Research Workflow
Use Matter42 like a lab notebook connected to analysis tools: create a project, upload files, ask the agent to parse and explore them, then run targeted tools with explicit scientific assumptions.
Projects
A project is the unit of organization for a sample set, collaboration, or analysis campaign. It holds:
- Chats with the research agent.
- Uploaded files and parsed datasets.
- Generated figures and tool results.
- Literature records when paper ingestion is enabled.
Keep related raw files in the same project, especially matched Raman and PL maps from the same flake. That lets the agent find project files, reuse parsed dataset IDs, and compare results across turns.
Upload Files
Open Files from the project sidebar and upload the raw measurement. You can also attach a project file directly from chat when the composer offers file picking.
After uploading, ask the agent to parse the file rather than pasting file contents into the message:
Parse the uploaded Raman map as MoS2 and set boundary_buffer_um=2.0.
For paired measurements, keep each measurement separate:
Parse the PL map and Raman map I uploaded. Explore both, then cluster the paired datasets.
Start A Chat
The best prompts specify the dataset, the material, the physical region, and the output you need. Matter42 can infer some context, but quantitative workflows improve when you provide the experimental assumptions.
Useful prompt patterns:
Explore this Raman map and tell me whether the E2g and A1g peaks look well resolved.
Segment the map first. Then estimate defect density with region_mode="interior" and selection_policy="largest_component".
Classify the dominant defect type. Use instrument_fwhm=1.5 cm^-1 and explain whether the result looks calibration-limited.
Run KMC for WS2 at 950 K, chalcogen-to-metal flux ratio 80, and 4 seeds.
Read Tool Cards
When the agent calls a tool, the chat shows a tool card with the input arguments, progress state, returned payload, and generated figures. For spectral and spatial tools, figures are interactive Plotly views. You can pan, zoom, hover pixels, and inspect legends without leaving the chat.
Common outputs include:
- Dataset summaries with
dataset_id, dataset kind, data type, axis units, and parsing notes. - Spatial heatmaps for intensity, peak center, FWHM, density, activity, regions, or clusters.
- Spectral plots with mean spectra, cluster spectra, or peak annotations.
- Statistics such as valid-pixel counts, feature medians, silhouette scores, density percentiles, and defect rankings.
- Suggested follow-up tools based on the result.
Use Region Controls
Region controls matter whenever the map contains tears, milled areas, boundary halos, edge artifacts, or disconnected islands.
Use segment_regions before quantitative analysis if you are not sure which pixels are clean. Then choose one of these analysis scopes:
region_mode="quality"for parse-time valid pixels.region_mode="interior"for intact regions away from boundaries.region_mode="transition"to study boundary halos.region_mode="damaged"to isolate masked or damaged regions.region_mode="all"when you intentionally want every pixel.
For PFIB samples, start with a nonzero boundary_buffer_um, often 2-3 um depending on the milling geometry and spatial resolution.
Ask Follow-Up Questions
Matter42 keeps the chat context and project files together, so you can refine an analysis without re-uploading data:
Repeat the density estimate with a 3 um boundary buffer and compare the median to the previous result.
Use the PL dataset as auxiliary evidence. Do the PL-quenched pixels overlap with high Raman linewidth?
Plot the spectrum for this dataset and overlay the annealed sample.
If the result seems surprising, ask for diagnostics rather than a new conclusion: valid pixel count, quality mask behavior, region summary, peak windows, instrument broadening, calibration range, or whether the model clipped at a boundary.
Literature And Notes
Use the Literature section for PDFs and extracted paper metadata when enabled. The agent can use parsed papers, protocols, and notes as context, but tool-backed quantitative conclusions should still come from parsed datasets and analysis tools.
API Keys
API keys, when exposed in your deployment, let external MCP clients save parsed datasets and results to your Matter42 account. Treat them like secrets and rotate keys that may have leaked.