One agent.
Every tool you trust.
Assemble your dream research agent à la carte — from the best open-source scientific tools, wrapped as skills, connected over MCP, and run inside your own walls. We're the orchestration layer and the science expertise. The ecosystem does the rest.
We don't reinvent the wheel. We connect them.
See how it worksThe open-source world has already built extraordinary scientific tools — for literature, chemistry, materials, biology, simulation, and more. Each has its own community driving support, innovation, and development.
The Scientist Assistant Agent gives you one simple, governed way to make them all work together. Each tool stays current with its own community — so your agent grows as the field grows, and never goes obsolete.
Declare. Connect.
Orchestrate. Grow.
A base framework that lets you declare the tools and agents you want in your environment — each wrapped in a uniform API so the Scientist Assistant Agent can call it as a skill. Build the agent you need; expand it whenever the field moves.
Declare your stack
Pick the tools your science needs from the vetted library. Your environment is unique to you — nothing you don't want, nothing you can't audit.
À LA CARTEWe wrap & connect
Each tool is wrapped in a uniform API and exposed to the agent as a skill — all connected over MCP, the open standard for tool use.
MCP TRANSPORTOne agent orchestrates
The Scientist Assistant Agent routes each request to the right tool, chains them together, and returns a grounded answer through a single web UI.
ORCHESTRATION LAYERIt grows with the field
Each tool evolves with its own community. Add new skills as the space evolves — your agent adapts instead of going obsolete.
NEVER OBSOLETEYour à la carte
skill catalog.
The open-source ecosystem a staff scientist can wire into their agent — grouped by research workflow. Every package is real, actively maintained, and backed by its own community. We vet, wrap, and connect each one; anything not yet on the shelf, we integrate.
PaperQA2
RAGAnswer questions over your paper corpus with grounded, in-text citations.
Future-House/paper-qaAi2 Scholar QA
SYNTHESISMulti-paper literature reviews with evidence and comparison tables.
allenai/ai2-scholarqa-libOpenScholar
SEARCHGrounded scientific Q&A retrieved over 45M+ open-access papers.
AkariAsai/OpenScholarPyAlex
METADATASearch and aggregate scholarly metadata across 270M+ works via OpenAlex.
J535D165/pyalexpyzotero
REFERENCESRead and write a researcher’s Zotero reference library.
urschrei/pyzoteroarxiv.py
PREPRINTSSearch arXiv and fetch paper metadata and PDFs.
lukasschwab/arxiv.pyDocling
PARSEConvert PDF, DOCX, PPTX, and HTML into structured Markdown and JSON.
docling-project/doclingMarker
CONVERTFast, high-fidelity PDF→Markdown, strong on math and tables.
datalab-to/markerGROBID
EXTRACTPull headers, references, and full text from scholarly PDFs as TEI/XML.
grobidOrg/grobidUnstructured
INGESTPartition 25+ document types into clean, LLM-ready chunks.
Unstructured-IO/unstructuredNougat
OCRTransformer OCR for academic PDFs, including equations.
facebookresearch/nougatPyMuPDF
EXTRACTFast text, image, and table extraction and PDF rendering.
pymupdf/PyMuPDFpdfplumber
TABLESExtract text and tables with exact layout coordinates.
jsvine/pdfplumberCamelot
TABLESExtract tables from text-based PDFs straight to DataFrames.
camelot-dev/camelotpandas
DATAFRAMESLoad, transform, and aggregate tabular data.
pandas-dev/pandasPolars
DATAFRAMESHigh-performance columnar DataFrames for large data.
pola-rs/polarsDuckDB
SQLRun fast analytical SQL over files and DataFrames in-process.
duckdb/duckdbSciPy
NUMERICSFit, optimize, and transform numerical data.
scipy/scipystatsmodels
STATSFit statistical models and return rigorous diagnostics.
statsmodels/statsmodelsscikit-learn
MLTrain and evaluate classical machine-learning models.
scikit-learn/scikit-learnpapermill
NOTEBOOKSParameterize and execute analysis notebooks programmatically.
