result.

2 Validate destination Require an absolute Workspace path, append the source filename for folder paths, and avoid overwriting unless requested.

3 Fetch politely Run as a background job, send a clear Vicena user agent, follow normal redirects, enforce size limits, retry transient failures, and stream bytes without crawling the host.

4 Verify and save Check the optional sha256, upload the original bytes to Workspace, and return a durable file link plus provenance metadata or a manual-upload reason.

Reusable Workspace dataset

The saved file can be opened from Files, cited in the chat answer, or moved into Compute with Workspace Compute Transfer.

direct_filesupporting_materialrepository_export

Input

A direct public HTTP(S) file URL plus an absolute Workspace destination path, such as /Datasets/study.csv or /Datasets/. Optional inputs include max_bytes, overwrite, and expected_sha256.

Output

A download job card that updates until the file is saved, fails, or asks for manual user upload. Successful jobs include a Workspace file link with source URL, final URL after redirects, MIME type, size in bytes, sha256 checksum, and a next-action field for the agent.

Limitations

This is a downloader plus explicit repository adapters, not a crawler. Unsupported HTML landing pages are rejected. It cannot bypass paywalls, login walls, robots restrictions, or rate limits. Current Workspace ingestion is capped at 50 MiB, so larger files return a manual-upload state. For multi-file repository pages, the resolver returns candidate URLs and a choose-candidate next action instead of auto-downloading.

Tips

Use Dataset Search for supported repositories, or Web Research for publisher supplements and ordinary websites, then use Dataset Download to save the verified file into Workspace.

Example prompts

› What can the Dataset Download tool do, and how should I use it?
› Download https://raw.githubusercontent.com/mwaskom/seaborn-data/master/iris.csv into /Datasets/iris.csv
› Download https://raw.githubusercontent.com/mwaskom/seaborn-data/master/penguins.csv into /Datasets/penguins.csv and report the sha256

5 agent tools

Lab Simulation & Validation

Simulate your protocol in a virtual lab before going to the bench. The AI validates each step against thermodynamic, kinetic, and chemical constraints.

🔬

Protocol Audit

Safety, yield, and feasibility check

Reviews an already described synthesis protocol or explicit procedure-safety question and flags physical impossibilities, proof gaps, and safety risks. It does not build or draft new protocols; use Protocol Builder for that.

View methodology and examples

How it works

The agent decomposes your protocol and runs deterministic checks on each step: yield validation (reactant MW vs product MW via RDKit), atom balance (conservation of mass from SMILES), chemical compatibility (curated rules + SMARTS-based classification, e.g. HF+glass, alkali+protic, chlorinated+polycarbonate), and boiling point lookup from the Thermo library. Safety data is pulled live from PubChem. Every finding cites its source. Tools return facts, not verdicts — the LLM synthesizes the final judgment.

Input

An existing protocol, procedure excerpt, or concrete safety/feasibility question with the relevant reagents, quantities, temperatures, vessels, and claimed yields where available.

Output

A report flagging impossibilities (yield > 100%, atoms from nowhere), incompatibilities (with cited rules), GHS hazards, and coverage gaps where the audit could not verify and you must apply your own judgment.

Limitations

~20 hardcoded compatibility rules (more coming via RHEACT/CAMEO). Thermo library covers ~70,000 compounds; unusual or proprietary molecules may hit coverage gaps. The audit complements your expertise, it does not replace it.

Tips

Use Protocol Builder to draft a protocol, then audit it here for physical issues before refining the procedure.

Example prompts

› What can the Protocol Audit tool do, and how should I use it?
› Check if 6.2g of naproxen methyl ester from 5g naproxen is physically possible
› Audit a Fischer esterification: refluxing 30 mL acetic acid with 50 mL ethanol and 2 mL conc. H2SO4 in a round-bottom flask at 85°C for 2 hours, expecting 40 g of ethyl acetate

🔥

Reaction Energetics

Virtual calorimeter and thermal safety

Calculates reaction enthalpy and the resulting adiabatic temperature rise. Detects thermal runaway risks, solvent boiling hazards, and exothermic safety concerns before you run the reaction.

View methodology and examples

How it works

Computes reaction enthalpy from standard formation enthalpies (Thermo database, 70,000+ compounds). Calculates adiabatic temperature rise using heat capacities of the reaction mixture. Compares the predicted temperature against solvent boiling points and known decomposition thresholds to flag runaway or boiling risks.

Input

A chemical reaction with reagent names or formulas. Optionally include quantities and solvent for adiabatic temperature rise calculations.

Output

Reaction enthalpy (kJ/mol), adiabatic temperature rise, and safety flags (exothermic warning, boiling risk, runaway risk).

Limitations

Relies on standard formation enthalpy data. Compounds not in the Thermo database cannot be analyzed. Does not model heat dissipation or cooling.

Tips

Use this before running any exothermic reaction. If the adiabatic temperature rise exceeds the solvent boiling point, you need active cooling or slower addition.

