# Attify > Attify independently audits what AI agents actually achieve, using real data from the companies that use them. AI agent verification: each job type an AI agent does gets a published standard that asks of a claimed outcome: did it happen, did it stick, did a human have to redo it. Attify reports the audited rate next to the vendor's claimed rate, with 95% intervals, against the fairest version of the vendor's own counting rule. Every result opens with a one-sentence verdict generated from the numbers by a fixed template, never by an LLM. Attify (attify.app) is not affiliated with attify.com. ## Standards - [Standard #1: Attify Audited Resolution Standard v0.11](https://attify.app/standards/support-resolution.md): Draft, open for comment. Every version at its permanent address (https://attify.app/standards/support-resolution/v/{version}); JSON with changelog: https://attify.app/api/v1/standards/support-resolution - [Standard #2: Attify Audited Meeting Standard v0.6](https://attify.app/standards/sales-meetings.md): Draft, open for comment. Every version at its permanent address (https://attify.app/standards/sales-meetings/v/{version}); JSON with changelog: https://attify.app/api/v1/standards/sales-meetings - [Standard #3: Attify Audited Booking Standard v0.6](https://attify.app/standards/bookings.md): Draft, open for comment. Every version at its permanent address (https://attify.app/standards/bookings/v/{version}); JSON with changelog: https://attify.app/api/v1/standards/bookings - [Program rules](https://attify.app/standards/program-rules.md): who counts toward an audited score, the minimums, publishing, statuses, revocation. - [How to cite a standard](https://attify.app/standards/how-to-cite), [releases (RSS)](https://attify.app/standards/feed.xml), the standards' defined terms with anchors: GET https://attify.app/api/v1/definitions. - Comments on a standard: email standards@attify.app (confidential on request); published responses at https://attify.app/standards/{job_type}/comments/{version}, also as comments.json. ## About Attify - [About Attify](https://attify.app/about.md): Attify independently audits what AI agents actually achieve, using real data from the companies that use them. - [Dor Tagger](https://attify.app/authors/dor-tagger.md): Founder of Attify. - [Corrections policy](https://attify.app/corrections.md): How to report an error in an Attify result, standard or page, how Attify investigates, and how corrections are published with the old and new numbers. - [Is Attify independent?](https://attify.app/independence.md): Who pays Attify, why no price depends on a result, what Attify will not do for a vendor, and what its method cannot see. - [Attify Index methodology](https://attify.app/index/methodology.md): How the Attify Index pools audited AI agent outcomes by job type each month, from contributing buyers only, and the anonymity rules every edition passes. - [Methodology](https://attify.app/methodology.md): How Attify checks an AI agent's claimed outcomes against the customer's own records, under a published standard per job type, with 95% intervals. - [Press](https://attify.app/press.md): Attify boilerplate, facts, how to cite its standards and results, and the press contact. - [Right of reply for vendors](https://attify.app/right-of-reply.md): How a vendor named on an Attify page or score can reply publicly, how fast Attify publishes the reply, and what a reply can and cannot change. ## Glossary - [Agent washing](https://attify.app/glossary/agent-washing.md): Agent washing is marketing ordinary automation (chatbots, scripts, assistants) as autonomous AI agents, or overstating what an agent does. The US FTC treats unsupported AI claims as deceptive. - [AI agent verification](https://attify.app/glossary/ai-agent-verification.md): AI agent verification: checking the outcomes an AI agent claims against the customer's own records, under a published standard, with stated uncertainty. - [Assumed resolution](https://attify.app/glossary/assumed-resolution.md): An assumed resolution is a conversation counted as resolved because the customer left without asking for more help, not because anyone confirmed it. - [Audited resolution](https://attify.app/glossary/audited-resolution.md): An audited resolution is a conversation an AI agent claimed as resolved that passes three checks: no repeat contact in 7 days, no handoff, and a judge's confirmation. - [Automated resolution](https://attify.app/glossary/automated-resolution.md): An automated resolution is a customer request an AI agent handled without a human. Zendesk and Ada use the term, and Zendesk bills per automated resolution. - [Booking accuracy](https://attify.app/glossary/booking-accuracy.md): Booking accuracy is whether an AI receptionist booked what the customer asked for: the right