tulip-frameworks

Tulip agents is an open-source, full-stack framework designed for building agentic AI systems with safety and control as core principles. The platform addresses the need for secure and reliable autonomous agents by integrating control mechanisms directly into its architecture, ensuring that consequential actions are executed only after passing user-defined policies. This approach aims to prevent unauthorized actions, such as those resulting from model jailbreaks, by keeping critical rules outside the model itself.

The framework offers a comprehensive suite of features for agent development, including support for tools, durable memory, retrieval-augmented generation (RAG), streaming, and both synchronous and asynchronous operations. Tulip agents can be configured for a variety of multi-agent orchestration patterns such as sequential, parallel, loop, graph, orchestrator, swarm, handoff, and agent-to-agent (A2A) communication. The cognitive router component dynamically selects the appropriate agent structure for each task, while the GSAR system grounds every claim or abstains from making unsupported assertions. The admission gate enforces policy checks before any significant agent action is taken.

Developers can easily define tools as functions and expose them to the agent, swap between supported models like OpenAI and Anthropic by specifying a string, and manage conversation memory and streaming events. The framework also provides advanced features such as state management, retry strategies, hooks, structured output, and support for multi-modal providers. Observability is built in, with capabilities for streaming events, event catalogues, and evaluation. For production environments, Tulip agents includes hardening features like checkpoint backends, security guardrails, red-teaming capabilities, tamper-evident audit trails, and integration with real-world gates for workflows such as payment refunds, infrastructure deployment, and GDPR-compliant data deletion.

Tulip agents can be installed via pip and integrates with a range of external services, including identity providers, threat intelligence platforms, and notification systems. Its open-source nature and modular design make it suitable for developers and organizations seeking to build, secure, and manage autonomous agent workflows across diverse domains.