Example prompts

› What can the Reaction Energetics tool do, and how should I use it?
› Is the neutralization of concentrated H2SO4 with NaOH safe in a 500 mL flask?
› Calculate the adiabatic temperature rise for nitration of toluene

⏱️

Reaction Kinetics

Rate constants, half-lives, and timing

Estimates how long a reaction takes and how temperature affects speed. Uses the Arrhenius equation and integrated rate laws to calculate rate constants, half-lives, and time-to-completion.

View methodology and examples

How it works

Uses the Arrhenius equation (k = A * exp(-Ea/RT)) to compute or adjust rate constants at different temperatures. Applies integrated rate laws for zero, first, and second-order reactions to calculate half-lives and time to reach a target conversion.

Input

Activation energy (Ea), pre-exponential factor (A), reaction order, and temperature. Or provide a known rate at one temperature to predict rates at another.

Output

Rate constant at the specified temperature, half-life, and estimated time to reach a target conversion percentage.

Limitations

Requires known kinetic parameters (Ea and A). For reactions where these are unknown, the tool can estimate from two data points at different temperatures.

Tips

Use this to plan reaction timing and temperature optimization. Combine with Reaction Energetics to check both speed and safety at your chosen temperature.

Example prompts

› What can the Reaction Kinetics tool do, and how should I use it?
› How long does a first-order reaction with Ea=85 kJ/mol take at 60C vs 80C?
› If a first-order reaction is 50% complete in 2 hours at 25C, how fast is it at 40C with Ea=65 kJ/mol?

💧

Solubility Predictor

Solvent compatibility and dissolution

Predicts whether a compound dissolves in a given solvent using polarity matching via LogP. Ranks common lab solvents by compatibility. Use before choosing a reaction or workup solvent.

View methodology and examples

How it works

Calculates the octanol-water partition coefficient (LogP) of the solute using RDKit descriptors. Compares it against the polarity profile of common lab solvents (water, ethanol, DCM, hexane, DMSO, etc.) to rank solvents by predicted compatibility. Retrieves additional solubility data from PubChem when available.

Input

A compound name, SMILES, or CAS number. Optionally specify a solvent to check, or ask for a ranked list of common solvents.

Output

A compatibility prediction (soluble/insoluble/partial) with a ranked list of solvents from best to worst match.

Limitations

Based on LogP polarity matching, which is a heuristic. Does not account for specific solute-solvent interactions, pH effects, or temperature dependence.

Tips

Use this before choosing a reaction solvent or planning a liquid-liquid extraction. For safety data on solvents, combine with the Safety Summary tool.

Example prompts

› What can the Solubility Predictor tool do, and how should I use it?
› Will ibuprofen dissolve in water or do I need an organic solvent?
› Rank solvents for dissolving polyethylene glycol 6000

📊

Spectroscopy Predictor

Predicted IR, NMR, and MS fingerprints

Predicts the expected spectral fingerprint (IR, NMR, MS) of a compound by identifying its functional groups. Verify whether your reaction produced the correct product or determine what peaks to look for.

View methodology and examples

How it works

Parses the molecular structure with RDKit and scans it against a library of SMARTS patterns to identify functional groups. Maps each group to its characteristic spectral signatures: IR absorption bands (cm-1), expected 1H/13C NMR chemical shift ranges, and mass fragmentation patterns. Returns a predicted fingerprint you can compare against your experimental spectrum.

Input

A compound name, SMILES string, or CAS number. You can also describe a reaction product and ask what peaks to expect.

Output

Predicted IR bands (cm-1 with assignment), expected NMR chemical shifts (ppm ranges per proton environment), and major mass spec fragments (m/z).

Limitations

Predictions are based on functional group identification, not full quantum mechanical calculations. Fine structure splitting in NMR and exact fragmentation patterns may differ from experiment.

Tips

Use this to verify reaction products: compare the predicted fingerprint against your measured spectrum. For full computational spectra, use the Science Computer with PySCF.

Example prompts

› What can the Spectroscopy Predictor tool do, and how should I use it?
› What IR peaks should I expect from the product of Fischer esterification?
› Predict the NMR spectrum of aspirin

13 agent tools

Computational Chemistry

Analyze molecular structures, predict reactions, calculate properties, and assess safety. Powered by RDKit, PubChem, and neural reaction models.

Chemistry Calculator

Quantitative chemistry calculations

Deterministic calculator for stoichiometry, yields, solution preparation, pH, simple equilibria, vapor pressure, and Ideal Gas Law pressure. It returns the quantitative result, assumptions, and model boundary the agent uses when a chemistry question needs numbers and units.

View methodology and examples

How it works

The calculator keeps the agent-facing request compact, then converts it into mode-specific internal requests before execution. It preserves valid formulas, resolves common names only when unambiguous, tracks excess reagents separately from measured amounts, validates required fields, and rejects ambiguous or under-specified systems instead of guessing. ChemPy handles equation balancing and stoichiometric math; Thermo provides property correlations for supported pure-substance vapor-pressure calculations.

