# PySpur

> Open-source visual playground for building, testing, evaluating and deploying AI agent workflows in Python or a drag-and-drop UI.

PySpur is an open-source agent builder from Sevn AI Inc. that lets AI engineers define test cases, build agentic workflows in Python code or a drag-and-drop UI, iterate on failures, and deploy the result as an API. It is Apache 2.0 licensed, and a hosted cloud version is offered through a request form.

## What It Is

PySpur is a visual playground for agentic workflows. The project describes its aim as reducing prompt trial-and-error, exposing step interactions that cause hidden failures, and replacing manual inspection of raw JSON output. Workflows are built from nodes, and new tools can be added by creating a single Python file.

## How the Workflow Works

The documented flow has four steps: define test cases for the agent, build the agent in Python or through the UI, run tests and debug at node level, then deploy to the cloud or on-premise. Each agent can be exported as a JSON file to share with a team or commit to git. Deployed agents can be published as an API, and execution traces are captured automatically.

## Capabilities

- Human-in-the-loop breakpoints that pause a workflow until a person approves it.
- Loops for iterative tool calling with memory.
- RAG pipelines that parse, chunk, embed and upsert data into a vector database.
- Multimodal inputs including video, images, audio, text and code, with file upload or pasted URLs.
- A UI editor for JSON Schema structured outputs.
- Evals on real-world datasets.
- Tools such as Slack, Firecrawl.dev, Google Sheets and GitHub.
- Support for more than 100 LLM providers, embedders and vector databases.

## Setup Path

PySpur requires Python 3.11 or higher. Install with pip, run `pyspur init` to create a project with a `.env` file, then `pyspur serve --sqlite` to start the app locally at port 6080. Configuring Postgres is recommended for a more stable experience, and provider API keys can be added in the UI or in the `.env` file. Python and TypeScript SDKs are listed as coming soon, and self-improvement is also listed as upcoming. Development setup is documented for Unix-like systems only.

## Features
- Visual drag-and-drop agent builder
- Python-based custom nodes
- Test case definition and node-level debugging
- Human-in-the-loop breakpoints
- Loops with memory
- RAG pipeline (parse, chunk, embed, upsert)
- Multimodal inputs
- Structured outputs with JSON Schema editor
- Automatic execution traces
- Evals on real-world datasets
- One-click deploy as API
- JSON export of agents for version control
- Support for 100+ LLM providers, embedders and vector DBs

## Integrations
Slack, Firecrawl.dev, Google Sheets, GitHub, OpenAI, Anthropic, Ollama, Gemini, DeepSeek

## Platforms
WINDOWS, WEB, API, VSC_EXTENSION, CLI

## Pricing
Open Source

## Version
v0.1.18

## Links
- Website: https://pyspur.dev
- Documentation: https://docs.pyspur.dev/
- Repository: https://github.com/PySpur-Dev/pyspur
- EveryDev.ai: https://www.everydev.ai/tools/pyspur