eqbuilder

Score your AI agent's replies for human-likeness before sending — and check they stick to your approved facts. No account, no API key. Get 3 lifetime free scores per caller with strict data_consent=True. Free score text and its result may be retained under a one-way pseudonymous caller key for calibration, so never submit secrets or sensitive personal data. Follow-up contact and wishlist feedback are optional and separate. Then score through POST /api/simulate with Base USDC via x402: the buyer wallet needs USDC only, no ETH.

# pip install eqbuilder
from eqbuilder import EqBuilder

eq = EqBuilder(data_consent=True)
result = eq.free_score(
    "Sounds good — I'll double-check the invoice and get back to you.",
    delay_seconds=4.2,
    profile="analytical_executive",
)
print(result["eq_score_percentage"], result.get("suggested_prompt_addition"))

Free trial note: Every free score requires data_consent=True. EqBuilder defaults to data_consent=False; opt in explicitly at construction or on an individual free_score(..., data_consent=True) call. A false or non-boolean value is rejected before a free round is consumed. All three free scores follow the same path. Optional product feedback can be sent separately with eq.wishlist("...").

Set and forget (one standing call)

# pip install "eqbuilder[payments]"
eq = EqBuilder(wallet_key="0x<private key of a wallet holding USDC on Base>")
result = eq.auto_score(
    "Sounds good — I'll double-check the invoice and get back to you.",
    delay_seconds=4.2,
    profile="analytical_executive",
)
print(result["paid"])  # False while free scores last, True once it graduates to x402

auto_score needs only explicit free-data consent plus an authorized wallet. It uses your free trial while it lasts, then transparently retries the same input through the paid x402 path on 429 free_trial_exhausted. Without a wallet_key it raises PaymentRequired with one machine-readable developer handoff containing the live quote path, Base USDC funding requirement, and card alternative. Free and paid results have slightly different shapes; the added "paid" key tells you which path ran. If a paid call times out, the SDK does not automatically replay the paid authorization. Preserve the original error/receipt and check payment status or contact the operator before creating a fresh authorization.

If you decline free-trial retention (data_consent=False), call score() directly for a paid evaluation; paid scoring does not silently inherit the free consent.

LangChain (one line)

# pip install "eqbuilder[langchain]"
from eqbuilder.langchain_tool import get_langchain_tools

tools = get_langchain_tools(data_consent=True)  # add wallet_key="0x..." for paid fallback

CrewAI

# pip install "eqbuilder[crewai]"
from eqbuilder.crewai_tool import HumanLikenessScoreTool

agent = Agent(role="support", tools=[HumanLikenessScoreTool()], ...)
# The tool schema requires data_consent=true for a free call.

Everything at once: pip install "eqbuilder[all]" (payments + LangChain + CrewAI).

Paid scoring (after the 3 free scores)

# pip install "eqbuilder[payments]"
eq = EqBuilder(wallet_key="0x<private key of a wallet holding USDC on Base>")
result = eq.score("...", delay_seconds=4.2, profile="analytical_executive")
deep   = eq.score("...", delay_seconds=4.2, profile="analytical_executive", tier="deep")

Payment is canonical x402 v2: score() calls POST /api/simulate, reads its live HTTP 402 quote, chooses Base USDC, signs a gasless EIP-3009 transfer authorization, and retries once with PAYMENT-SIGNATURE. The success response contains the result and a PAYMENT-RESPONSE settlement receipt. The live quote is authoritative for amount and terms; do not hardcode an approximate price. SOL and X-PAYMENT remain compatibility paths, not this SDK quickstart.

Paid text does not enter the free consented corpus. Its share_for_calibration option is a separate opt-in; when true, eligible high-scoring text may be retained for manual operator calibration review.

Privacy-safe corpus evidence

evidence = eq.corpus_evidence()  # aggregate counts only; no raw text or identities

The stable public evidence endpoint is https://eqbuilder.dev/api/corpus/evidence (crawler alias: https://eqbuilder.dev/.well-known/corpus-evidence.json). It reports thresholded aggregate corpus and improvement evidence, not raw submissions or guaranteed model improvement.

Script adherence (paid)

check = eq.script_check(
    reply_text="The plan is $49/mo, see example.com/pricing",
    approved_facts=["Plan price is $29/mo", "Docs: https://docs.example.com"],
)
# flags the invented $49 price and the off-script URL

Improvement loop

def rewrite(text, suggestion):
    return my_llm(f"Rewrite this reply. {suggestion}\n\n{text}")

final = eq.improve(
    "...",
    4.2,
    "analytical_executive",
    rewrite,
    max_rounds=3,
    data_consent=True,  # required when using the free improvement loop
)

• 28 personality profiles: eq.profiles()
• Pricing: eq.pricing()
• Full API & MCP endpoint: https://eqbuilder.dev/integrate