service, time, provider, location and price, without staff having to fix it. - [Booking rate](https://attify.app/glossary/booking-rate.md): Booking rate is the share of calls or conversations that ended in a booked appointment. Its denominator decides the number, and the agent being measured may classify the calls itself. - [Confidence interval (95%)](https://attify.app/glossary/confidence-interval.md): A 95% confidence interval is the range of values consistent with a measured rate given its sample size. Attify prints one next to every rate and every gap. - [Containment rate](https://attify.app/glossary/containment-rate.md): Containment rate is the share of conversations an AI agent handled without passing them to a human, whether or not the customer's issue was resolved. - [Cost per audited outcome](https://attify.app/glossary/cost-per-audited-outcome.md): Cost per audited outcome is what you paid for an AI agent in a month divided by the outcomes that held up in an audit, rather than the outcomes the vendor claimed. - [Customer-confirmed vs. customer satisfaction](https://attify.app/glossary/customer-confirmed-vs-satisfaction.md): Customer-confirmed means the customer explicitly answered that the issue was resolved. A satisfaction rating (CSAT) is a separate signal and never counts as confirmation. - [Customer independence](https://attify.app/glossary/customer-independence.md): Customer independence is the rule that only genuine, separate customers count toward a vendor's audited score: four checks stop a vendor from posing as its own customer. - [Data provenance: connected, witnessed, uploaded](https://attify.app/glossary/data-provenance.md): Data provenance says where the data behind an audited result came from: read from the customer's own systems (connected), streamed by the vendor as it happened (witnessed), or a file (uploaded). - [Deflection rate](https://attify.app/glossary/deflection-rate.md): Deflection rate is the share of customers who found help without opening a ticket. It measures avoided tickets, not resolved issues, and has no single standard formula. - [Evidence tier](https://attify.app/glossary/evidence-tier.md): An evidence tier says how strong the proof behind an audited number is: rules only, judge-confirmed, or confirmed by the customer or the customer's records. - [First contact resolution (FCR)](https://attify.app/glossary/first-contact-resolution.md): First contact resolution is the share of issues resolved in the first contact, measured by customer survey or by the absence of a repeat contact within a window. - [Handoff](https://attify.app/glossary/handoff.md): A handoff is when an AI agent passes the customer to a person or another channel instead of finishing the job. Vendors differ on which handoffs void a resolution. - [Human audit](https://attify.app/glossary/human-audit.md): A human audit is people labelling a random sample of what an LLM judge decided, blind to the judge, so the judge's agreement with people is measured rather than assumed. - [Judge agreement](https://attify.app/glossary/judge-agreement.md): Judge agreement is how often an LLM judge's decisions match people's labels on the same records, reported as a percentage with an interval and as Cohen's kappa. - [LLM-as-a-judge](https://attify.app/glossary/llm-as-a-judge.md): LLM-as-a-judge is using a language model with a written prompt to grade another system's output. Its verdicts are only as trustworthy as its measured agreement with people. - [Meeting held rate](https://attify.app/glossary/meeting-held-rate.md): Meeting held rate is the share of booked sales meetings that actually took place. For AI SDRs it is read from calendar, video-call and CRM records, never the vendor's dashboard. - [Meeting-to-opportunity rate](https://attify.app/glossary/meeting-to-opportunity-rate.md): Meeting-to-opportunity rate is the share of held sales meetings that became a pipeline opportunity. Attify reports it as the qualified tier, never as the headline. - [Meetings booked](https://attify.app/glossary/meetings-booked.md): Meetings booked is the count AI SDR vendors report: meetings an AI agent scheduled. Vendors count at booking, before anyone knows whether the meeting will happen. - [No-show rate](https://attify.app/glossary/no-show-rate.md): No-show rate is the share of booked appointments or meetings the customer missed without cancelling. It varies about tenfold between sectors, so compare within a business, not across them. - [Outcome-based pricing](https://attify.app/glossary/outcome-based-pricing.md): Outcome-based pricing bills for an AI agent's results (a resolution, a booked meeting) instead of seats or usage. The vendor's definition of the outcome sets the bill. - [Reconciliation](https://attify.app/glossary/reconciliation.md): Reconciliation is the customer's