Vera verifies autonomous AI agents before they act.
Vera is an open-source control layer for production AI agents. It intercepts tool calls, runs Qwen-powered multi-agent review, applies policy gates, requests human approval for risky actions, and records an auditable trace of every decision.
Built for the Global AI Hackathon Series with Qwen Cloud.
Primary track: Track 3, Agent Society
Secondary fit: Track 4, Autopilot Agent
An autonomous DevOps agent receives a production incident. Without Vera, the agent may perform risky changes directly. With Vera, every proposed action is routed through a review society:
- Planner proposes next steps.
- Risk reviewer evaluates blast radius.
- Policy reviewer applies guardrails.
- Memory reviewer recalls similar incidents.
- Judge approves, blocks, or escalates.
The dashboard shows the full trace, policy decisions, approvals, and final outcome.
- Frontend: Next.js dashboard
- Backend: FastAPI or Node API
- AI: Qwen Cloud OpenAI-compatible API
- Storage: SQLite for local MVP, Postgres-ready schema later
- Policies: JSON/YAML rule files
- Deployment: Alibaba Cloud backend proof
apps/ Frontend apps
services/ Backend services and agent runtime
policies/ Policy rules and guardrail definitions
scenarios/ Demo incidents and workflow fixtures
eval/ Evaluation harness and baseline comparisons
infra/ Deployment assets
docs/ Product, architecture, and submission docs
Groundwork phase. Product scope, requirements, pages, resources, and submission plan are being locked before implementation.