Open Data in Slicelytics from AI Agents

Two Ways for an Agent to Open Your Data

An agent opens data in Slicelytics either by building a link or by writing files into a folder that Slicelytics watches. Links need no setup and work in any modern browser, but browsers and operating systems limit how long a link can be. The watched folder has no such limit, which makes it the way to go for larger datasets and for scripts that write a new file on every run.

Watch a Folder: The Way to Open Larger Datasets

Watching a folder connects Slicelytics to a folder on your computer. Your agent, or any script, writes data files into it, and Slicelytics lists them, newest first, a couple of seconds after they are written. Nothing is uploaded and no link is built, so the size of a dataset is limited only by your computer. NDJSON files are streamed in the background, with a progress bar, so even runs of several gigabytes open without freezing the page, including the run your agent picks to compare against.

1. Pick a folder once: in Chrome or Edge, open Slicelytics, click "Watch a folder" in the Explore view, and pick a folder, e.g. ~/Slicelytics. Tell your agent which folder you picked.
2. Let your agent write files: each run goes into the folder under a new name, e.g. run-42.ndjson. Supported formats are .json, .ndjson, .jsonl, .csv, and .toon, plain or compressed as .gz or .zst.
3. Watch new files appear: Slicelytics checks the folder every two seconds and adds new files to the front of the list. Click one to load it.
4. Keep it connected: the folder is remembered across visits. Install Slicelytics as an app, or choose "Allow on every visit" when asked to reconnect, and you never have to pick it again.

New Files Appear Without Replacing Your View

While you are looking at a dataset, a new file only joins the list - Slicelytics never swaps the data under you unasked. When you do want to see a run, ask your agent ("show me run 42"). It writes a small request file into the folder, and Slicelytics loads that run within a couple of seconds:

// slicelytics.show.json
{ "run": "run-42.ndjson" }

If Slicelytics is closed when a new run arrives, it opens on the newest run next time. If nothing new has arrived, it reopens on what you were last looking at.

A Ready-Made View for Every Run

A run can say how it should be shown. Next to run-42.ndjson, your agent can write run-42.view.json with the page to open, the properties to show, value filters, the x-axis, and the chart type and metric for each property. It applies whenever that run loads:

// run-42.view.json
{
  "v": 1,
  "view": "visualize",
  "props": ["p95_ms"],
  "filters": { "endpoint": "/api/search" },
  "x": "day",
  "chart": "line"
}

To compare two runs side by side, the view names the other run, e.g. { "v": 1, "compare": "run-41.ndjson" }, and Slicelytics opens the Compare view with both.

Open Data with a Link

For small and medium datasets, or when your agent can only reply with text, a link is the quickest route. The agent compresses your file into the part of the link after #, which browsers never send to a server, and can add a view that sets the page, properties, value filters, x-axis, and chart. A Compare link carries both datasets. Agents find the link format at slicelytics.com/llms.txt, and people in the link format reference.

If you have Slicelytics installed and already open, a link you click doesn't replace what you are looking at: the app offers it with a "Show" button. If you are signed out, the link brings you back to its data and view once you have signed in.

Add the Slicelytics Skill to Your Agent

The Slicelytics agent skill is published on GitHub at github.com/duryno/Slicelytics-public. It teaches your agent when to build a link and when to use the watched folder, how to set up the view, and to leave your data as it is. It comes with a small Python script that builds Slicelytics links. In Claude Code, install it as a plugin:

/plugin marketplace add duryno/Slicelytics-public
/plugin install slicelytics@slicelytics

For other agents, such as Codex or Cursor, copy the slicelytics folder from .agents/skills into the agent's skills folder. The repository's README has the steps for each.

The same repository is the place for support requests, feedback, and feature requests: open an issue there.

What to Ask Your Agent

• "Plot p95 latency per day from runs.csv in Slicelytics."
• "Show only the failed requests in responses.ndjson in Slicelytics."
• "Compare run-41 and run-42 in Slicelytics."
• "Write each experiment run to ~/Slicelytics so I can follow along."

The instructions at slicelytics.com/llms.txt tell agents to pass your files as they are: no sorting, cleaning, or summarising unless you ask. Filtering and counting you want to see go into the view, which leaves the file untouched.

Common Questions

Which AI agents work with Slicelytics?

Any agent that can follow the instructions at slicelytics.com/llms.txt and either print a link, open one on your computer, or write files into a folder - for example Claude Code. Agents that support skills can use the Slicelytics skill.

Does watching a folder work in Firefox or Safari?

Not yet. Watching a folder needs the File System Access API, which only Chrome and Edge provide. In other browsers, your agent can open data with a link.

Do I need a subscription?

Yes, opening data needs a signed-in Slicelytics subscriber. A link opened before signing in or subscribing survives login and checkout, so you land on it afterwards.