monthly check that data its vendor streamed to Attify matches its own records: a total, then a seeded random spot check. - [Repeat contact rate](https://attify.app/glossary/repeat-contact-rate.md): Repeat contact rate is the share of conversations followed by another contact from the same customer within a window, such as 7 days. It catches resolutions that did not hold. - [Resolution rate](https://attify.app/glossary/resolution-rate.md): An AI support agent's resolution rate is the share of conversations it counts as resolved. Each vendor defines resolved differently, so rates are only comparable under one rule. - [Show rate](https://attify.app/glossary/show-rate.md): Show rate is the share of scheduled meetings or appointments where the other person attended. Published benchmarks are thin and mostly come from scheduling vendors' own data. ## Guides - [How to audit conversations your AI agent marked resolved](https://attify.app/guides/audit-ai-resolved-conversations.md): A step-by-step audit of AI-resolved support conversations: draw a seeded random sample, label it blind with written rules, measure agreement, and report the share with an interval. - [How to calculate cost per audited outcome](https://attify.app/guides/calculate-cost-per-audited-outcome.md): Divide what you paid for an AI agent in a month by the outcomes that passed an audit, not the ones it claimed. Formulas for per-outcome, per-interaction, seat and flat pricing. - [How to check whether your AI SDR's meetings actually happened](https://attify.app/guides/check-ai-sdr-meetings-happened.md): Match every meeting an AI sales agent counted as booked to your own calendar and CRM, and check who booked it, who the prospect was, whether it was cancelled and who attended. - [How to check an AI vendor's claims before buying](https://attify.app/guides/check-ai-vendor-claims-before-buying.md): Get the outcome definition in writing, ask for results from customers' own systems with sample size and period, test on your own data, and put audit rights in the contract. - [How to run an AI agent pilot and decide on evidence](https://attify.app/guides/evaluate-ai-agent-pilot.md): Set the outcome rule and the pass mark before the pilot starts, keep a control group, measure in your own systems, and wait for outcomes to settle before deciding. - [How to measure your AI receptionist's real bookings](https://attify.app/guides/measure-ai-receptionist-bookings.md): Check every appointment your AI receptionist counted as booked in your own booking system: did the AI book it, did staff fix it, was it cancelled, did the customer come. - [How to measure your AI support agent's real resolution rate](https://attify.app/guides/measure-ai-support-agent-resolution-rate.md): Check the conversations your AI agent counted as resolved for repeat contacts within 7 days, handoffs, and a sampled read of whether the issue was solved. - [How to prove AI agent ROI to finance](https://attify.app/guides/prove-ai-agent-roi-to-finance.md): Build the ROI case on audited outcomes, not claimed ones: what the agent cost, what the outcomes that held would have cost otherwise, and what failed claims cost to clean up. - [Questions to ask before renewing an AI agent contract](https://attify.app/guides/questions-before-renewing-ai-agent-contract.md): Before you renew an AI agent, answer from your own records: what share of its outcomes held up, what each audited outcome cost, where it failed, and whether that is changing. - [How to send your AI agent's data to Attify](https://attify.app/guides/send-agent-data-to-attify.md): Four ways to get an AI agent's interactions to Attify: connect the customer's own systems, connect an agent platform account, stream through the Ingest API or SDKs, or send OpenTelemetry traces. - [How to start verification for your AI agent](https://attify.app/guides/start-ai-agent-verification.md): Open a verification program, invite independent customers to connect their own systems, meet the minimums, and choose whether to publish your audited score, with a live badge and a signed attestation. - [What each AI agent platform counts as an outcome](https://attify.app/guides/what-platforms-count-as-an-outcome.md): How AI support, sales and receptionist platforms define the outcome they report and bill: a resolution, a booked meeting, a booked appointment, from their own documentation. - [Why vendor-reported AI agent metrics differ from what happened](https://attify.app/guides/why-vendor-ai-metrics-differ-from-reality.md): Vendors count outcomes at the earliest moment, treat silence as success, use short windows and their own graders. Each choice is reasonable alone; together they open a gap. ## Ways to verify an AI agent, compared - [AI agent benchmarks vs. production verification](https://attify.app/compare/agent-benchmarks.md): Public