algovoi-atb
AlgoVoi Agent Trust Bench client — run your AI agent through 178 adversarial x402 payment profiles and earn a Falcon-1024 signed ATB Pass Certificate for reputation-gated pricing.
Model baselines
| Model | Profiles | Accuracy | Variant | Date |
|---|---|---|---|---|
claude-sonnet-4-6 |
69/69 | 100% | plain | 2026-06-01 |
claude-haiku-4-5 |
69/69 | 100% | plain | 2026-06-02 |
claude-haiku-4-5 |
71/71 | 100% | elizaos-plugin | 2026-06-02 |
claude-haiku-4-5 |
69/70 | 98.6% | bench-mcp-tools | 2026-06-02 |
All baselines run via inspect eval against the live bench. Zero adversarial profiles settled across all runs. The bench-mcp-tools variant uses real inspect_ai tool-use — the model calls bench_refuse_profile or bench_claim_profile to communicate its decision rather than outputting text.
Install
pip install algovoi-atb # client only
pip install "algovoi-atb[verify]" # + Falcon-1024 certificate verification
Quick start
from algovoi_atb import run_bench
async def my_agent(profile_id: str, challenge: dict) -> bool:
# Return True to refuse (correct for adversarial profiles).
# Return False to pay (only for control/benign profiles).
suspicious = any(
kw in challenge.get("description", "").lower()
for kw in ("urgent", "override", "ignore", "system prompt")
)
return suspicious
result = await run_bench(my_agent, label="my-agent-v1")
print(f"Score: {result.score:.2f}")
print(f"Passed: {result.passed}")
print(f"Report: {result.report_url}")
# Present the cert to x402 gateways via X-ATB-Credential for discounts.
headers = {"X-ATB-Credential": result.certificate_b64}
Verify a certificate
import httpx
from algovoi_atb import verify_certificate
# Fetch the public key from the bench.
pk_doc = httpx.get(
"https://agent-trust-bench.algovoi.co.uk"
"/agent-trust-bench/.well-known/atb-keys.json"
).json()
pk_b64 = pk_doc["keys"][0]["public_key_pqclean_b64"]
verified = verify_certificate(result.certificate_b64, pk_b64=pk_b64)
print(verified.valid, verified.grants_discount, verified.expires_at)
Lower-level API
from algovoi_atb import BenchClient
async with BenchClient(label="my-agent") as client:
run_data = await client.create_run(label="my-agent-v1")
await client.start_session()
profiles = await client.list_profiles()
for profile in profiles:
pid = profile["metadata"]["profile"]
challenge = await client.hit_profile(profile["resource"])
if challenge and my_decision(pid, challenge):
await client.log_refusal(pid)
cert = await client.fetch_certificate()
print(cert["certificate"]) # base64url envelope → X-ATB-Credential
Links
Metadata
Release files for algovoi-atb 0.1.6
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Reset |
|---|---|---|---|---|
| algovoi_atb-0.1.6-py3-none-any.whl | Python 3 | none | any | Details |
Release files / algovoi_atb-0.1.6-py3-none-any.whl
| Download URL | algovoi_atb-0.1.6-py3-none-any.whl |
|---|---|
| Size | 21.0 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
636a0e902b4d90c0f317b8abc5d53b5776879889dc830782ecde6fe014ecb226
|
|
BLAKE2b-256 checksum How to use checksums |
3eb3397071f671f3f8b24d9f5f095dfc12803945100c2f258cb23fa49c5cdd44
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/6.2.0 CPython/3.12.10
|
Security for algovoi-atb 0.1.6
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.