Quickstart - VectorShift

Prerequisites: Python 3.10+ and a VectorShift account. If you haven’t installed yet, see Installation.

Run your first pipeline

Pick where you’re writing code — the pipeline you build is identical in each, and shows up in the platform editor either way.

Local
Google Colab
Replit

Install the SDK

pip install vectorshift

Set your API key

Generate one at Account → API Keys, then export it:

export VECTORSHIFT_API_KEY="vs-..."

Build, save, and run

Paste this into a file (hello.py) and run it.

from vectorshift import Pipeline

p = Pipeline.new(name="Hello pipeline")
query = p.add(name="query").input(input_type="string")
llm = p.add(name="llm").llm(provider="openai", model="gpt-5.1", prompt=query.text)
p.add(name="answer").output(value=llm.response)

p.save()
result = p.run({"query": "What is VectorShift in one sentence?"})
print(result["outputs"]["answer"])

python hello.py

See it in the editor

Open the platform — your new pipeline is editable in the visual editor, identical to one built there.

Install the SDK

In a new Colab notebook, run this in the first cell:

!pip install vectorshift

Set your API key

Generate one at Account → API Keys. Use getpass so the key isn’t saved in the notebook:

import os
from getpass import getpass
os.environ["VECTORSHIFT_API_KEY"] = getpass("VectorShift API key: ")

Prefer Colab’s 🔑 Secrets panel? Store it as VECTORSHIFT_API_KEY, then os.environ["VECTORSHIFT_API_KEY"] = userdata.get("VECTORSHIFT_API_KEY") after from google.colab import userdata.

Build, save, and run

Paste this into the next cell and run it.

from vectorshift import Pipeline

p = Pipeline.new(name="Hello pipeline")
query = p.add(name="query").input(input_type="string")
llm = p.add(name="llm").llm(provider="openai", model="gpt-5.1", prompt=query.text)
p.add(name="answer").output(value=llm.response)

p.save()
result = p.run({"query": "What is VectorShift in one sentence?"})
print(result["outputs"]["answer"])

See it in the editor

Open the platform — your new pipeline is editable in the visual editor, identical to one built there.

Create a Python Repl

Start a new Python Repl at replit.com.

Add the SDK

Add vectorshift from the Packages pane, or just import it — Replit installs it on first run. You can also run it in the Shell:

pip install vectorshift

Set your API key

Open Tools → Secrets, add a secret named VECTORSHIFT_API_KEY with your key from Account → API Keys. Replit injects secrets as environment variables, so the SDK picks it up automatically.

Build, save, and run

Paste this into main.py and click Run.

from vectorshift import Pipeline

p = Pipeline.new(name="Hello pipeline")
query = p.add(name="query").input(input_type="string")
llm = p.add(name="llm").llm(provider="openai", model="gpt-5.1", prompt=query.text)
p.add(name="answer").output(value=llm.response)

p.save()
result = p.run({"query": "What is VectorShift in one sentence?"})
print(result["outputs"]["answer"])

For the best developer experience, enable the mypy plugin. It catches wrong field names on node builders, mismatched types between wired nodes, and agent tools missing required inputs — at edit time, before you save. Drop a mypy.ini with plugins = vectorshift.mypy_plugin and most editors (VS Code, Cursor, PyCharm, Zed) light up SDK-aware errors inline as you type.

Now try

Stream tokens

Swap run for stream and yield tokens as they arrive.

Run in the background

start() returns immediately; poll or stream for results.

Turn it into an agent

Add tools and multi-turn conversation.

What’s next

Concepts

Pipeline vs Agent vs Session vs Knowledge Base.

Pipeline reference

Every public method.

Building with AI

Want ChatGPT or Claude to build a pipeline, agent, knowledge base, or table for you? Paste this prompt first, then describe what you want — it points the model at the SDK docs and llms.txt so it follows the documented patterns.

You are building with the VectorShift Python SDK. Before writing code, read
https://docs.vectorshift.ai/llms.txt and the SDK reference pages it links.
Follow the documented patterns exactly: use the builder API, correct node and
field names, and the right object type for my request (pipeline, agent,
knowledge base, or table). Install with `pip install vectorshift` and
authenticate via the VECTORSHIFT_API_KEY environment variable. Then build,
save, and run what I describe, returning one complete runnable script.

My request:

​Run your first pipeline

​Now try

Stream tokens

Run in the background

Turn it into an agent

​What’s next

Concepts

Pipeline reference

​Building with AI

Run your first pipeline

Now try

What’s next

Building with AI