Skip to content
Pixie logo Pixie

Data & secrets

set-secret

non-destructive (writes .env)

Trigger: “Set OPENAI_API_KEY for whisper”, “Add my Anthropic key to the RAG tool”, “Clear the X secret”.

Steps:

  1. Identify the tool (uses list-tools if ambiguous).
  2. Identify the secret key.
  3. Read the value from stdin without echoing.
  4. Write KEY=value to tools/<id>/.env (creates chmod 600 on POSIX).
  5. Confirm.

Never:

  • Logs the value.
  • Echoes it back.
  • Stores it in pixie.db or anywhere outside tools/<id>/.env.

After set-secret, the tool’s settings page in the dashboard shows the secret as “set” (green pill). To replace it, run set-secret again or use the “Replace” button in the settings UI.

To clear a secret, run with an empty value or use the “Clear” button.

fetch-dataset-from-url

non-destructive

Trigger: any HTTPS URL pointing at a data file, with a request to “fetch” / “download” / “import” it for a named tool.

Downloads to tools/<id>/data/ with:

  • Safe extraction (refuses paths containing ..).
  • SHA-256 checksum verification (if you provide an expected hash).
  • A small note appended to data/README.md recording source URL and date.

Doesn’t handle Kaggle (use fetch-dataset-from-kaggle) or HuggingFace (use fetch-dataset-from-huggingface) — those need authenticated API calls.

fetch-dataset-from-kaggle

non-destructive

Trigger: mentions of Kaggle, a dataset slug like username/dataset, or a competition name.

Uses your local Kaggle credentials (typically ~/.kaggle/kaggle.json) to download the dataset or competition data into tools/<id>/data/. Refuses Kaggle notebooks (those are scripts, not datasets — use add-tool-from-notebook).

fetch-dataset-from-huggingface

non-destructive

Trigger: mentions of HuggingFace, an org/dataset slug from the HF hub.

Uses huggingface_hub with your local HF_TOKEN (if set) to pull public or gated datasets into tools/<id>/data/. Refuses HF models — those should be loaded by the tool itself via transformers, not copied into the tool folder.

import-dataset-from-local

non-destructive

Trigger: a local filesystem path. “Use ./data/iris.csv as the dataset for tool foo”.

Copies / symlinks / moves a local file or folder into tools/<id>/data/. Asks before overwriting an existing same-named file.

How tools find their data

Pixie sets cwd=tools/<id>/ when spawning a tool, so relative paths work:

import pandas as pd
import pathlib


DATA_DIR = pathlib.Path(__file__).parent / "data"


iris = pd.read_csv(DATA_DIR / "iris.csv")

data/ is in .gitignore by default — large datasets shouldn’t live in your tool’s source repo. If you want to ship a small reference fixture, put it in reference/ instead (see Fixtures & reference validation).