$ pip install nyra

Engineering from
First Principles.

The generative AI ecosystem relies on bloated wrappers and black boxes. We got tired of debugging magic prompts, so we built a completely deterministic, global multi-agent architecture.

Zero Dependencies.

Nyra is 100% Python native. By stripping away heavy frameworks and abstract wrappers, Nyra allows elite developers to engineer intelligence systems with absolute visibility into the execution stack.

The Cognitive Loop.

Instead of hoping an LLM gets the right answer, Nyra enforces a rigorous architectural loop. Tasks are routed sequentially through the Planner, the Reasoner, the Orchestrator, and finally, the Critic.

If a task fails validation, the Critic forces the Reasoner to try again. Deterministic control, every single time.

The Proof.

See the simplicity of the API and the depth of the execution trace for yourself. No magic, just code.

1. The Setup (Code)
from nyra.agent import Agent from nyra.providers.nvidia import NvidiaLLM API_KEY = "your-nvidia-api-key" # Boot up the Agent agent = Agent() agent.llm = NvidiaLLM(api_key=API_KEY) # Connect the cognitive modules agent.planner.llm = agent.llm agent.reasoner.llm = agent.llm agent.critic.llm = agent.llm agent.orchestrator.llm = agent.llm # Register Specialized Sub-Agents agent.register_sub_agent( "ResearchBot", "Researcher", ["research_web", "extract_statistics"] ) # Run the deterministic loop objective = "Research India's EV market growth from 2020 to 2025." agent.run(objective)
2. The Execution (Trace)
[Planner] 1. Research India's EV market growth from 2020 to 2025 2. Extract statistics on India's EV market growth from 2020 to 2025 3. Calculate CAGR of India's EV market growth [Orchestrator] Routing Step 1 to -> ResearchBot [Reasoner] ResearchBot executing: "India's EV market growth from 2020 to 2025" [Result] Task 'India's EV market growth' successfully resolved by sub-agent 'ResearchBot'.