crewai-wzrd 0.3.0

WZRD velocity signals, oracle inference, and earn-loop tooling for CrewAI agents

crewai-wzrd

WZRD velocity signal and earn-loop tooling for CrewAI. Gives agents real-time model momentum data for task-aware model selection, plus a native one-shot earn cycle that wraps wzrd-client.

Free tier. No auth required. Signals cached 60s.

Install

pip install crewai-wzrd

WzrdEarnTool uses the same shared wzrd-client loop as the CLI, so it will resolve the agent keypair from the usual WZRD_* env vars or auto-generate a dedicated wallet on first use.

Usage

from crewai import Agent, Task, Crew
from crewai_wzrd import WzrdEarnTool, WzrdVelocityTool

researcher = Agent(
    role="Model Selection Specialist",
    tools=[WzrdVelocityTool(), WzrdEarnTool()],
    goal="Pick the best AI model for each task and harvest CCM when the work is done"
)

task = Task(
    description="Find the fastest model for code generation right now",
    agent=researcher,
    expected_output="Model name with supporting velocity data"
)

crew = Crew(agents=[researcher], tasks=[task])
result = crew.kickoff()

The tool auto-detects capability from the task description (code, reasoning, chat, vision) and filters signals accordingly.

WzrdEarnTool runs one full earn cycle through wzrd-client: it authenticates, picks a model, runs server-witnessed inference, reports the result, and checks/claims rewards when available. The tool returns the captured loop log as plain text so the agent can inspect the cycle outcome.

What it returns

WZRD Velocity Signal (capability: code)
Top models by momentum:
  1. deepseek/deepseek-coder-v2 - trend: surging, score: 0.87, confidence: normal
  2. anthropic/claude-3.5-sonnet - trend: accelerating, score: 0.82, confidence: normal
  3. openai/gpt-4o - trend: stable, score: 0.79, confidence: normal
Recommended: deepseek/deepseek-coder-v2 - momentum is surging.
Signal timestamp: 2026-04-02T12:00:00Z

Links

• WZRD Protocol
• Signal API docs