Pyspur
A visual playground for agentic workflows: Iterate over your agents 10x faster
Last verified:
What is Pyspur?
PySpur is an open-source visual AI agent builder in Python that enables AI engineers to build, test, and iterate on agentic workflows 10x faster. It provides a drag-and-drop visual playground where users can construct agent workflows by connecting modular nodes representing LLM calls, tool integrations, decision logic, and more. The platform solves common AI agent development pain points like prompt hell, workflow blindspots, and terminal testing nightmares by offering step-by-step execution visibility and node-level debugging.
Key features include drag-and-drop workflow building, built-in test case definition and execution, RAG support (parse, chunk, embed, upsert to vector DB), loop support for iterative tool calling with memory, multimodal input handling (PDFs, videos, audio, images, text, code), structured JSON output editing, 10+ tool integrations (Slack, Firecrawl, Google Sheets, GitHub), evaluation on real-world datasets, one-click API deployment, and compatibility with 100+ LLM providers, embedders, and vector databases. Workflows export as JSON for easy sharing and Git version control.
PySpur is designed for AI engineers building production AI agents who need rapid iteration without reinventing the wheel. It supports both self-hosted deployment (via pip or Docker with SQLite or PostgreSQL) and cloud options. The platform is Apache 2.0 licensed, backed by Y Combinator, and offers quick pip installation with Python 3.12+. Coming soon are SDKs for Python and TypeScript for deeper codebase integration.
Pyspur pricing
Pricing model: Freemium
PySpur is fully open-source and free to use under the Apache 2.0 license with no licensing costs. Users can self-host on their own infrastructure via pip install or Docker with SQLite or PostgreSQL. The core platform has no paid tiers for self-hosted use. Cloud hosting plans may apply separately for PySpur Cloud, though specific pricing details are not published on the website. Users only pay for their own LLM provider API costs (OpenAI, Anthropic, etc.) which they configure in the app.
Pyspur pros
- Drag-and-drop visual workflow builder for rapid iteration
- Iterate over agents 10x faster than traditional methods
- Node-level debugging with step-by-step execution visibility
- Built-in test case definition and execution framework
- RAG support with parsing, chunking, embedding, and vector DB upsert
- Loop support for iterative tool calling with memory
- Multimodal support for video, images, audio, text, and code
- Structured JSON output editor with schema validation
- 10+ built-in tool integrations (Slack, Firecrawl, Google Sheets, GitHub)
- One-click API deployment to cloud or on-premise
- Compatible with 100+ LLM providers, embedders, and vector DBs
- Easy custom node creation with single Python file
- Workflow export as JSON for team sharing and Git version control
- Open-source under Apache 2.0 license with no licensing costs
- Quick pip install with Python 3.12+ (pyspur init, pyspur serve)
- Docker deployment option for scalable production systems
- Local model support via Ollama integration
- Backed by Y Combinator with active development
- Vendor-unified nodes for easy model comparison and switching
Pyspur cons
- Development on Windows/PC not supported (Unix-like systems only)
- Requires Python 3.12 or higher for pip installation
- Smaller community and ecosystem compared to alternatives
- Limited to web platform with no native desktop or mobile support
- Recommends PostgreSQL configuration for stable production use
- Narrower use case focused only on agent workflow development
- SDKs for Python and TypeScript still coming soon
- Requires manual API key configuration for LLM providers
Frequently asked questions about Pyspur
What is PySpur?
PySpur is an AI agent builder in Python that provides a visual playground for agentic workflows. AI engineers use it to build agents, execute them step-by-step, and inspect past runs. It enables rapid iteration on AI agents visually without reinventing the wheel, solving problems like prompt hell, workflow blindspots, and terminal testing nightmares.
How do I install PySpur?
The quickest way is using pip: run 'pip install pyspur', then 'pyspur init my-project' to create a new project with an .env file, then 'pyspur serve --sqlite' to start the server at http://localhost:6080. For production, Docker deployment is recommended using the provided start script which sets up PostgreSQL backing.
What Python version does PySpur require?
PySpur requires Python 3.12 or higher for pip installation. Python 3.11 is mentioned in some documentation as the minimum for certain installations.
Can I add custom tools or nodes to PySpur?
Yes, adding new tools is a matter of creating a single Python file. You can add custom nodes using a simple Python decorator with @NodeRegistry.register(), specifying the category, display_name, logo, and implementing the run() method. This makes extensibility easy for AI engineers.
How does PySpur handle RAG?
PySpur includes built-in RAG support with a two-step process: Step 1 creates a document collection through chunking and parsing, Step 2 creates a vector index through embedding and vector DB upsert. This allows agents to retrieve and use external knowledge.
Can I deploy PySpur agents as an API?
Yes, PySpur offers one-click API deployment using the 'Deploy' button in the top bar. You can publish your agent as an API and integrate it wherever you want, either to the cloud or on-premise environments with auto-generated infrastructure.
What LLM providers does PySpur support?
PySpur supports 100+ LLM providers, embedders, and vector databases through vendor-unified nodes. You add API keys for providers like OpenAI, Anthropic, and others in the API Keys tab of the app, and changes take effect immediately. You can quickly jump between models to test which works best.
How do I share and version control PySpur workflows?
Every agent can be exported as a JSON file containing the workflow definition with nodes, coordinates, and schemas. Share this JSON with your team and commit it to Git for version tracking and collaboration. Workflows are stored as files that integrate naturally with Git.
Can I use local models with PySpur?
Yes, PySpur supports local models via Ollama. Configure by starting Ollama service with OLLAMA_HOST='0.0.0.0' ollama serve, update your .env file with OLLAMA_BASE_URL=http://host.docker.internal:11434, download models using 'ollama pull <model-name>', then select Ollama models from the sidebar for LLM nodes.
What is the licensing for PySpur?
PySpur is fully open-source under the Apache 2.0 license, which provides clear patent protections. This means it's free to use, modify, and distribute with no licensing costs. The Apache 2.0 license is more permissive than many alternatives and explicitly protects users from patent claims.