agent benchmarks such as tau-bench, tau2-bench and CRMArena-Pro test models and agent setups on simulated users and sandboxes. They do not measure a vendor's deployment. - [How AI receptionist vendors define a booking](https://attify.app/compare/ai-receptionist-booking-definitions.md): AI receptionist and voice-agent vendors bill per call, minute, caller, location or contract, and apart from ServiceTitan none of the pages read defines a booking. Marketplaces that bill per booking draw the line later. - [AI risk and security certification vs. outcome verification](https://attify.app/compare/ai-risk-certification.md): AIUC-1, SOC 2 and ISO/IEC 42001 check an AI agent's safety, security and management controls. Outcome verification checks whether the agent's claimed outcomes held up. - [How AI SDR vendors define a booked meeting](https://attify.app/compare/ai-sdr-meeting-definitions.md): How AI sales agent vendors count and bill a meeting, from their own documentation: at booking, by status, by reply label or by attribution window. Few bill per meeting. - [Analyst reports and review sites vs. audited outcomes](https://attify.app/compare/analyst-reports-and-reviews.md): Forrester Waves score vendors from questionnaires, demos and references; G2 reviews collect moderated user opinion. Neither is a measured outcome rate with a denominator and an interval. - [Booked vs. held vs. qualified meetings](https://attify.app/compare/booked-vs-held-vs-qualified-meetings.md): A booked meeting is on the calendar, a held meeting took place with the prospect, a qualified meeting led to an opportunity. AI SDR vendors count the first; sales teams are measured on the others. - [COPC certification vs. AI agent verification](https://attify.app/compare/copc.md): The COPC CX Standard sets management requirements for customer experience operations, human and AI. Verification measures one AI agent's outcomes against a published standard. - [LLM evaluation and observability platforms vs. outcome verification](https://attify.app/compare/evals-and-observability.md): Braintrust, LangSmith, Langfuse and Arize let teams score their own agent with evaluators they choose. Outcome verification checks claimed outcomes under a published rule, in the customer's records. - [AI agent pilots and holdout tests vs. standing verification](https://attify.app/compare/pilots-and-holdouts.md): A pilot with a control group shows whether an AI agent beats your current process for a few weeks. Standing verification keeps measuring its outcomes after the pilot ends. - [Pre-purchase AI agent evaluations vs. production verification](https://attify.app/compare/pre-purchase-evaluations.md): Evaluations such as G2 Agent Evaluations and Vals AI test agents on private tasks in a controlled environment before purchase. Production verification checks a deployed agent's real outcomes. - [Conversation QA tools vs. AI agent verification](https://attify.app/compare/qa-tools.md): QA tools such as Zendesk QA and Observe.AI score every conversation against a rubric the customer writes. Verification checks whether claimed outcomes held, under a published standard. - [Resolution vs. containment vs. deflection](https://attify.app/compare/resolution-vs-containment-vs-deflection.md): Deflection counts tickets never opened, containment counts AI conversations never handed to a human, resolution counts conversations the vendor's rule calls resolved. Only an audited resolution checks the issue stayed solved. - [Modeled ROI studies vs. audited outcomes](https://attify.app/compare/roi-studies.md): A Forrester Total Economic Impact study models the financial impact of a technology on a composite organization from customer interviews. Audited outcomes measure what one deployment achieved. - [How AI support agents define and bill a resolution](https://attify.app/compare/support-ai-agent-outcome-definitions.md): Each AI customer-support vendor's own definition of a resolution: the window, which handoffs void it, whether the customer confirms, and what is billed. From their documentation, dated. - [Vendor dashboards vs. independent verification](https://attify.app/compare/vendor-dashboards.md): An AI vendor's dashboard reports outcomes under the vendor's own definition, from the vendor's platform. Independent verification tests that count against the customer's own records. ## Research - [Research: how AI receptionists count and bill a booking](https://attify.app/research/bookings.md): AI receptionist and voice-agent vendors bill per call, minute, caller, location or contract, and all but one define no booking; per-booking marketplaces draw the line later. - [Research: how AI sales agents count and bill a meeting](https://attify.app/research/sales-meetings.md): AI SDR vendors count a meeting at booking and bill seats, credits or