AGENT NATIVE OFFERS

← Leaderboard

Dify Human-Dependent

AUDIE Score: 58/100 · Audited 2026-04-06 · Website: https://dify.ai · Machine-readable: JSON

Pillar Scores

P1 Signal Architecture — 18/25
P2 Clarity Stack — 16/25
P3 Trust Envelope — 9/20
P4 Velocity Triggers — 6/10
P5 Gravity Design — 9/20

Executive Summary

Dify scores 58/100 — **HUMAN-DEPENDENT** tier. The platform excels at MCP integration and transparent pricing, making it accessible to AI agents in theory. However, it lacks critical infrastructure for autonomous agent operation: no scoped permissions, no machine-readable audit logs, and no programmatic renewal signals. Agents evaluating Dify must currently escalate to human operators for permission management, action verification, and subscription renewal. The top priority is implementing agent-scoped permissions and machine-accessible audit logs; both are medium-effort, high-impact fixes that would elevate Dify into the EMERGING tier (65–84).

Strongest Signals

Critical Gaps

Priority Actions

    All 20 Criteria

    P1-A Structured Data — 2/5
    Dify mentions JSON Schema for structured LLM outputs in their docs and feature releases (X: "Structured Outputs with JSON Schema Editor"), but no organization schema.org markup detected on homepage. No Offer/Product schema found.
    P1-B Machine-Readable Pricing — 4/5
    Pricing clearly displayed on https://dify.ai/pricing with distinct tiers (Sandbox, Professional, Team). Message credits, API limits, storage quoted in tables. Not in schema.org/Offer format, but well-structured in HTML.
    P1-C llms.txt / Agent Layer — 3/5
    Docs reference llms.txt endpoint at https://docs.dify.ai/llms.txt, but this is library documentation, not a published agent identity layer. No /llms.txt file found at root domain.
    P1-D API / MCP Availability — 5/5
    Native MCP integration documented. Dify publishes agent capabilities including tool integration, workflow nodes, and plugin ecosystem. MCP support confirmed in docs: "Bridge Your Systems / Platforms with Native MCP Integration" and "Publish as an Universal MCP Server."
    P1-E Discoverability (GEO) — 4/5
    5M+ GitHub downloads, 800+ contributors, high-quality structured documentation. Strong SEO and AI retrieval signals. Optimized for LLM consumption via knowledge base integration and structured outputs.
    P2-A Offer Completeness — 4/5
    What (agentic workflows, RAG, LLM applications), Who (developers, enterprises), How Much (Sandbox free, Pro $59/mo, Team $159/mo) all stated. However, scattered across multiple pages—not unified single source.
    P2-B Scope & Limits — 4/5
    Message credits per tier explicit: Sandbox 200, Pro 5,000, Team 10,000. API call limits (Sandbox 5,000/month unlimited), storage (50MB to 20GB), request rate limits (10 to 1,000/min) all stated. Clear and structured on pricing page.
    P2-C Substitution Rules — 1/5
    No documented fallback behavior if monthly message credits exhausted. No equivalent-product logic or tier escalation rules published.
    P2-D Conditional Logic — 3/5
    Pricing is straightforward (no hidden "contact sales"), but no agent-optimized conditional logic gates documented. No explicit eligibility rules or gating conditions.
    P2-E Semantic Precision — 4/5
    "Message credits," "knowledge requests," "workspace" all defined in context. Some marketing language remains ("Enterprise-grade security," "Scalable, Stable, Secure") without measurable definitions.
    P3-A Verifiable Performance — 4/5
    Dify statuspage (https://dify.statuspage.io/) shows 100% uptime over past 90 days. G2 rating: 4.1/5 from 20 verified reviews (55% 5-star). SOC 2 Type I/II, ISO 27001:2022, GDPR certified. No published SLA percentage found.
    P3-B Scoped Permissions — 2/5
    Team member counts and app limits mentioned (Sandbox 1 member, Pro 3, Team 50), but no agent-specific permission scoping (time-bounded, action-bounded, amount-bounded) documented.
    P3-C Audit Trail — 1/5
    "Unlimited Log History" mentioned for Pro/Team tiers, but no machine-accessible audit log endpoint or transaction history API documented for agents.
    P3-D Behavioral Consistency — 2/5
    Compliance certifications (SOC 2, ISO 27001, GDPR) suggest stability. No explicit change policy, version control of terms, or notice period documented.
    P4-A Friction-Free Activation — 4/5
    Sandbox tier requires no credit card ("No credit card required for the free tier's 200 OpenAI calls trial"). Instant access. Upgrade to Pro requires payment but is automated. Full autonomous activation not gated by human approval.
    P4-B Agent Decision Signals — 2/5
    Free tier exists (200 messages), but signal not explicitly optimized for agent decision-making. No published urgency triggers (deprecation, capacity limits, pricing tier thresholds in agent-legible format).
    P5-A Integration Depth — 3/5
    Knowledge base storage (50MB–20GB) creates data lock-in. Integrations with external APIs and LLM providers via plugin ecosystem. Mild switching cost due to workflow data trapped in Dify system.
    P5-B Agent Memory Layer — 3/5
    Workspace and team member history stored. Not agent-specific memory. No agent-accessible context persistence layer documented.
    P5-C Programmatic Renewal — 1/5
    No programmatic renewal API documented. No agent-accessible re-up signals or usage-based triggers.
    P5-D Compounding Value — 2/5
    Knowledge base grows over time, improving RAG quality. Not explicitly signaled as agent-readable compounding value metric.

    Rubric v1 (April 2026). Scores reflect the company's state on the audit date and may have improved since.