Chemistry Calculator Flow

6 modes

The tool turns a natural-language chemistry question into a validated calculation plan, runs deterministic calculators, and returns model assumptions and boundaries so the agent can interpret the result without inventing extra chemistry.

1 Read the calculation target Identify whether the user is asking for yield, balancing, vapor pressure, gas pressure, dilution, solution preparation, pH, or equilibrium.

2 Normalize species and units Preserve valid formulas, resolve common names only when unambiguous, convert amounts and temperatures, and keep excess reagents distinct from measured masses.

3 Validate required inputs Check mode-specific fields before execution. Missing reactants, products, constants, volumes, temperatures, or ambiguous species produce structured validation errors instead of silent guesses.

Stoichiometry engine

Stoichiometry

Used for balancing equations, limiting-reagent analysis, theoretical yield, and combustion-style product calculations.

1 Build reaction model Use validated reactants and products; infer only narrowly supported textbook products such as complete combustion products when the request clearly specifies that reaction class.
2 Balance with ChemPy Solve the symbolic equation balance. If balancing fails, return a validation error rather than assuming a 1:1 reaction.
3 Find limiting reagent Convert provided masses to moles, mark excess reagents as non-limiting, and identify the reagent that constrains reaction extent.
4 Calculate target amount Return requested product yields or required reactant amounts with moles, grams, the balanced equation, and gas-generation boundary notes when relevant.

Physical chemistry engines

Phys chem

Used for vapor pressure, ideal-gas pressure, solution preparation, dilution, pH, and simple equilibrium questions.

1 Choose calculation mode Dispatch to vapor_pressure, pressure_volume, solution_prep, pH, or equilibrium based on the typed validated request.
2 Resolve constants Use property correlations or user-supplied constants when the model is in scope; reject systems where the needed constant or species identity is missing.
3 Solve with units Compute the numeric result in chemistry-native units such as grams, moles, molarity, atm, psi, mmHg, kPa, or pH.
4 Report boundaries State assumptions such as pure-substance vapor-liquid equilibrium or ideal-gas behavior, and distinguish those from sealed-vessel safety or protocol feasibility.

Agent interpretation boundary

Boundary

The calculator provides quantitative evidence; the final agent may add qualitative scientific interpretation but must not invent additional quantitative or safety conclusions.

1 Allowed interpretation The agent can explain reaction class, excess-reagent assumptions, theoretical-vs-isolated yield, equilibrium partial pressure, exothermic reaction class, and why gas generation may matter.
2 Needs another tool Heat released, adiabatic temperature rise, closed-vessel pressure beyond supplied ideal-gas inputs, flammability risk, reaction rate, toxicity, compatibility, and protocol feasibility require a dedicated Vicena tool.
3 Next-step routing If the user asks for one of those follow-up analyses, the agent should call or recommend Reaction Energetics, Reaction Kinetics, Safety Summary, Protocol Audit, or a related tool with the missing inputs.

Calculation plus bounded interpretation

The chat agent answers from the calculator result, adds supported qualitative context, and routes unsupported quantitative or safety claims to the right tool instead of treating calculator output as a full simulation.

stoichiometryvapor_pressurepressure_volumesolution_preppHequilibrium

Input

A quantitative chemistry question, optional reagent amounts, product species, solution targets, equilibrium constants, or temperature and gas conditions.

Output

Balanced equation, limiting reagent, reaction extent, theoretical yield, required solution mass, pH, equilibrium result, vapor pressure, or ideal-gas pressure with units, assumptions, and validation errors for corrected follow-up calls.

Limitations

Requires calculation-safe species identities and known constants for equilibrium systems. It does not estimate heat release, adiabatic temperature rise, vessel compatibility, reaction timing, toxicity, protocol feasibility, or real experimental yield; those require dedicated tools or additional inputs.

Tips

Use this for routine quantitative chemistry. For thermal safety use Reaction Energetics, for timing use Reaction Kinetics, and for protocol or hazard conclusions use the relevant audit or safety tool.

Example prompts

› What can the Chemistry Calculator tool do, and how should I use it?
› Balance the combustion of propane and calculate the yield from 50g
› How many grams of NaCl do I get from 10g NaOH and excess HCl?

Chemical Intelligence

Compound identity, properties, hazards, and fingerprints

Builds a deterministic compound fact pack from PubChem, RDKit, Thermo, and SMARTS patterns. It resolves one compound, computes descriptors, checks functional groups, summarizes available GHS hazards, estimates solvent compatibility, predicts analytical fingerprints, and compares molecular similarity when references are provided.

View methodology and examples

How it works

The MCP tool keeps the public interface compact, then normalizes the request into a single compound identity and optional fact sections. It preserves explicit SMILES, rejects ambiguous bare SMILES-like inputs, resolves common names or CAS numbers through PubChem when available, computes local molecular descriptors with RDKit, uses SMARTS for functional groups and spectroscopy hints, asks PubChem for GHS records, uses Thermo for available pure-compound physical properties, and reports data gaps instead of guessing.

Chemical Intelligence Flow

6 modes

The tool resolves one compound, builds deterministic fact sections, and exposes assumptions and data gaps so the chat agent can explain the chemistry without inventing unsupported facts.

