Build your own stateful agent framework
We see agents as the intersection of an LLM, storage, and orchestration. Build your own agent framework with Pixelagent - an open-source blueprint that handles data infrastructure while giving you complete control. Multi-provider, multimodal support, memory..
Pixelagent blueprints can be transformed into PyPI packages, enabling teams to extend the build-your-own philosophy to deployment and sharing. Agents become reusable, versioned components that other developers can install and customize.
Pixelagent's declarative model supports custom agentic strategies through system prompts, step-by-step reasoning patterns, and composable extensions. Examples include ReAct planning loops, self-improvement reflection, and team workflows—all built on Pixeltable's data infrastructure.
Pixelagent is model-agnostic and extensible to multiple providers (Anthropic, OpenAI, AWS Bedrock). Agents defined once can swap providers by changing imports, enabling cost optimization and model experimentation without architectural changes.
Pixelagent automatically persists agent memory in Pixeltable tables with complete traceability of messages, tool calls, and performance metrics. Users can access conversation history, tool call logs, and customize memory retention (e.g., n_latest_messages) without manual database management.
Pixelagent provides a declarative Python framework that unifies LLM, storage, and orchestration into a single interface. Engineers can build agents with native multimodal support (text, images, audio, video), implement custom memory systems with semantic search, add tool-calling via Python UDFs, and extend with reasoning/reflection loops—all without vendor lock-in.