EasyViz

Scientific plotting guided by data and literature.

EasyViz guides local Agents that can run Python and inspect images, from data or references to editable scientific figures.

Latest stable release.

| Track | Start with | The Agent does | | --- | --- | --- | | Create | Data and a scientific question | Choose a useful chart, apply suitable design and review the image. | | Reproduce | A reference image and your data | Read the reference, implement its layers and compare the result. |

Receive a reviewed SVG, separate caption, runnable code/settings and traceable plotting inputs at your chosen size and typography. Request PDF, PNG or TIFF when needed; an editable .ev project and local workbench are optional.

For custom Create panels, the Agent plans the main comparison and supporting layers before the first render.

Install

Give your Agent this request:

Install or update https://github.com/idealstarry/EasyViz in my local
ChatGPT Desktop. Follow INSTALL.md and verify the installation for me.

npm · Node.js 20+ and Python 3.11+:

ChatGPT Desktop / Codex:

npx easyviz codex

DeepSeek Harness:

npx easyviz deepseek

Installing the public npm package needs no npm account. It includes all three skills; drawing dependencies are installed separately. See installation and update options. Start a new chat after installing.

Use

Create

Use EasyViz to inspect /path/to/my-data and recommend useful figures.
Use an 80 × 70 mm panel and 8 pt text. Inspect and refine the actual images
before delivery, export SVG and save a separate caption and runnable code.

Reproduce

Use EasyViz to reproduce this reference with my measurements.csv.
Keep its layers and axis structure, use a 100 × 76 mm panel and 8 pt text,
and write plotting code for any unsupported layers. Export SVG and save a separate caption.

On first use in each chat, the Agent briefly explains drawing, chat-based tweaks and the optional workbench. It inspects the data, then asks only the consequential missing questions, within at most two clarification rounds. You can ask it to recommend a design. It selects suitable literature mechanisms, checks the actual images and corrects concrete findings before the first finished delivery. Small changes can be requested directly in chat; cosmetic edits preserve adopted statistics and scientific meanings.

Examples

Create

Compartment response · 88 Ccl2 observations, mean ± SEM and separate units. Data and code.

Expression and genotype contrasts · 174 TPM values and an aligned descriptive comparison. Data and code.

Signature relationships · Gene overlaps, paired coefficients and descriptive sign fractions. Data and code.

Replicate treatment bars, basic panels and other cases cover simpler questions.

Reproduce

Paired effect lanes · Original Figure 3a,b from Vabistsevits et al., Nature Communications (2024).

| Original literature panel | EasyViz reconstruction | | --- | --- | | | |

Data, code and declared adaptations.

Integration rates · The selected depot radar from Massier et al., Nature Communications (2023).

| Original literature panel | EasyViz reconstruction | | --- | --- | | | |

Data, code and declared adaptations.

Broken-axis replicate bars · Figure 2c from Huang et al., Nature Communications (2023).

| Original literature panel | EasyViz reconstruction | | --- | --- | | | |

Data, code and declared adaptations.

These reconstructions retain all selected supplied values and adopt editable 8 pt text. Their case pages record measured geometry, intentional changes and remaining differences. The original excerpts are CC BY 4.0. Time-course reproduction is also available.

Published Source Data or author code is optional for your own reproduction. Browse all cases.

Colors

Choose a palette for the marks and scientific meaning. Defaults are editable.

| Use | Palette | Preview | | --- | --- | --- | | Summary areas with dark boundaries | progeny-summary | | | Two observation classes | scwat-blue-pink | | | Increasing magnitude | somerville-sky | | | Values around an adopted center | scwat-blue-white-coral | |

All palettes, HEX values and literature sources.

Optional local workbench

For ordinary plotting and small changes, stay in the Agent chat. To annotate several locations or compare attempts visually, ask:

Open the EasyViz workbench for this project.

The Agent opens the local page only on request. It can register the current validated SVG or prepare an editable .ev project when requested; ordinary delivery needs no browser, running service or MCP connection.

Use the local page at http://127.0.0.1:PORT/ to annotate figures, and the original Agent chat to apply your edits:

1. Choose a registered figure or import a valid .ev project.

2. Select SVG elements or draw numbered regions, write a comment for each, then click Save drafts to save all completed comments together.

3. Return to the same chat that created the figure and send:

   Apply the saved EasyViz workbench comments to this figure.
   Update its plotting code, render a fresh attempt and review the result.

4. The Agent reads the saved comments directly; no copying or retyping is needed. Compare the new attempt in the workbench, accept it or request another round of changes. An accepted version can be restored later.

With an original Agent connected and actively waiting through MCP, Submit edits can hand off the comments without another chat message. After that Agent's turn ends, use step 3 unless a supported host's scheduled continuation has been configured and verified. Saving comments alone does not start an idle Agent.

The editable canvas uses mapped SVG elements; raster layers remain raster. PDF and PNG are additional exports. Click the figure name to rename it; the .ev project and Agent context keep that name. Workbench guide · Optional scheduled checks.

For MCP setup, ask your Agent to follow INSTALL.md. A separate Codex worker is available only when explicitly requested; importing an .ev project alone does not run its code.

Guides

Chart library · Data exploration and analysis · First Create delivery · Reference to code · Development · Validation and limits

Original code and documentation: MIT. Third-party materials retain their own terms; see source attribution.