The Economic Control Plane for AI Agent Fleets

See which agents are pulling their weight, where spend is wasted, and what to fix next.

Run your first audit

Detects local OpenClaw, Hermes, and Claude Code data. Cursor users: audit a usage CSV export.

Using a coding agent? Install the skill

Then ask your agent to audit your AI spend — it runs Xerg and explains the findings.

Backed by

a16z Scout, the NVIDIA Inception Program, and Cloudflare Launchpad

Same source · latest 2 audits

nemoclaw-56d456

OpenClaw Agent - Outbound Marketing

Improving
Baseline
Mar 1 - Mar 31
Total spend $28.4k
Waste $8.2k
Waste rate 28.9%
Current
Apr 1 - Apr 30
Total spend $24.9k
Waste $5.4k
Waste rate 21.8%
Waste delta -$2.8k 34.1% less waste
Waste rate delta -7.1% down vs baseline
Spend delta -$3.5k lower total spend
Top workflow reply_router
Top issue context bloat
Change trim reply context + cap retrieval
Result $2.8k less waste vs baseline

Works with these agent frameworks

How it works

One command.
Sixty seconds to first signal.

01

Run your first audit

Xerg finds your local OpenClaw, Hermes, or Claude Code data, runs the audit locally, and prints the report right away. No account required. Nothing leaves your machine unless you choose to push.

02

Fix one thing and compare

Take the top recommendation, make one workflow or model change, then re-run to see the before/after dollar delta against your saved local baseline.

OPTIONAL
03

Push to the dashboard

Pro

With a paid Pro plan, connect your Xerg account and push audits for shared history, daily trends, and Ask Xerg.

What Xerg does

Finds confirmed waste. Suggests the first fix.
Building toward policy controls.

Every tool shows a spend total. Xerg shows where dollars are wasted, what to test first, and how the numbers change when you re-run.

01

Seven spend categories

Xerg surfaces seven finding types in dollars. Retry waste and loop waste are confirmed waste. Context bloat, downgrade candidates, idle spend, cache carryover, and max-mode concentration are savings opportunities to test.

02

Dollars, not tokens

Token counts are telemetry. Dollar impact is priority. Xerg reports spend, confirmed waste, and opportunity impact in USD.

03

Local-first audits

No account required for local audits and compare. Xerg runs locally against your OpenClaw, Hermes, or Claude Code data. Nothing leaves your machine unless you choose to push.

04

Built for agent workflows

Agent runs have measurable economics: logs, calls, retries, loops, and outcomes. Xerg turns that runtime data into waste signals you can act on.

05

Optional hosted workspace

Push results when you want shared history, dashboard visibility, and hosted follow-up. Hosted features are optional and separate from local audits.

Coming next

Policy controls

We’re building toward budget guardrails and automated controls like loop protection and model guardrails.

Start finding waste today.

The CLI is free and runs locally. Upgrade for hosted visibility,
historical tracking, and economic governance.

View pricing