activity; 9 of the 13 per-meeting firms whose terms were read tie the fee to a held meeting. Read 31 October 2026. - [Research: how AI support vendors define and bill a resolution](https://attify.app/research/support-resolution-definitions.md): What 14 AI customer-support vendors' own documents say a resolution is, when it is decided, what voids it and how it is billed, read on 8 November 2026. - [Research: other ways to verify an AI agent](https://attify.app/research/ways-to-verify-ai-agents.md): What benchmarks, evaluations, certifications, ROI studies, reviews, QA tools, eval platforms, dashboards and pilots each measure, from their own documentation. ## Calculators - [Audit sample size calculator](https://attify.app/tools/audit-sample-size.md): How many AI conversations, meetings or bookings to label to estimate a share within a margin you choose, and how wide the interval is for a sample you already have. - [Cost per audited outcome calculator](https://attify.app/tools/cost-per-audited-outcome.md): What each real outcome from your AI agent costs: enter what you pay, what it claimed and what share passed the audit, and see the cost per audited outcome next to the price per claimed one. - [Outcome-based pricing break-even calculator](https://attify.app/tools/outcome-pricing-break-even.md): At what audited share does a per-outcome AI price beat what the same outcome costs you today? And when is per-conversation pricing cheaper than per-resolution pricing? ## Who it is for - [Attify for developers](https://attify.app/for/developers.md): For developers connecting an AI agent to Attify: the Ingest API, Python and TypeScript SDKs, OpenTelemetry, platform connectors, and the read-only API and MCP server agents use. - [Attify for finance](https://attify.app/for/finance.md): For finance teams paying for AI agents: cost per audited outcome against the price per claimed one, what was billed for claims that failed, and a monthly verdict you can check. - [Attify for investors](https://attify.app/for/investors.md): For investors in AI-agent companies: signed traction reports and audited outcome scores built from customers' own systems, with customers numbered and every figure's sample, period and provenance. - [Attify for operations leaders, clinics and service businesses](https://attify.app/for/operations.md): For practices, clinics and service businesses using an AI receptionist: which bookings were real, which needed staff to fix, which were no-shows, and what each kept appointment cost. - [Attify for procurement](https://attify.app/for/procurement.md): For procurement teams buying AI agents: what to require in writing, which assurance answers which question, and how to check a vendor's audited score yourself. - [Attify for sales leaders](https://attify.app/for/sales-leaders.md): For sales leaders paying for an AI SDR: how many of its booked meetings were real, held meetings with new prospects, what each cost, and how that compares with your own reps. - [Attify for support leaders](https://attify.app/for/support-leaders.md): For heads of support running an AI agent: the share of its resolutions that held for 7 days with no handoff, why the rest failed, and what each real resolution cost, every month. - [Attify for AI vendors and founders](https://attify.app/for/vendors.md): For AI-agent startups: what an audited score takes, what buyers and investors see, what happens if your audited rate is below your claim, and the path while you have fewer than three customers. ## Blog - [How the Audited by Attify badge works: a live status and a signed attestation](https://attify.app/blog/launch-badge.md): An Attify badge is drawn live from a vendor's current status, links to its public profile, and is backed by a signed attestation anyone can check at /verify. - [The Attify Index: pooled, anonymous, and empty until it isn't](https://attify.app/blog/launch-index.md): The Attify Index will publish audited AI agent outcomes by job type each month, pooled from buyers who contribute. It shows no number until enough companies take part. - [Three standards for what AI agents actually achieve](https://attify.app/blog/launch-standards.md): Attify publishes three standards, for customer-support resolution, meetings booked by AI sales agents and appointments booked by AI receptionists. What they count, and why. ## Job types Results compare only within one job type. Each job's checks, in its standard's order (GET ${SITE}/api/v1/jobs has their questions and evidence tiers): - support-resolution (Standard #1, customer support, the default): the customer connects its help desk; results count conversations. Checks: repeat_contact, handoff, judge_unresolved. Takes uploads (the free check). - sales-meetings (Standard #2, sales meetings): the customer connects its CRM and calendar; results count