1 Resolve identity Accept an explicit compound field or extract a single compound from the request. Explicit SMILES is preserved, names and CAS numbers are resolved through PubChem, and ambiguous bare SMILES-like inputs are rejected.

2 Compute local descriptors Parse the resolved structure with RDKit to calculate formula, exact mass, molecular weight, LogP, TPSA, H-bond counts, rotatable bonds, and Lipinski flags.

3 Build fact sections Run SMARTS functional-group matching, PubChem GHS lookup, Thermo physical-property lookup, LogP solvent ranking, spectroscopy hints, and optional Morgan-fingerprint similarity.

4 Report boundaries Return provenance, assumptions, model boundaries, and data gaps so the final answer can distinguish verified facts from interpretation.

Grounded agent interpretation

The agent uses the fact pack as evidence for qualitative explanation, and routes unsupported quantitative safety or protocol claims to dedicated tools.

identitypropertieshazardssolubilityspectroscopysimilarity

Input

A compound-level request with a compound name, CAS number, or explicit SMILES. Optional fields can request sections, solvents for focused solubility guidance, or reference compounds for similarity.

Output

A grounded fact pack covering identity, identifiers, RDKit properties, Thermo physical properties when available, functional groups, PubChem/GHS hazards, LogP-based solubility guidance, SMARTS-based spectroscopy hints, similarity scores, assumptions, provenance, and data gaps.

Limitations

This is a fact-pack tool, not a protocol audit, toxicity model, measured solubility database, or full spectral simulator. Hazard records may be missing from PubChem, solubility is estimated from polarity/LogP, and spectra are functional-group hints rather than measured or quantum-calculated spectra.

Tips

Use this before asking the agent to reason about an unfamiliar compound. For numeric lab math use Chemistry Calculator, for reaction heat use Reaction Energetics, and for kinetics use Reaction Kinetics.

Example prompts

› What can the Chemical Intelligence tool do, and how should I use it?
› Build a chemical intelligence fact pack for aspirin
› Analyze caffeine and show solubility guidance for water and ethanol

Reaction Intelligence

Reaction prediction, feasibility, compatibility, energetics, and kinetics

Builds a reaction-level fact pack that interprets a reaction, checks balance and atom conservation, predicts products when SMILES inputs are available, suggests retrosynthesis routes for targets, screens compatibility, and runs energetics or kinetics sections when the required inputs are supplied.

View methodology and examples

How it works

The MCP tool keeps a compact public schema, normalizes the request into reaction sections, and calls deterministic or model-backed helpers only when the required inputs are present. ChemPy balances formula equations, RDKit checks atom conservation and yield feasibility, SMARTS rules handle simple textbook reactions, ReactionT5v2 handles complex prediction and retrosynthesis, curated compatibility rules screen reagent and vessel pairs, and Thermo/Arrhenius engines are used for thermal or timing sections.

Reaction Intelligence Flow

6 modes

The tool turns a reaction request into bounded evidence sections so the chat agent can interpret the chemistry without treating model suggestions or limited checks as final verdicts.

1 Normalize reaction request Extract reaction text, reactants, products, target molecule, solvent, vessel, amounts, conditions, and requested analysis sections.

2 Run scoped helpers Call balance, atom conservation, simple reaction rules, neural prediction, retrosynthesis, compatibility, yield, energetics, or kinetics helpers only when their required inputs are present.

3 Preserve gaps Return missing inputs and model boundaries explicitly instead of guessing products, rates, heat release, route feasibility, or safety.

4 Ground final answer Expose provenance and recommended next tools so the final answer can distinguish verified checks from qualitative interpretation.

predictionretrosynthesisbalancecompatibilityenergeticskinetics

Input

A reaction-level request with optional reaction equation, reactants, products, target molecule, solvent, vessel, conditions, amounts, and requested sections.

Output

A structured reaction fact pack with interpretation, prediction or route suggestions, balance checks, yield feasibility, compatibility checks, energetics or kinetics output, assumptions, boundaries, data gaps, and provenance.

Limitations

This is not an executable protocol or proof of safety. Neural predictions and retrosynthesis are suggestions; compatibility rules have limited coverage; energetics is screening-level; kinetics requires explicit rate constants or Arrhenius parameters; yield checks only bound physical feasibility.

Tips

Use this for reaction-level grounding before the agent explains feasibility. Use Chemistry Calculator for routine numeric stoichiometry and Protocol Audit for procedure-level review.

Example prompts

› What can the Reaction Intelligence tool do, and how should I use it?
› Build a reaction intelligence fact pack for ethanol plus acetic acid
› Check atom balance and glass compatibility for ethanol plus acetic acid forming ethyl acetate and water

🧪

Chemical Analyzer

Properties and hazards lookup

Retrieves physical properties and safety data for chemicals from PubChem. Returns boiling points, molecular weights, GHS hazard classifications, and reactive group information.

