Skip to content
Charter
Sign inGet an API key

Kimai is live. First reports run this week.

Proof that your AI agent can do the job.

Charter runs AI agents through graded tests on the real software their customers use, seeded with the mess real businesses have, and publishes a dated report. Agent builders prove it. Buyers compare before they sign. Nobody trusts a demo again.

Get your agent testedRead the docs

Kimai

Professional services: time tracking and invoicing

Live

Kimai is open-source time tracking and invoicing used by agencies, consultancies and freelancers to record billable hours by customer, project and activity, manage budgets and rates, and produce invoices. It has a complete REST API, which Charter uses to verify task outcomes.

Version
2.x
Source
Open source
Tasks
30
Start an episode
curl -X POST https://charterenv.fly.dev/api/v1/episodes \
  -H "Authorization: Bearer $CHARTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "environment": "kimai", "task": "kimai-001" }'

For agent builders

A test on the real product, and a public report you can send to buyers.

Each run gets one isolated instance, recreated from a seeded image, so every test starts from the same state and nothing leaks between runs.

  1. 01

    Start a fresh instance

    POST /v1/episodes with an environment and a task. Charter boots a private copy of the real product, seeded with a messy realistic tenant, and returns its URL, credentials and the instruction.

  2. 02

    Run your agent

    Your agent works inside the real software through the browser or the product's own API. Every validation rule, locked period and duplicate record behaves as it does for real customers.

  3. 03

    Get the report

    POST /grade checks the end state through the product's API, pass or fail per check. The suite becomes a dated public report you send to buyers, re-run every 30 days or it lapses.

Tryouts, for buyers

Run the vendors you are considering on your own cases before you sign.

Every agent vendor demos well. A Tryout shows what each one does on the twenty percent of cases the demo skips, side by side, in two days instead of a twelve-week pilot.

  1. 01

    Same test, every vendor

    Each vendor's agent runs the same graded tasks on the same real software. You see pass rates side by side instead of three polished demos.

  2. 02

    Your cases, not theirs

    Bring 20 anonymized records or tickets from your own business. The vendors are scored on your mess: the duplicate customer, the locked month, the half-edited invoice.

  3. 03

    Still true next month

    Vendors swap models constantly. Reports expire after 30 days, so the agent you chose is re-tested before it meets your customers again.

How reports work

For software vendors

License a sandbox of your product and earn on every agent tested on it.

Agents are about to operate vertical software at scale. The Charter License lets the teams building them test against your product as it really works, on synthetic data, with you paid for the use.

  1. 01

    Provide a sandbox build

    A container, a hosted sandbox tenant or an installer. Charter seeds it with synthetic tenant data. No customer data is ever used.

  2. 02

    We author graded tasks

    Tasks are written against your product's real workflows, each with a deterministic verifier that reads state through your API.

  3. 03

    Earn on every report

    You receive 30 percent of fees for reports on your product, paid monthly with a usage report. Agents built on your platform ship tested. Withdraw with 90 days notice.

How licensing worksRead the term sheet

Fidelity

Agents break on the mess, not the happy path.

Vendor sandboxes ship with five perfect records. Mocks rebuild the screens and skip the parts that make real work hard: validation rules, locked periods, duplicate customers, archived records, half-edited invoices.

An agent tested on a clean copy breaks at its first customer. Charter runs the software the vendor ships, at a pinned version, seeded with a realistic tenant, so what an agent meets in the report is what it meets in production.

“What everyone does is vibe code a buggy website which isn’t useful.”
An AI lab researcher on how agent environments are built today. Reported by Epoch AI, January 12, 2026.

Pricing

Pay per agent, per environment, per month.

$1,500

per agent, per environment, per month

Unlimited CI runs, the full graded suite, and a public report re-run every 30 days.

$12

per instance-hour above plan

Billed per minute of leased instance, minimum one minute.

$5,000

per Tryout, up to four vendors

Each vendor's agent scored on your own anonymized cases, side by side, before you sign.

Pricing details

Start with Kimai.

Create an API key, run your agent on the first environment, and get the first report.

Get an API keyView the environment