meetings. Checks: not_ai, not_new, stepped_in, cancelled, no_show, not_sales. - bookings (Standard #3, bookings): the customer connects its booking system; results count appointments. Checks: not_ai, not_new, staff_fixed, cancelled, no_show, wrong_booking. ## Vendor verification Vendor verification rests on connected data (for support resolution, every conversation their customers' help desks hold for a rolling 30 days, the AI-handled ones counted; for sales meetings and bookings, every meeting or appointment their customers' own systems hold for the period) or on witnessed data their customers reconciled (Standard #1 v0.11, "Vendor verification"; program rules: https://attify.app/standards/program-rules). Uploaded data never counts. A customer counts only when it is independent of the vendor: a separate company on its own domain, real history before its invitation, Attify's direct confirmation when drawn, and the vendor's attestation that its customers are genuine (a customer proven fake revokes the score). A vendor's first judged score is private: the vendor sees every number and chooses whether to publish it. Once it does, every later score publishes after a 7-day review window, whatever it shows, and a published score can never be hidden or edited. A score publishes only once the judge is validated against people for the standard version and the score's own human audit is complete. Statuses: Claim confirmed ยท within 5.0 points (Attify is 95% sure the vendor's count overstates audited outcomes by at most the claim tolerance, 5.0 points since 2026-11-09: the gap's upper 95% limit, pooled with each customer as a cluster, is within it; the tolerance is a published parameter, and every badge and attestation states the one it was decided with), Claim not confirmed (otherwise), Lapsed (the vendor stopped verification, no fresh data for 14 days, or below the minimums of 3 customers and 500 conversations; the published history stays public), Revoked (by Attify, with a reason). A status is always shown with the audited rate, the gap to the claim, the job type and the period. Rates are ranked and compared only within one job type, and for support resolution next to the contact-reason mix. - GET https://attify.app/api/v1/vendors?job_type=support-resolution&status=claim_confirmed: the registry (proven.vendor/1), each with job_type, standard version, evidence tier, provenance, rank within the job type and (support resolution) contact-reason mix. job_type is support-resolution, sales-meetings or bookings. - GET https://attify.app/api/v1/vendors/{slug}: a vendor's profile: the verdict sentence, both handoff outcomes (support resolution), history, the public changelog and the signed attestation. - GET https://attify.app/api/v1/attestations/{id}: a signed attestation (proven.attestation/1, Ed25519) and whether it is current. POST https://attify.app/api/v1/verify with {"attestation": ...}: the signature, then whether it is current, superseded, expired or revoked, in one sentence (proven.verification/1). - https://attify.app/.well-known/proven-keys.json: the public keys (JWKS). The signed bytes are the payload serialized with sorted keys, no whitespace, UTF-8, integral numbers without a fraction (what JSON.stringify writes once the keys are sorted). - https://attify.app/api/v1/openapi.json: the OpenAPI document for these. https://attify.app/badge/{slug}.svg: the live badge; https://attify.app/badge/{slug}.json: its hover card's data, read live (status, audited and claimed rates with intervals, period, standard version, last checked, profile and /verify links). - MCP (Streamable HTTP, POST https://attify.app/mcp), every tool read-only. No sign-in: get_verdict, list_findings, get_evidence, compare_subjects, explain_check, get_standard, ask_attify, get_index, define_term, lookup_vendor, verify_attestation, explain_provenance, get_traction_report. With sign-in, the person's own organization: get_my_account, get_monthly_verdict, get_agent_outcomes, compare_my_agents, get_renewal_packs, get_program_status, get_badge_embed, list_traction_reports, get_ingest_health (OAuth 2.1 with PKCE, discovered from https://attify.app/.well-known/oauth-protected-resource/mcp). What each returns: https://attify.app/docs/agents. Adding it to ChatGPT, Claude or Grok: https://attify.app/apps. ## Witnessed data: agents on their own platforms An agent that does not run inside a help desk Attify can read streams each interaction to Attify as it happens, signed with its deployment's key and timestamped on arrival: witnessed data (schema proven.ingest/1). It counts toward a verification only after the customer consents in Attify and reconciles it each month (its own total from its records or the vendor's invoices, then a random spot check against its own systems); until then it feeds the vendor's private report, and its verdict opens "On witnessed data not yet reconciled". An event later than 5 minutes is accepted and flagged; a late interaction is never scored as witnessed. Docs: https://attify.app/docs/ingest. - POST https://api.attify.app/v1/ingest/events with {"events": [...]