View methodology and examples

How it works

Queries the PubChem REST API (116 million compounds) by name, CAS number, or SMILES. Returns physical constants (melting point, boiling point, density, molecular weight), GHS hazard pictograms, H-statements, and reactive group classifications.

Example prompts

› What can the Chemical Analyzer tool do, and how should I use it?
› What are the physical properties and hazards of dimethyl sulfoxide?
› Look up the boiling point and GHS classification of acetonitrile

🔬

Reaction Predictor

Neural network reaction prediction

Predicts products of complex chemical reactions using a trained neural network. Optimized for multi-step and pharmaceutical reactions like Suzuki coupling, Heck reactions, and Sonogashira coupling. ~93% accuracy on complex patent reactions.

View methodology and examples

How it works

Converts reactants to SMILES notation and feeds them through a Molecular Transformer, a sequence-to-sequence neural network trained on 1.2 million patent reactions. The model treats reactions as translations between molecular languages. Predicted products are validated for chemical validity using RDKit before being returned.

Input

Reactant names, SMILES strings, or a natural language description of the reaction. You can specify catalysts and conditions.

Output

Predicted product(s) as SMILES with names and a confidence indicator. The product is validated for chemical correctness.

Limitations

Optimized for organic synthesis. Less reliable for inorganic, enzymatic, or radical reactions. Accuracy is ~93% on complex patent-style reactions, lower on unusual chemistries.

Tips

Use this for complex, multi-step reactions. For simple textbook reactions (acid-base, esterification), use Simple Reactions instead for higher accuracy.

Example prompts

› What can the Reaction Predictor tool do, and how should I use it?
› Predict the product of Suzuki coupling between phenylboronic acid and 4-bromoanisole
› Predict the product of a Heck reaction with styrene and iodobenzene

🔬

Retrosynthesis

Work backwards from target molecule

Suggests reactants needed to synthesize a target molecule. Proposes synthetic routes by working backwards from your desired product, identifying feasible starting materials and reaction conditions.

View methodology and examples

How it works

Takes a target molecule in SMILES format and runs it through the Molecular Transformer in reverse mode. The model proposes disconnections and suggests commercially available starting materials. Multiple synthetic routes are ranked by feasibility.

Input

A target molecule as a name, SMILES string, or drawn structure.

Output

One or more proposed synthetic routes with starting materials and reaction types. Routes are ranked by feasibility.

Limitations

Works best for drug-like organic molecules. Very large molecules (polymers) or inorganic compounds may produce unreliable suggestions.

Tips

Use this when you know what you want to make but not how to make it. Combine with the Reaction Predictor to validate each proposed step.

Example prompts

› What can the Retrosynthesis tool do, and how should I use it?
› How can I synthesize ibuprofen from simple starting materials?
› Suggest a retrosynthetic route to paracetamol

🔬

Simple Reactions

Textbook reaction predictions

Predicts products of common textbook reactions: esterification, acid-base neutralization, SN2 substitution, hydrolysis, and oxidation/reduction of simple substrates.

View methodology and examples

How it works

Uses rule-based reaction templates implemented in RDKit for well-characterized reaction types. Unlike the neural predictor, this is deterministic and optimized for simple, well-understood transformations where accuracy is near 100%.

Input

Reactant names or formulas and the reaction type (e.g. "esterification", "neutralization", "SN2").

Output

The predicted product with a balanced equation.

Limitations

Only covers well-known textbook reaction types. For complex or novel reactions, use the Reaction Predictor.

Tips

Use this for common organic and inorganic reactions where accuracy matters more than novelty. The rule-based approach is deterministic, so identical inputs always give identical outputs.

Example prompts

› What can the Simple Reactions tool do, and how should I use it?
› What is the product of ethanol and acetic acid with an acid catalyst?
› Predict the product of SN2 reaction between NaBr and 1-chlorobutane

🔬

Functional Group Identifier

SMARTS substructure matching

Identifies functional groups present in a molecule from its SMILES representation. Uses SMARTS pattern matching to scan for common organic functional groups.

View methodology and examples

How it works

Parses the input molecule with RDKit and runs it against a curated library of SMARTS patterns covering 60+ functional groups: alcohols, amines, carbonyls, halogens, heterocycles, and more. Reports all matches with their positions in the molecule.

Input

A molecule name, SMILES string, or CAS number.

Output

A list of all functional groups found in the molecule, with their names and positions.

Limitations

Identifies standard organic functional groups. Unusual or very complex heterocyclic motifs may not be covered.

Tips

Use this to understand an unfamiliar molecule before running other tools. Knowing the functional groups helps predict reactivity, solubility, and spectral features.

Example prompts

› What can the Functional Group Identifier tool do, and how should I use it?
› What functional groups are present in aspirin?
› Does SMILES CC(=O)Nc1ccc(O)cc1 contain any amine groups?

🔬

Molecular Descriptors

Drug-likeness and Lipinski analysis

Calculates molecular descriptors and checks Lipinski's Rule of Five for drug-likeness. Returns LogP, molecular weight, hydrogen bond donors/acceptors, and polar surface area.

