Mirrorthink
Mirrorthink (formerly MirrorThink) is an AI for scientific research with virtual lab simulations, literature search, protocol building, and chemistry tools. Used by 1000+ institutes.
Last verified:
What is Mirrorthink?
Mirrorthink is an autonomous AI research agent built specifically for scientists and engineers in the science and engineering sectors. It turns a research question into literature search results, experimental protocol drafts, and computational analysis steps in one unified workspace. The AI agent reasons about your intent, selects the right tool from a catalog of 24 scientific tools, executes it, and iterates until it produces a verified, cited answer.
Key features include automated literature review across PubMed, arXiv, Google Scholar, and Google Patents with adaptive query refinement and full citation chains; a Protocol Builder that extracts experimental variables from papers with confidence scoring; virtual lab simulation that validates protocols against thermodynamic, kinetic, and chemical constraints before bench work; an agentic JupyterLab environment with 30+ pre-installed scientific packages (NumPy, SciPy, RDKit, PySCF, ASE); and a Chemistry Lab with 20+ computational chemistry tools for reactions, properties, spectra, solubility, and quantum chemistry including DFT, Hartree-Fock, and MP2 calculations.
Mirrorthink is designed for researchers at universities (MIT, Stanford, Harvard, Oxford, Cambridge, Caltech), scientists at 1000+ institutes worldwide, PhD students, computational chemists, chemical engineers, and anyone doing real science who needs AI that grounds answers in physical constraints like thermodynamics, kinetics, solubility, and molecular structure instead of hallucinating.
The platform introduces Workspace Files in version 3.0 that save protocols, audits, PDFs, plots, data files, and notes as durable files with stable paths, plus long-running chat that queues follow-up prompts and keeps Files, Computer, and Tools open in side tabs. Every finding links to its original source with DOIs, authors, publication dates, and patent numbers, and when a tool has no data it explicitly says so instead of faking safety.
Mirrorthink pricing
Pricing model: Free
Freemium pricing model with a free tier that includes literature review, patent search, computational chemistry, and persistent JupyterLab computer for trying the tool. Paid plans start from $5/month with monthly billing frequency. Paid options provide expanded usage capacity for research workloads beyond free tier limits.
Mirrorthink pros
- Autonomous AI agent that reasons through research questions and selects tools dynamically
- Up to 10 reasoning cycles per question with scratchpad tracking goals and findings
- Multi-source literature search across PubMed, arXiv, Google Scholar, and Google Patents in parallel
- Adaptive query refinement that reformulates queries until coverage is comprehensive
- Full citation chains with DOIs, authors, publication dates, and patent numbers for every finding
- Protocol Builder extracts variables by variable with confidence scoring from independent sources
- Virtual lab simulation validates thermodynamics, kinetics, solubility before bench work
- Agentic JupyterLab writes cells, runs code, fixes errors, shows every step live
- 30+ scientific packages pre-installed including NumPy, SciPy, RDKit, PySCF, ASE
- Persistent cloud environment where notebooks, data, and packages survive across chats
- 20+ computational chemistry tools for reactions, properties, descriptors, spectra, retrosynthesis
- Quantum chemistry with DFT, Hartree-Fock, and MP2 calculations via PySCF
- Protocol Audit catches impossible reactions, validates yield and atom balance, checks GHS hazards
- Physics and chemistry aware answers grounded in real thermodynamic and kinetic constraints
- Workspace Files give protocols, audits, plots, and notes durable stable paths
- ReactionT5v2 neural predictor for patent-style coupling reactions
- Evidence tools expose assumptions, provenance, gaps, and structured outputs agent can reason over
- Variable-by-variable extraction for temperatures, concentrations, durations, reagents cross-referenced independently
- Solvent compatibility and precipitation risks flagged before starting experiments
- Used by researchers at MIT, Stanford, Harvard, Oxford, Cambridge, Caltech and 1000+ institutes
Mirrorthink cons
- New feature rough on edges for complex workflows like computational fluid dynamics
- Limited to science and engineering sectors not general purpose AI
- Requires internet connection for cloud JupyterLab and database searches
- Quantum chemistry calculations may be slow for large molecular systems
- Only 24 tools in catalog compared to broader AI platforms with more integrations
- Free tier has limited usage capacity for heavy research workloads
- Protocol extraction depends on literature quality may miss unpublished methods
- Virtual lab simulation estimates not experimental guarantees
- Persistent environment means errors or bad packages stay unless manually cleaned
- Learning curve for scientists new to JupyterLab and computational chemistry tools
Frequently asked questions about Mirrorthink
What is Vicena and what does it do?
