Xerg serves as an economic control plane specifically engineered for AI agent fleets. Founded by Jason Curry and Dr. Charles Parker, the platform addresses the challenge of spiraling costs associated with AI agents by providing transparency into where money is spent, what resources are wasted, and how to implement changes to improve efficiency. The platform is named after exergy—the thermodynamic concept representing the portion of energy capable of performing useful work—reflecting its goal to distinguish productive agent activity from wasted tokens, retries, and idle loops. By offering tools that allow developers to see spend as a map rather than just a summary bill, Xerg enables teams to trace every dollar to specific agents, workflows, and executions.

The tool functions by auditing agent-runtime economics, normalizing data locally, and offering actionable insights through both local command-line interactions and hosted workspaces. It processes logs and traces from common agent frameworks and coding assistants, identifying inefficiencies such as context bloat and redundant API calls. The platform ensures that users do not need to share sensitive prompts or raw provider logs, as it focuses on economic metadata and aggregated findings. This audit process allows teams to classify waste in dollar amounts, propose validated spending policies, and verify improvements with before-and-after comparisons.

Some of the key features are:

• Local-First Audits: Run unlimited audits on local machine data without creating an account or sending sensitive information to a cloud service.
• Spend Mapping: Visualize expenses by tracing every call, tool, and token to the specific agent or workflow that caused it.
• Waste Classification: Identify and categorize waste, including retry-waste, loop-waste, and context bloat, with associated dollar amounts.
• Policy Recommendations: Generate draft spending policies and optimization recommendations validated through actual usage data.
• Agent Integration: Seamlessly integrate with popular tools like OpenClaw, Hermes, Claude Code, and Cursor via CLI, MCP, or custom schemas.
• Team Collaboration: Utilize hosted workspaces to share audit maps, track spend trends over time, and manage enterprise-grade security features like SSO/SAML.

The system is designed to be operational from install to governance. Developers begin by installing the CLI, which allows them to run an immediate audit. Once they have collected data, they can push these audits to a secure workspace to collaborate with their team. The platform also provides an MCP server and Slack integrations for Pro users, facilitating continuous monitoring and automated alerts regarding agent spend. Enterprise teams can further leverage self-hosted deployment options, custom retention policies, and SLAs to support larger, complex AI infrastructures.

Some common use cases include:

• Reducing Agent Overhead: Identifying and eliminating repetitive API calls or bloated context windows that drive up inference costs.
• Monitoring Multi-Agent Fleets: Providing engineering leads with a centralized dashboard to track costs across diverse agents running in different development environments.
• Automating Cost Governance: Implementing CI/CD gates that fail builds or deployments when the estimated cost or waste threshold for an agent is exceeded.
• Cost-Optimization Research: Testing different prompt or retrieval strategies and using Xerg to measure the economic impact of these changes on agent performance.