View methodology and examples

How it works

Computes 2D molecular descriptors using RDKit: LogP (Wildman-Crippen), molecular weight, number of hydrogen bond donors and acceptors, topological polar surface area, and rotatable bond count. Evaluates Lipinski's Rule of Five and Veber's rules to assess oral bioavailability.

Input

A molecule name, SMILES string, or CAS number.

Output

A table of descriptors (LogP, MW, HBD, HBA, TPSA, rotatable bonds) and a pass/fail assessment against Lipinski and Veber rules.

Limitations

Evaluates drug-likeness based on physicochemical properties only. Does not predict biological activity, toxicity, or metabolic stability.

Tips

Use this early in drug design to filter candidates. Molecules that fail Lipinski's rules are unlikely to be orally bioavailable.

Example prompts

› What can the Molecular Descriptors tool do, and how should I use it?
› Does caffeine pass Lipinski's Rule of Five?
› Is aspirin drug-like? CC(=O)Oc1ccccc1C(=O)O

🔬

Molecular Similarity

Tanimoto fingerprint comparison

Calculates structural similarity between two molecules using Morgan fingerprints and Tanimoto coefficient. Returns a score between 0 (completely different) and 1 (identical).

View methodology and examples

How it works

Generates Morgan circular fingerprints (radius 2, 2048 bits) for each molecule using RDKit. Computes the Tanimoto coefficient (intersection over union of bit vectors) to quantify structural similarity. This is the same method used in pharmaceutical virtual screening.

Input

Two molecule names, SMILES strings, or CAS numbers to compare.

Output

A Tanimoto similarity score between 0.0 (completely different) and 1.0 (identical), with a qualitative assessment.

Limitations

Measures 2D structural similarity only. Molecules with similar shapes but different connectivity (3D similarity) are not captured.

Tips

Scores above 0.85 generally indicate very similar molecules. Use this to find structural analogs or check if two molecules are variants of the same scaffold.

Example prompts

› What can the Molecular Similarity tool do, and how should I use it?
› How similar are ibuprofen and naproxen?
› Compare the structures of caffeine and theobromine

🔬

Safety Summary

GHS hazard information

Retrieves GHS safety information from PubChem for any chemical. Returns hazard statements, signal words, and pictogram descriptions. Accepts chemical names, SMILES, or CAS numbers.

View methodology and examples

How it works

Queries PubChem's GHS classification data. Returns the signal word (Danger/Warning), all H-statements (hazard), P-statements (precaution), and pictogram codes. Accepts input as chemical name, SMILES, or CAS number.

Input

A chemical name, SMILES string, or CAS number.

Output

Signal word (Danger/Warning), GHS pictograms, hazard statements (H-codes), and precautionary statements (P-codes).

Limitations

Returns the GHS classification from PubChem. Compounds not registered in PubChem will not have data. Does not assess mixture hazards.

Tips

Always check safety before handling a new reagent. Combine with the Solubility Predictor to assess both compatibility and safety of your chosen solvents.

Example prompts

› What can the Safety Summary tool do, and how should I use it?
› What are the safety hazards of hydrofluoric acid?
› Give me the GHS classification for sodium cyanide

🔬

Molecule Converters

Name, SMILES, and CAS conversion

Converts between molecule representations: common names to SMILES, SMILES to names, and molecules to CAS registry numbers. Uses PubChem as the reference database.

View methodology and examples

How it works

Resolves chemical identifiers through PubChem's standardization pipeline. Converts between IUPAC names, common names, CAS registry numbers, and SMILES/InChI representations. Handles synonyms and trade names.

Input

A molecule in any format: common name, IUPAC name, CAS number, or SMILES string.

Output

The molecule in the requested format (SMILES, name, or CAS), with the PubChem CID for reference.

Limitations

Relies on PubChem's compound registry. Very new or proprietary compounds may not be found.

Tips

Use this to translate between formats when other tools require SMILES input. Most chemistry tools in Vicena also accept names directly, but SMILES is unambiguous.

Example prompts

› What can the Molecule Converters tool do, and how should I use it?
› What is the SMILES for ibuprofen?
› Convert CAS 50-78-2 to a chemical name

⚛️

Computational Chemistry

Advanced molecular modeling and notebook workflows

Runs local molecular analysis in the Science Computer and routes supported advanced molecular modeling to managed Rowan workflows. Inputs, workflow state, raw results, notebooks, and scientific limits remain visible.

View methodology and examples

How it works

Vicena uses Deep Work with the Science Computer for bounded local work and routes eligible advanced jobs through the managed Rowan gateway. It keeps inputs, workflow UUIDs, budgets, raw files, tables, plots, structures, and interpretation visible.

Input

A scientific goal with a molecule, SMILES, protein, paper, assay condition, or property you want to compare or compute.

Output

A notebook and persisted outputs such as raw results, CSV/JSON files, plots, molecular structures, scores, and interpretation.

Limitations

Compute credits, per-job limits, available Rowan workflows, sandbox packages, and account permissions apply. Start with lightweight descriptors or setup notebooks before expensive docking, MD, or batch runs.