Vicena is an autonomous AI research agent designed specifically for the science and engineering sectors. It turns a research question into literature search results, experimental protocol drafts, and computational analysis steps in one workspace. The AI agent reasons about your intent, selects the right tool from a catalog of 24 scientific tools, executes them, and iterates until it produces a verified, cited answer grounded in real physics and chemistry constraints.
Which databases does Vicena search for literature?
Vicena searches PubMed, arXiv, Google Scholar, Google Patents, and the open web in parallel. The AI agent evaluates results and reformulates queries adaptively until coverage is comprehensive, then links every finding to its original source with DOIs, authors, publication dates, and patent numbers for full citation chains.
What is the agentic JupyterLab feature?
The agentic JupyterLab gives Vicena a persistent cloud JupyterLab environment for scientific Python. The agent builds notebooks cell by cell, writes code, checks output, fixes errors, and refines while you see every step live. It comes with 30+ scientific packages pre-installed including NumPy, SciPy, matplotlib, RDKit, PySCF, and ASE, and your notebooks, data, and installed packages persist across chats.
How does the virtual lab simulation work?
Virtual lab simulation validates your protocol in a virtual lab before going to the bench. The AI validates each step against thermodynamic constraints including energy balances, equilibrium constants, and Gibbs free energy; kinetic feasibility including reaction rates, activation energies, and time-to-completion estimated from literature; and solubility and compatibility including solvent compatibility, miscibility, and precipitation risks flagged before you start.
What computational chemistry tools are available?
Vicena includes 20+ computational chemistry tools for reactions, properties, descriptors, spectra, solubility, and retrosynthesis. This includes quantum chemistry with DFT, Hartree-Fock, and MP2 calculations using PySCF where the AI picks the method and basis set, writes the code, runs it, and returns results. There is also a ReactionT5v2 neural predictor for patent-style coupling reactions.
What is the Protocol Audit feature?
Protocol Audit validates synthesis protocols before the bench by checking yield, atom balance, reagent-vessel compatibility, boiling points, and GHS hazards. It catches impossibilities before you run them, and every finding cites its source. When a tool has no data it says so instead of faking safety, and PubChem and RDKit integration catches impossible reactions and validates molecular structures.
How does Vicena extract experimental protocols from papers?
When you describe the experiment you want to reproduce, the AI agent extracts every variable from the literature delivering a complete protocol where nothing is hallucinated. It does variable-by-variable extraction for temperatures, concentrations, durations, and reagents with each parameter found and cross-referenced independently. Every value gets a confidence score by how many independent sources agree so you know what to trust.
What is new in Vicena version 3.0?
Version 3.0 turns Vicena from single chat into a working research environment with three key features: Workspace Files that save protocols, audits, uploads, plots, and notes as normal files with stable paths; targetable state-aware Jupyter work organized around the notebook you are using with compact transcripts and visible UI state; and long-running chat that queues follow-up prompts, jumps to latest result, and keeps Files, Computer, and Tools open in side tabs.
Is Vicena free to use?
Yes, Vicena offers a free tier that includes literature review, patent search, computational chemistry, and a persistent JupyterLab computer. You can describe your research problem and the agent gets to work without paying. Paid plans start from $5/month for expanded usage capacity if you need more than the free tier provides.
Who uses Vicena and which institutes are on board?
Vicena is used by researchers at MIT, Stanford, Harvard, Oxford, Cambridge, Caltech, and 1000+ institutes worldwide. It is designed for scientists, students, PhD researchers, computational chemists, chemical engineers, and anyone in the science and engineering sectors who needs AI grounded in real physical constraints rather than general purpose chat.