Qova Docs

Trust-based task delegation for CrewAI crews. When multiple agents are available, QovaTrustDelegation assigns tasks based on each agent's on-chain credit score. High-value tasks go to high-scored agents. Low-value tasks can be handled by newer agents still building reputation.

Install

bun add @qova/crewai @brnmwai/qova-core

How It Works

For each agent in availableAgents, reads their score from ReputationRegistry on-chain
Classifies the task by value tier (highValue / standard / lowValue)
Filters agents whose score meets the tier threshold
Among qualifying agents, picks the highest-scored one
After task completion, records the outcome via TransactionValidator

Usage

import { QovaTrustDelegation } from "@qova/crewai";

const delegation = new QovaTrustDelegation({
  chain: "base-sepolia",
  scoreThresholds: {
    highValue: 800,     // Only AAA/AA/A agents handle these
    standard: 650,      // BB+ agents for routine tasks
    lowValue: 450,      // CCC+ agents for low-risk work
  },
});

// Delegation reads each agent's score and picks the best fit
const assignedAgent = await delegation.assignTask(task, availableAgents);

Delegation Flow

Task arrives with value tier
    |
    v
[Read scores for all available agents]  -->  ReputationRegistry on Base
    |
    v
[Filter by tier threshold]
    |-- highValue task (e.g. $10,000 swap)  -->  Only agents scoring 800+
    |-- standard task (e.g. $500 payment)   -->  Only agents scoring 650+
    |-- lowValue task (e.g. $10 transfer)   -->  Only agents scoring 450+
    |
    v
[Pick highest-scored qualifying agent]
    |
    v
[Execute task]  -->  [Record outcome to TransactionValidator]

Threshold Mapping

Tier	Score Threshold	Credit Grade	Example Tasks
High Value	800+	A or above	Large swaps, bridge operations, contract deployments
Standard	650+	BB or above	Routine payments, transfers, API calls
Low Value	450+	CCC or above	Data queries, read operations, logging

Features

Score-based task assignment - highest-trust agent gets the task
Tiered delegation thresholds - different minimums for different task values
Score updates after completion - successful tasks improve the agent's reputation
Crew-level aggregation - view the overall trust level of an entire crew
Fallback handling - configurable behavior when no agent meets the threshold