Tips

Start from the scientific decision you need to make. Vicena can prepare the notebook, keep workflow UUIDs and raw outputs, and help iterate without you having to babysit infrastructure.

Example prompts

› What can Computational Chemistry do, and how should I use it?
› Compare aspirin, ibuprofen, and naproxen using descriptors and explain which differences matter
› Prepare a conformer search for aspirin from SMILES CC(=O)Oc1ccccc1C(=O)O and summarize the lowest-energy structures

4 agent tools

Science Computer

A persistent cloud computer with JupyterLab where you and the AI agent collaborate in real notebooks. Write code, run simulations, and visualize results together.

💻

Compute Shell

Terminal for packages, quick commands, and inline calculations

Run shell commands in an isolated Linux environment. Use it to install packages, inspect the compute environment, manipulate sandbox files, run quick calculations, and execute small scripts with inline output.

View methodology and examples

How it works

Each user gets a dedicated Linux container with sandbox storage. The agent runs shell commands directly and reads stdout/stderr. Packages installed via pip or apt persist inside the Science Computer. The container is isolated from other users and separate from Workspace Files.

Input

Natural language instructions for what you want to install, inspect, calculate, or run quickly. You can also paste shell commands directly.

Output

The command output (stdout/stderr) displayed inline in chat. Files created by commands remain in the Science Computer filesystem, not Workspace Files.

Limitations

CPU-only environment (no GPU). Long-running computations may time out after 2 minutes per command. Network access is available for downloading packages and data. Direct transfer between Science Computer files and Workspace Files is not available yet.

Tips

Use this for package installation, quick one-off commands, and short inline computations. For iterative computation with plots and tables, use Jupyter instead.

Example prompts

› What can the Compute Shell tool do, and how should I use it?
› Run Python to generate a sine-wave plot from synthetic data and display the image
› Use RDKit to calculate the molecular weight and LogP of caffeine from SMILES Cn1cnc2c1c(=O)n(C)c(=O)n2C

🖥️

Workspace State

Read-only visible UI state

Read the current known user-facing workspace state without changing anything. It reports which Jupyter notebook the UI is showing, which notebook the notebook tool last touched, and the previous visible notebook when known.

View methodology and examples

How it works

The tool reads canonical state recorded by the app backend from JupyterLab UI events and notebook-tool actions. It never opens notebooks, switches tabs, lists files, runs code, or mutates workspace state.

Input

Ask about the current visible notebook, previous visible notebook, or ambiguous phrases like current notebook and active tab.

Output

A compact JSON state snapshot with known surfaces such as Jupyter visible_notebook, previous_visible_notebook, and tool_active_notebook.

Limitations

Returns only state the app has observed from the browser and tool calls. Unknown fields mean that surface has not reported state yet.

Tips

Use this before acting on ambiguous UI-relative requests. Use Jupyter for actual notebook edits or execution.

Example prompts

› What can the Workspace State tool do, and how should I use it?
› Which notebook am I looking at right now?
› Before appending to a notebook, check the current visible workspace state

Jupyter

Visible Python notebooks for simulations, analysis, and plots

Work in a shared JupyterLab notebook when the code, plots, and outputs should be visible, editable, and reusable. The agent can create, inspect, edit, and run cells while returning compact transcripts and displayed images.

View methodology and examples

How it works

Runs a real JupyterLab instance in the Science Computer. The agent uses a small tool family: Jupyter Manager for notebook lifecycle and whole-notebook reads, Jupyter Editor for cell edits, and Jupyter Runner for execution and output inspection. Named notebooks become tool-active for later calls. If the browser reports a newer UI-active notebook, pathless writes fail as ambiguous instead of guessing; choose notebook_path, notebook_target="tool_active", or notebook_target="ui_active".

Notebook Workflow

6 modes

Use notebooks for visible computational work. Use Compute Shell only for support tasks such as package installs, file checks, or short one-off commands.

1 Choose the notebook Create a named notebook, pass notebook_path, or choose notebook_target when the UI-active and tool-active notebooks differ.

2 Read or edit cells Append, insert, replace, edit, delete, move, read one cell, or read the whole notebook. Reads and edits never execute code.

3 Run intentionally Run one existing cell or all cells. The transcript includes execution counts, text, errors, and image markers.

4 Keep the record clean Continue by appending new work, replacing the relevant cell, or reading user-side edits before changing them.

New notebook

create_notebook

Best when the user asks for a fresh notebook with a known structure. It never overwrites an existing file.

1 Provide cells Pass notebook_path and an ordered cells list. Then run the saved code cells when outputs are needed.
2 Return the transcript Report the path, cell counts, source, execution counts, outputs, and any displayed images.

Continue a notebook

active target

Best when the user says continue, append, current notebook, or active notebook.

1 Resolve the target Omit notebook_path only when there is no newer UI-active notebook. Otherwise choose tool_active, ui_active, or a named path.
2 Append or edit directly Use append_cell for new work, replace_cell for a known bad cell, or edit_cell for a small source change, then run the relevant cell.