} (1 to 500), signed: Proven-Key, Proven-Timestamp, Proven-Signature (v1=hex HMAC-SHA256 of "."). POST /v1/ingest/stream takes NDJSON. Answers proven.ingest.result/1 with Attify's signed receipt. These go to the API's own host with the deployment's key, never the access gate or the API key. - POST https://api.attify.app/v1/otlp/v1/traces: OpenTelemetry traces (OTLP/HTTP, protobuf or JSON), Authorization: Bearer , mapped from the gen-ai semantic conventions v1.41.1; the agent's claim needs the attribute proven.outcome. - https://attify.app/docs/ingest/schema.json: the event's JSON Schema (urn:proven:ingest-event:1). https://attify.app/docs/ingest/openapi.json: the OpenAPI document of the Ingest API, deployments and reconciliation. - Every verdict states its provenance (connected, witnessed or uploaded) and, for witnessed data, reconciled (true or false); every program customer states its stream's health (late events, gaps, sequence breaks) and its reconciliation. ## Traction reports A traction report states what a company's AI agent did at its customers, from connected data, or witnessed data its customers reconciled (never uploads; the company chooses whom to invite, and its customers are not put through the independence checks of a vendor verification): units of work per customer per month and week, retention month over month with expansion and contraction, pilot to production (the first week above a production threshold), concentration (top customer, top three), and audited outcome rates per customer-month at the rules tier. Judge-tier numbers appear only from a score published under the same gates as vendor verification. Every number carries its sample size, period, evidence tier and provenance; claims stated by an investor or the company are tested as supported, not supported or not testable; the opinion and summary are templates over the numbers. Shape: proven.traction/1 (schemas/traction.schema.json). - A share link (https://attify.app/r/{token}) opens the report as a page; its JSON is GET https://attify.app/api/proven/v1/shared/{token} (signed in when the link has an allowlist). Every opening is logged for the company. - The report's attestation (proven.traction-attestation/1, Ed25519, with the SHA-256 of the canonical report JSON) checks at POST https://attify.app/api/v1/verify: current, expired or withdrawn, in one sentence. - Investors request a report from a company by email (POST /v1/requests, an investor organization's session); the company accepts, its customers connect and consent, and the report is shared back to the investor's organization. ## The Attify Index Audited outcome rates by job type, pooled from the buyers who contribute anonymous data: connected or reconciled witnessed data only, never a company or vendor named. An edition publishes for a job type only when at least 5 companies contribute and none holds more than half of its outcomes. It regenerates each month; each edition keeps its permanent address (https://attify.app/index/{YYYY-MM}, with .json and .csv twins) and never changes. https://attify.app/index, or GET https://attify.app/api/v1/index (attify.index/1: per job type the audited share with its 95% interval, each company a cluster, the companies and outcomes behind it, the provenance mix, the failure modes' shares, the method and the citable findings). - Customer support: The Attify Index for customer support publishes when 5 companies contribute; 0 contribute now. - Sales meetings: The Attify Index for sales meetings publishes when 5 companies contribute; 0 contribute now. - Bookings: The Attify Index for bookings publishes when 5 companies contribute; 0 contribute now. ## Developer docs - [Developer docs](https://attify.app/docs): the four ways to send an agent's outcomes, and the read side. - [Ingest API](https://attify.app/docs/ingest.md), [for agents](https://attify.app/docs/agents), [schemas](https://attify.app/docs/schemas). - Sources, each with the fields read and the claim field: [Bland AI](https://attify.app/docs/sources/bland-ai.md), [Cal.com](https://attify.app/docs/sources/cal-com.md), [Calendly](https://attify.app/docs/sources/calendly.md), [ElevenLabs](https://attify.app/docs/sources/elevenlabs.md), [Freshdesk](https://attify.app/docs/sources/freshdesk.md), [Google Calendar](https://attify.app/docs/sources/google-calendar.md), [Gorgias](https://attify.app/docs/sources/gorgias.md), [HubSpot](https://attify.app/docs/sources/hubspot.md), [Intercom](https://attify.app/docs/sources/intercom.md), [Langfuse](https://attify.app/docs/sources/langfuse.md), [LangSmith](https://attify.app/docs/sources/langsmith.md), [Microsoft 