All 20 Criteria
P1-A Structured Data — 2/5
Every Actor page on the Apify Store automatically generates a machine-readable .md version (e.g., apify.com/apify/website-content-crawler.md), which is a notable innovation in content addressability. However, no schema.org markup (Product, Offer, AggregateRating) found on the main pricing or homepage. The .md-page pattern is ahead of most competitors but doesn't meet the schema.org standard required for higher scores.
P1-B Machine-Readable Pricing — 4/5
apify.com/pricing.md returns a full structured markdown pricing document with tables for all plans, compute unit (CU) rates, proxy pricing by type, and add-ons. This is a dedicated machine-readable pricing endpoint — rare and genuinely useful for agents needing to evaluate Apify without parsing JavaScript-rendered HTML. One point short of 5/5 because it uses markdown tables rather than schema.org/Offer or JSON.
P1-C llms.txt / Agent Layer — 5/5
docs.apify.com/llms.txt provides the docs index for LLM consumption, and docs.apify.com/llms-full.txt provides the complete documentation dump. Additionally, every Actor page generates its own .md version automatically. The apify.com llms.txt links directly to these. This is the most complete tiered llms.txt implementation in this audit series — root-level pointer to both summary and full content, plus per-Actor machine-readable pages.
P1-D API / MCP Availability — 5/5
Comprehensive API layer: OpenAPI v2 spec at docs.apify.com/api/v2.md, Python and JavaScript API clients (open-source), REST API, and MCP server at mcp.apify.com with Docker image published on Docker Hub for local deployment. The MCP server supports dynamic Actor discovery — agents can search, retrieve details on, and add new Actors at runtime. Additionally supports x402 (USDC on Base blockchain) and Skyfire for autonomous agentic payments, enabling agents to pay for Actor runs without human authorization. This is the most agent-payment-native infrastructure found in any audit to date.
P1-E Discoverability (GEO) — 3/5
robots.txt is minimal (allow all, sitemap reference only — no AI-bot-specific directives). llms.txt exists at root via apify.com llms.txt. Strong content presence in developer communities and SOC 2 certifications add trust signals. Not proactively AI-retrieval optimized beyond llms.txt presence. Would benefit from explicit AI crawler allowances in robots.txt and structured entity content.
P2-A Offer Completeness — 4/5
apify.com/pricing.md is a single authoritative document covering all plan tiers, CU pricing, proxy pricing, add-ons, and billing FAQs. Individual Actor pricing is per-Actor (some free, some paid/rented) — this variability is disclosed but means per-Actor total cost requires querying Apify Store. Offer completeness is high for the platform itself; Actor-level pricing adds an inherent variable.
P2-B Scope & Limits — 3/5
Concurrent run limits (25–256 by plan), max RAM per plan (8–256 GB), and CU pricing are documented in pricing.md. However, API rate limits for the REST API are not prominently documented for agent-facing usage. Compute unit consumption per Actor type is variable and requires test runs to determine — disclosed in FAQs but not pre-declarable. Adequate but not fully declarative.
P2-C Substitution & Fallback Rules — 1/5
No substitution or fallback guidance found. Agents building workflows on Apify Actor runs have no documented behavior for Actor failure, result timeout, or service degradation. This is a common gap across audited companies but notable given Apify's enterprise positioning.
P2-D Conditional Logic Transparency — 3/5
Overage behavior is clearly explained: free plan users are blocked; paid plan users are charged overage and notified. Upgrade/downgrade process is documented (prorated, same billing cycle). Enterprise requires "contact sales" but is disclosed. Actor rent pricing varies per-Actor — disclosed but variable. Conditions are findable; not all machine-readable.
P2-E Semantic Precision — 3/5
Platform-level pricing is precise (1 CU = 1 GB RAM × 1 hour, specific dollar amounts per plan). Actor-level descriptions vary significantly in precision — some Actors have precise rate/result counts, others use marketing language. The platform core is precise; the marketplace layer introduces variability.
