Averth

Cost analysis for AI agents

Know what your
AI agent runs cost.

Give Averth a batch of runs. See the cost per usable result, what drove it, and which past runs a policy would have stopped.

Get started

Available as a ChatGPT plugin. Open source on GitHub

What Averth does

From run records to a cost breakdown.

  1. 01

    Analyze the costs

    Models, retries, tools, and human review. Find what each accepted outcome cost.

  2. 02

    Replay a policy

    Test cost or token-yield thresholds against past runs. See the spend involved and successful runs caught in the cutoff.

  3. 03

    Understand the layers

    Explore context growth, tool spend, token yield, costly outliers, and model and human contributions.

Analysis of data you supply. No automatic account access or live budget enforcement.

New: evidence verifier

Check what the evidence supports.

V0 judges the checks you write against the runs and artifacts you supply, and tells you what the evidence supported for each one, what failed, and what was never observed or could not be verified.

Write a check, supply the runs and the repo snapshot, get a verdict with the evidence cited. Full detail in Docs.

Start here

Bring a batch of runs.

Use the ChatGPT plugin

With Averth enabled in ChatGPT, supply a JSON array of sanitized run records or an Averth ledger export. Start with:

Analyze these agent runs. Show cost per accepted outcome, the cost breakdown, and the expensive tail. Treat missing outcomes as unknown.
Keep your traces local

Use the open-source CLI to process a trace export on your own machine.

pip install git+https://github.com/Wesley3141/Averth.git
averth trace runs.json --format langfuse
What should I share?

Sanitized cost and outcome metadata only. Remove customer identifiers, sensitive free text, credentials, and personal data. Averth processes submitted ledgers in memory without storing them; ChatGPT conversation retention is separate.

For enterprise teams

See Averth with your team.

Walk through cost analysis and historical policy replay for your agent workflows.

Request a demo