Skip to main content

​
Vibe Coding with Pixeltable

Pixeltable’s declarative primitives make it ideal for LLM-assisted development. Instead of writing complex orchestration code, you describe what you want—and LLMs can generate correct Pixeltable code on the first try. 
# LLMs can easily generate this pattern
import pixeltable as pxt
from pixeltable.functions.huggingface import sentence_transformer

t = pxt.create_table('myapp.docs', {'text': pxt.String}, if_exists='ignore')
t.add_embedding_index('text', embedding=sentence_transformer.using(model_id='all-MiniLM-L6-v2'))

​
Agent Instructions Files

We provide instruction files for AI coding agents (Cursor, Claude Code, Windsurf, etc.) to understand the Pixeltable codebase:

AGENTS.md

Instructions for AI agents: project structure, conventions, and patterns

CLAUDE.md

Claude-specific agent instructions (same content as AGENTS.md)
Add these to your project or point your AI assistant to them for better Pixeltable code generation.

​
Plain text docs

You can access all of our documentation as plain text markdown files by adding .md to the end of any url. For example, you can find the plain text version of this page itself at https://docs.pixeltable.com/overview/building-pixeltable-with-llms.md. This helps AI tools and agents consume our content and allows you to copy and paste the entire contents of a doc into an LLM. This format is preferable to scraping or copying from our HTML and JavaScript-rendered pages because:
  • Plain text contains fewer formatting tokens.
  • Content that isn’t rendered in the default view (for example, it’s hidden in a tab) of a given page is rendered in the plain text version.
  • LLMs can parse and understand markdown hierarchy.
We host two files to assist AI tools and agents:
  1. https://docs.pixeltable.com/llms.txt: This file instructs how to retrieve plain text versions of our pages. It follows the emerging standard for making websites and content more accessible to LLMs.
  2. https://docs.pixeltable.com/llms-full.txt: This file contains more comprehensive instructions and metadata about our documentation structure.
These files help AI tools navigate and understand our documentation more effectively.

​
Hosted MCP server

We provide a hosted MCP server that AI applications can connect to directly for searching our documentation. MCP Server URL: 
https://docs.pixeltable.com/mcp
Available Tools:
ToolDescription
SearchPixeltableDocumentationSearch across the Pixeltable documentation to find relevant information, code examples, API references, and guides. Returns contextual content with titles and direct links.
Use this URL to connect AI applications like Claude Desktop, Cursor, or Windsurf to our documentation.

​
Self-hosted MCP server

For developers who want more control or need to call the Pixeltable API directly, we provide a self-hosted Pixeltable MCP server. This server provides AI agents tools for calling the Pixeltable API and searching our knowledge base.

Pixeltable MCP servers

View the source code and contribute to our Pixeltable MCP servers
 The self-hosted server gives agents direct access to Pixeltable’s multimodal capabilities:
  • Audio: Transcription, speaker diarization, audio search
  • Video: Frame extraction, scene detection, object tracking
  • Images: Object detection, classification, similarity search
  • Text: RAG retrieval, document parsing, semantic search

​
Pixelagent Toolkit

If you’re building agentic software, we provide an SDK for adding Pixeltable functionality to your agent’s capabilities.

Pixelagent

Agent engineering blueprint powered on Pixeltable
 Pixelagent provides:
  • Multi-agent systems: Coordinate multiple agents with shared memory
  • Tool calling: Register UDFs and queries as callable tools
  • State management: Automatic persistence of conversation history
  • Observability: Full lineage tracking of agent decisions
Learn more in our Pixelagent documentation.

​
See also

Last modified on January 29, 2026