P3-A Verifiable Performance Data — 4/5
status.apify.com is a dedicated Atlassian Statuspage with real-time and historical incident data. StatusGator has independently monitored Apify since April 2022. Self-reported 99.95% uptime. SOC 2 Type II, GDPR, and CCPA certified — third-party validated compliance. G2, Capterra, GetApp, and SoftwareAdvice reviews exist. Trusted by Intercom and European Commission. Strong third-party verification footprint.
P3-B Scoped Permission Model — 4/5
API token with organization-level scoping. x402 protocol integration enables agents to make bounded, on-chain USDC payments for Actor runs — this is the most explicit agent-scoped permission model found in this audit series. Skyfire integration provides managed payment tokens with spending controls. An agent can be authorized to spend up to X USDC on Actor runs without needing a human-held API key. Notable innovation.
P3-C Audit Trail / Transaction Log — 3/5
Usage data is accessible in the Apify Console billing section and via the API (Actor run history, dataset contents). Run-level logging is a core platform feature. However, a dedicated machine-accessible audit log API (timestamped actions, payment records, key usage) is not explicitly documented for agent system consumption. Better than most audited companies; not quite full audit-trail grade.
P3-D Behavioral Consistency Signals — 3/5
changelog at apify.com/change-log provides timestamped platform updates. REST API has been at v2 for multiple years — stable versioning track record. Platform has been operating since ~2015 with enterprise customers (Intercom, European Commission). No explicit API deprecation policy or notice period stated, but demonstrated multi-year stability is evidence.
P4-A Friction-Free Activation — 4/5
Free plan with no credit card required and instant account creation. API token issued immediately in the console. x402 agentic payment support means an agent can autonomously pay for Actor runs without requiring a pre-provisioned human API key. Minor friction: initial account registration requires email/OAuth signup — a human action needed at least once. Post-setup, fully autonomous operation is supported.
P4-B Agent Decision Signals — 4/5
Free plan provides a clear autonomous trial signal. Dynamic Actor discovery via MCP lets agents find and evaluate Actors programmatically before committing to use. x402 USDC payment mechanism allows agents to make cost-bounded autonomous run decisions. Minor gap: no explicit "when to upgrade" signal or webhook for usage threshold notifications that agents can subscribe to.
P5-A Integration Depth / Switching Cost — 4/5
10,000+ Actors in the Apify Store create a data network effect: each Actor's run history, dataset, and schedule is tied to the Apify platform. Crawlee (open-source scraping library) ties developer workflows to the ecosystem. Integrations with Vercel AI SDK, OpenAI Agents SDK, Mastra, LangChain, and others embed Apify at the framework level. Switching requires migrating Actor configurations, stored datasets, scheduled runs, and proxy configurations.
P5-B Agent Memory / Personalization Layer — 3/5
Apify Storage (Datasets, Key-Value Stores, Request Queues) provides persistent, API-accessible storage that agents can use as memory between runs. A key-value store is explicitly designed for state persistence. Agents can read, write, and query their own stored data via the REST API between Actor runs. This is a genuine agent-accessible memory layer — one of the few in this audit series.
P5-C Programmatic Renewal Signals — 3/5
All paid plans auto-renew without human action. Billing management is available via the console API. Usage thresholds trigger notifications (documented in FAQs). No dedicated agent-facing renewal API or webhook for "subscription expiring soon" events found, but the auto-renewal mechanism is solid and the billing API is accessible.
P5-D Compounding Value Signal — 3/5
Actor ecosystem grows over time (10,000+ Actors as of audit date). Scheduled runs accumulate historical datasets that become more valuable over time. Platform monitoring tools provide historical performance data per Actor. These compounding effects are real but not surfaced as agent-readable signals — agents cannot query "how much more valuable is this platform than last month" via API.
Rubric v1 (April 2026). Scores reflect the company's state on the audit date and may have improved since.