Inspect existing work

read actions

Best when the user asks what is in the notebook, what changed, or what the last output shows.

1 Read without running Use read_notebook or read_cell. Set include_outputs=true when outputs or images matter.
2 Answer from state Use the returned cell source, text output, errors, tables, and displayed images. Do not infer from filenames.

Visible computational record

The user sees a live notebook with cells and outputs inside the Science Computer session folder. Notebook files are separate from Workspace Files.

opencreatereadwrite cellsrunreset

Input

Requests for visible Python work: create a notebook, build a simulation, analyze data, plot results, append a cell, replace a known cell, run all cells, or inspect the last displayed notebook image.

Output

A live Jupyter notebook plus a concise transcript with the action, path, cell counts, changed or read source, execution counts, text/error output, and image markers. Displayed notebook images are attached to the tool result.

Limitations

CPU-only environment. Very large datasets (>1 GB) and GPU training are not supported. Notebook files live in the Science Computer session folder, not Workspace Files. Direct transfer between notebooks and Workspace Files is not available yet. Destructive changes must be explicit: create_notebook never overwrites, delete_cells requires indices, and replace_notebook requires cells.

Tips

Use this for reusable multi-step computation with plots, tables, and code the user may edit. Use read_notebook or read_cell before modifying user-side edits. Use Compute Shell for package installs, quick file checks, and support commands that should not become notebook cells.

Example prompts

› What can the Jupyter tool do, and how should I use it?
› [USE JUPYTER] Build a notebook that fits this Michaelis-Menten dataset and reports Km, Vmax, residuals, and a plot: substrate_uM=[5,10,20,50,100,200], rate=[0.08,0.15,0.27,0.50,0.72,0.88].
› [USE JUPYTER] Create a notebook that analyzes this calibration curve, fits a linear model, reports R^2 and LOD, and plots the fit with residuals: concentration=[0,1,2,5,10], signal=[0.03,0.18,0.35,0.82,1.61].

📁

Computer File System

Temporary sandbox file operations

Read, write, and manage files inside the active compute sandbox or Jupyter-backed working directory. This is separate from the Files sidebar and Workspace Files.

View methodology and examples

How it works

Provides structured filesystem access within the compute environment. The agent can read, write, move, delete, and search sandbox files when a task explicitly involves a filesystem path, uploaded dataset, notebook output, or project source file.

Input

Natural language instructions for sandbox file operations, or specific computer/Jupyter file paths to read or write.

Output

File contents, directory listings, or confirmation of write/move/delete operations.

Limitations

This is not the Vicena Workspace Files filesystem. Direct copy/import/export/move between the Science Computer and Workspace Files is not available yet. For notes, protocol certificates, saved canvas files, or vague requests like “create a new file,” use the Files workspace instead.

Tips

Use this for data files used by computation. Use workspace files for documents you want to manage from the Vicena Files sidebar.

Example prompts

› What can the Computer File System tool do, and how should I use it?
› Create /tmp/example.csv with columns time,temperature and three rows, then read it back and summarize the columns
› List files in the current Jupyter working directory

Science Stack

42 databases, libraries & models

The databases, libraries, and models that power Vicena's tools. All are available in the Scientific Simulations notebook environment for direct use in your notebooks.

Library

by Project Jupyter

The open-source interactive development environment used by millions of scientists. Originally developed at UC Berkeley, Jupyter is the standard for reproducible computational research across all scientific disciplines.

Official project

Model

Describe your research problem and the agent selects the right tools automatically. Most tools are free, no credit card required.

Try it free

The scientific tools behind real work.

What kind of scientific work do you want to do?

Computational Modeling & Simulation

Research Tools, Workspace & Foundations

CFD & Aerodynamics

OpenFOAM

snappyHexMesh

trimesh

SU2

Optical Simulation

MCX / pmcx

Meep

MPB

Mitsuba

pymoo

SALib

Semiconductor Design

Icarus Verilog

Verilator

Yosys

OpenROAD

KLayout

DEVSIM

Circuit Design

ngspice

DEVSIM

NumPy / SciPy

JupyterLab

Computational Chemistry & Molecular Dynamics

Rowan

AIMNet2 / fairchem

xTB / CREST

PySCF / GPU4PySCF

Psi4

Quantum ESPRESSO

OpenFold3

Boltz

Chai-1r

ColabFold / MMseqs2

OpenMM / OpenFF

Vina / QVina2 / Gnina

TMD

geomeTRIC / Sella / ASE

RDKit

Workspace

Workspace Files

How it works

Workspace File Flow

Empty workspace

Existing workspace

Computer boundary

Input

Output

Limitations

Tips

Example prompts

Research & Discovery

Literature Review

How it works

Literature Review Flow

Indexed paper pass

Scientific freshness supplement

Input

Output

Limitations

Tips

Example prompts

Patent Research

How it works

Patent Research Flow

Indexed patent pass

Patent freshness supplement

Input

Output

Limitations

Tips

Example prompts

Protocol Builder

How it works

Protocol Builder Flow