nteract/papermillnbconvert
NOTEBOOKSExecute notebooks in a kernel and export results.
jupyter/nbconvertData Formulator
EXPLOREAI-driven, natural-language data exploration and charting.
microsoft/data-formulatorPandasAI
NL QUERYAsk questions of a DataFrame in plain English.
sinaptik-ai/pandas-aiNumPy
ARRAYSThe foundation for numerical array computation.
numpy/numpySymPy
SYMBOLICSymbolic algebra, calculus, and equation solving.
sympy/sympyJAX
AUTODIFFDifferentiable, accelerated array computing on GPU/TPU.
jax-ml/jaxDeepXDE
PINNsSolve forward and inverse PDEs with physics-informed neural nets.
lululxvi/deepxdeFEniCSx
FEMSolve partial differential equations by finite elements.
FEniCS/dolfinxAstropy
ASTRONOMYCoordinates, units, FITS, tables, and cosmology.
astropy/astropyRDKit
CHEMCompute molecular properties and handle SMILES.
rdkit/rdkitOpen Babel
CHEMInterconvert 100+ chemical file formats.
openbabel/openbabelASE
SIMULATIONSet up, run, and analyze atomistic simulations.
DTU GitLab - asepymatgen
MATERIALSMaterials structure analysis; query the Materials Project.
materialsproject/pymatgenMDAnalysis
MDAnalyze molecular-dynamics trajectories.
MDAnalysis/mdanalysisOpenMM
SIMULATIONRun GPU-accelerated molecular dynamics.
openmm/openmmPySCF
QUANTUMRun quantum-chemistry electronic-structure calculations.
pyscf/pyscfmatminer
MATERIALSFeaturize materials data for machine learning.
hackingmaterials/matminerChemCrow
AGENTAn LLM agent that reasons with chemistry tools.
ur-whitelab/chemcrow-publicBiopython
BIOSequence I/O, BLAST, and NCBI / PubMed access.
biopython/biopythonBiomni
AGENTA general-purpose biomedical research agent.
snap-stanford/Biomniscanpy
SINGLE-CELLAnalyze single-cell RNA-seq and annotate cell types.
scverse/scanpypysam
GENOMICSRead and manipulate sequencing alignments (BAM / VCF).
pysam-developers/pysamColabFold
STRUCTUREFast AlphaFold2 protein-structure prediction.
sokrypton/ColabFoldOpenFold
STRUCTURETrainable PyTorch reproduction of AlphaFold2.
aqlaboratory/openfoldMNE-Python
NEUROAnalyze EEG / MEG neuro time-series.
mne-tools/mne-pythonxarray
GEOWork with NetCDF and labeled, gridded N-D data.
pydata/xarrayObsPy
SEISMOLOGYFetch and process seismic waveform data.
obspy/obspyyt
SIMULATIONAnalyze and visualize 3D volumetric simulation data.
yt-project/ytSnakemake
WORKFLOWDefine and run reproducible HPC / cloud pipelines.
snakemake/snakemakeNextflow
WORKFLOWRun portable, container-native bioinformatics pipelines.
nextflow-io/nextflowMLflow
TRACKINGLog experiments, parameters, and models.
mlflow/mlflowDVC
VERSIONINGVersion datasets and models; reproduce pipeline stages.
iterative/dvcPrefect
ORCHESTRATIONOrchestrate dependent tasks with retries and scheduling.
PrefectHQ/prefectDagster
DATA ASSETSBuild typed, observable data-asset pipelines.
dagster-io/dagsterApache Airflow
SCHEDULERSchedule and monitor batch DAG workflows.
apache/airflowHydra
CONFIGCompose and sweep experiment configurations.
facebookresearch/hydraMatplotlib
PLOTTINGGenerate publication-quality plots.
matplotlib/matplotlibPlotly
INTERACTIVEGenerate interactive, shareable web charts.
plotly/plotly.pyBokeh
INTERACTIVEBuild interactive browser visualizations and dashboards.
bokeh/bokehseaborn
STATS VIZMake statistical plots quickly with great defaults.