365](https://attify.app/docs/sources/microsoft-365.md), [Retell AI](https://attify.app/docs/sources/retell-ai.md), [Salesforce](https://attify.app/docs/sources/salesforce.md), [Vapi](https://attify.app/docs/sources/vapi.md), [Voiceflow](https://attify.app/docs/sources/voiceflow.md), [Zendesk](https://attify.app/docs/sources/zendesk.md). ## Buyers A company that pays for an AI agent connects its own systems and gets a monthly dashboard per agent: claimed against audited, what each audited outcome costs against the price per claimed one, where claims failed, and a signed renewal pack (attify.renewal_pack/1) that checks at POST https://attify.app/api/v1/verify. The dashboard is https://attify.app/connect, signed in; the verdict on the 1st of each month arrives by email. ## Pricing [Plans and prices](https://attify.app/pricing.md). The price never depends on a result. Buyers pay by audited AI interactions a month, every one Attify audits whatever its result; vendors by the customers a program covers; traction reports by stage. A 14-day Team trial needs no card. Every vendor is free until vendor pricing starts; founding vendors (the first 10 to open a program) keep their terms after: free for 12 months from the first published score, then 50% off. Polar is the merchant of record. - Buyers, Free: Free. The monthly verdict; Audited against claimed; Benchmark against other companies; Needs your anonymous data, withdrawable. - Buyers, Team: $249 a month. Up to 5,000 audited AI interactions a month; Cost per audited outcome; Where it fails, with suggestions; Judged audits, alerts, the renewal pack. - Buyers, Business: $990 a month. Up to 50,000 audited AI interactions a month; Everything in Team; Several agents or vendors side by side. - Buyers, Enterprise: Custom. More volume, custom terms. - Vendors, Founding vendor: Free. Verification program, badge and traction reports; Free for 12 months from your first published score; Then 50% off, recorded on your account. - Vendors, Starter: Free today (listed at $500 a month). Up to 10 connected customers or reconciled deployments; Judged scores, the badge and a public profile. - Vendors, Growth: Free today (listed at $1,500 a month). Up to 50 connected customers or reconciled deployments; Everything in Starter. - Vendors, Enterprise: Custom. More customers, custom terms. - Vendors, Traction report, seed: Free today (listed at $2,500 a report). One signed traction report; Share links with expiry and a view log; Free when an investor requests it. - Vendors, Traction report, Series A and later: Free today (listed at $5,000 a report). One signed traction report; Share links with expiry and a view log; Free when an investor requests it. - Anyone, Upload check: Free. One export, no account; Claimed against audited, rules tier; A self-reported preview, never a verification. ## How results are read Every result states its provenance next to its evidence tier: "uploaded" (a check: self-reported, whoever uploaded the file chose what it holds), "connected" (Attify read every record of the period from the customer's own system) or "witnessed" (the vendor's agent streamed it, with "reconciled" saying whether the customer checked it). Only connected data, or reconciled witnessed data, can support a vendor verification or badge. Errors share one shape: {"error": {"code", "message", "status", "details"}}. - Read: https://attify.app/api/v1 (the OpenAPI document is https://attify.app/api/v1/openapi.json) and the MCP server above. Job types, their checks, evidence tiers and upload fields: GET https://attify.app/api/v1/jobs. - Act: people run the free check at https://attify.app/check/support-resolution, connect a help desk, CRM, calendar or agent platform at https://attify.app/connect, and stream an agent through the Ingest API above. The product's own API behind those pages takes Attify's key, never issued to agents. ## Optional - [Sample verdict](https://attify.app/demo/verdict.json), [findings](https://attify.app/demo/findings.json) and [verdict card](https://attify.app/demo/verdict_card.png): the home page's sample audit (proven.verdict/1, proven.findings/1), on synthetic demo data: invented conversations, judged by a scripted demo judge. - [Registry](https://attify.app/registry), [vendor profiles](https://attify.app/registry) and [Verify](https://attify.app/verify): the same, as pages. - [For AI vendors](https://attify.app/vendors): AI agent verification for vendors: an audited score from customers' own systems, a live badge, a signed attestation; the founding cohort's terms. - [For buyers](https://attify.app/buyers): AI agent verification for companies using AI agents: audited against claimed outcomes, cost per audited outcome, the free check.