Capability tests
Does the system do the task it claims to do, under the stated conditions?
Diagnostic audit · founding lane
Find out whether your AI system actually works.
Give GLEE a repository, endpoint, agent card, or evaluation trace. We attack it as an independent evaluator and return a report you can inspect, act on, and publish with its limits attached.
Start the $19 probe
Inquire about the $99 founding audit ↗
See the request brief
Inspect frozen Bench releases
Request-only today. Your agent works in the demo. Does it survive reality? The $19 probe uses manual Base-USDC settlement today. A separate $0.50 machine probe is already priced and negotiable — but its payment settlement and execution are not yet released to buyers. Do not include credentials or signing material in the request.
Audit output
The report follows the system’s real failure modes.
Scope is frozen before work starts. Findings carry the evidence needed to reproduce or challenge them.
Tool use and state
Where do tools, memory, context, or state transitions lose the thread?
Reliability and recovery
What fails on retries, partial completion, timeouts, or realistic edge cases?
Efficiency per closure
What did the system spend in time, tokens, retries, and human attention to finish?
Security boundaries
Which trust, permission, secret, or input boundaries deserve a closer look?
Reproducible evidence
Receive test inputs, observed failures, limitations, and a receipt-style handoff.
Start small
Buy an output, not a philosophy.
Each door is sized to answer one commercial question: is this system worth trusting, fixing, comparing, or calling?
$19 USD
One narrow question, one bounded test profile, concise findings, and a next move.
Pay and start the probe$99 USD
A serious first-system diagnostic across capability, failure, state, recovery, efficiency, and trust boundaries.
Inquire about the founding audit ↗$250 USD
Human/agent collaboration, broader test coverage, more reproduction, and deeper comparison.
Request deep review ↗From $500 USD
Bespoke benchmark design, multi-system comparison, or an experimental evaluation with a defined protocol.
Request a custom scope ↗First transaction · manual rail
Send $19, then send the question.
The first probe uses direct Base-USDC settlement so the payment can be independently verified against GLEE’s designated receiving address. There is no hosted checkout here, and this offer does not depend on the machine lane’s x402 pricing — that lane is priced but not yet open for payment.
19 USDC on Base
- Network
- Base mainnet · chain ID 8453
- Token
- USDC
- Token contract
0x833589fCD6eDb6E08f4c7C32D4f71b54bdA02913- Receiving wallet
0xf409052bF8B6bd53B238791c678Ab7b4777175a5
Check before sending. Confirm the network, token contract, amount, and receiving address in your wallet. Never send credentials or signing material.
Make the brief actionable.
- Send exactly 19 USDC on Base mainnet.
- Open the official request path.
- Put the transaction hash first, then include the target, question, and test profile.
Manual verification. GLEE verifies the hash, chain, token, amount, recipient, and confirmation before starting. A transfer alone is not a completed audit.
Copy this into the request
Make the first paid question concrete.
The account-backed request lets GLEE price the first useful chunk before any detailed teardown begins.
GLEE Bench Quick System Probe
Target: [repo, endpoint, agent card, or evaluation trace]
Question: What should this system reliably do?
Profile: [capability / tools / memory / recovery / efficiency / security]
Assertions: [what you believe is true, if any]
Useful output: evidence-backed findings with reproducible cases
Budget: $19 USD quick probe
Payment: 19 USDC on Base mainnet
Transaction hash: [paste the canonical transaction hash first]
Authority: [who can approve the invoice]Machine lane · pricing live, execution gated
A machine can ask the same question for cents.
POST /bench/probe is the machine interface: send a target, test profile, and optional assertions; receive executed tests, failures, latency, conformance, evidence, and a verifiable receipt.
Priced, not yet open for payment. The endpoint is live and answers with a real x402 challenge — $0.50 in USDC on Base — and the payment address is armed. What is not yet released is what comes after: settling the payment and running the probe. Our own canary deliberately stops at the challenge, so we will not take funds on a path we have not run end to end. Do not send payment yet.
Boundary
Diagnostic, not certification.
GLEE Bench reports what was tested, what failed, what was observed, and what remains unknown. It does not confer regulatory approval, security certification, or a permanent quality label.
Customer material stays within the agreed transfer path. Do not paste credentials, private keys, wallet recovery words, or secrets into the request.
Every claim about the audit earns its status from fresh evidence. A plan, a payment, or a passing unit test alone is not a verified system result.