mwaskom/seabornStreamlit
WEB APPExpose a result as an interactive web app.
streamlit/streamlitGradio
UIWrap a model or tool in a UI and API endpoint.
gradio-app/gradioQuarto
REPORTINGRender reproducible scientific reports to PDF / HTML.
quarto-dev/quarto-cliMCP Python SDK
MCPExpose any tool to the agent over the open MCP standard.
modelcontextprotocol/python-sdkFastMCP
MCPTurn any Python function into an MCP skill.
jlowin/fastmcpFastAPI
APIWrap a tool behind a uniform, async HTTP API.
fastapi/fastapiLangChain
FRAMEWORKTool-calling and agent glue.
langchain-ai/langchainLangGraph
ORCHESTRATIONStateful, multi-agent orchestration graphs.
langchain-ai/langgraphLlamaIndex
RAGBuild retrieval and data agents.
run-llama/llama_indexvLLM
SERVINGServe local models behind an OpenAI-compatible API.
vllm-project/vllmOllama
LOCAL LLMRun local models for offline or air-gapped use.
ollama/ollamaRay
DISTRIBUTEDScale tool execution across a cluster.
ray-project/rayBioMCP
MCPQuery biomedical databases (PubMed, trials, variants) over MCP.
genomoncology/biomcparXiv MCP
MCParXiv search, download, and analysis as an MCP tool.
blazickjp/arxiv-mcp-servermcp.science
MCPReady-made scientific MCP servers (Materials Project, DFT, Jupyter).
pathintegral-institute/mcp.scienceE2B
SANDBOXSafely execute agent-generated code in secure sandboxes.
e2b-dev/code-interpreterOpen Interpreter
EXECUTERun tasks via natural-language code execution.
openinterpreter/open-interpreterDon't see a tool you rely on? That's what our custom integration service is for — we'll vet it, wrap it, and connect it to your agent.
Other tools can orchestrate. Our difference is singular: we focus on science and scientists. We know your environment, your review processes, and what "trustworthy" has to mean in a lab.
Built by people who
know scientists.
Science-native
The tools are curated for research, not generic business workflows. We know what a staff scientist actually needs on the bench.
Yours, on your terms
Every environment is unique to the scientist and can run fully on-prem — including air-gapped networks. Your data never has to leave.
Vetted, not just bundled
Every tool in a stack has been reviewed by our team for fit, quality, and security — so IT can trust what's running.
Never obsolete
Because each tool evolves with its own community, your agent keeps improving as the field does. No single vendor to outpace it.
Connected by MCP
Skills speak the open Model Context Protocol — an interoperable standard, not a proprietary format you're locked into.
Custom integration
Need something niche or in-house? Our custom integration service vets, wraps, and connects it to your agent.
Deployable inside
your walls.
Designed to pass review at a national lab first — and to hold up anywhere security and provenance matter.
On-prem & air-gapped
Runs entirely on lab-controlled infrastructure. No dependency on outside services for sensitive work.
You own the keys & data
Bring-your-own-key and full data residency. Your data and credentials stay under your control.
Open standard transport
Tools connect over MCP — interoperable and inspectable, with no proprietary lock-in.
Auditable & licensed
Every tool is open source with a known license and community. What runs is transparent and reviewable.
The agent is the thin part.
We are the orchestration layer and web UI. The heavy lifting is done by mature open-source tools, each maintained by its own community — so there's less bespoke code for your team to trust, and more battle-tested software underneath.
We know scientists.
That's the whole point.
Ricochet is led by Casey Cobb — CEO, Berkeley Lab Affiliate, conference keynote speaker, and author of Flipping the Script. For close to two decades, Ricochet has helped science organizations turn information chaos into well-managed, integrated systems.
The Scientist Assistant Agent is that philosophy applied to the AI era: don't reinvent the wheel, connect the best of what exists, and wrap it in the judgment and governance a lab can actually trust.
Ready to assemble your agent?
Tell us about your science and your environment. We'll show you the stack we'd build — and how it runs inside your walls.