Smartflow is an AI governance control plane built for AI traffic. Kong AI Gateway is an AI governance layer built on Kong's API gateway, extending an established API management platform to LLM, MCP and agent-to-agent traffic. Both enforce policy inline. They differ in what the surrounding platform is for, and in what the audit record is designed to prove.
The short version
Kong's position in 2026 is stronger than most comparisons acknowledge. AI Gateway 3.14 shipped native A2A support in April, AI Gateway 2.0 reached general availability in September, and the AI plugin family now covers semantic caching, semantic prompt and response guards, PII sanitization, token-based rate limiting and MCP proxying. If the organization already runs Kong, the AI story is a license away rather than a new platform.
Smartflow is the choice where AI governance is the primary requirement rather than one traffic class among many, and where the evidence a regulator asks for has to come out of the system rather than be assembled from it.
Feature Comparison
| Capability | Smartflow | Kong AI Gateway |
|---|---|---|
| Architecture | Purpose-built for AI governance. Rust. | API management platform extended for AI. Apache-2.0 core with an AI plugin family. |
| AI traffic understanding | Token-level visibility, prompt understanding, model-aware routing. | AI-aware plugins with token-level metering, global token budgets with per-model caps, and dynamic model routing on a model alias. |
| Semantic caching | MetaCache resolves a lookup in four phases, first hit wins: intent signature, intent near-miss, exact key, then a VectorLite BERT semantic match. Hit rates are workload-dependent. | AI Semantic Cache plugin, shipped in Gateway 3.8 on 11 September 2024. Vector-database backed. Enterprise license required. |
| Guardrails | Visual policy editor covering PII, topic restriction, jailbreak detection and output moderation, with no code required. | AI Prompt Guard is open source. Semantic prompt guard, semantic response guard, PII sanitizer, prompt compressor, LLM-as-judge and custom guardrail are enterprise-licensed. |
| Agent identity | AIDA cryptographic agent credentials, scoped and independently revocable. | API key, OAuth 2.1 and RFC 8693 token exchange. Principal-aware policies via Kong Identity in AI Gateway 2.0. |
| MCP and A2A | MCP JSON-RPC gateway with tool caching. A2A orchestration under agent identity. | AI MCP Proxy with OAuth 2.1 and scope-based tool filtering. AI A2A Proxy, generally available April 2026. Both enterprise-licensed. |
| Audit log retention | Customer-controlled. No vendor-imposed ceiling. | Konnect retains audit logs for seven days, after which they are permanently deleted. Self-managed Gateway retention is configurable via audit_log_record_ttl, with optional RSA signing of entries. |
| Regulatory evidence | Examination packages and model inventory mapped to FINRA 3110, HIPAA, SOX and the EU AI Act. | General API security and audit logging. AI-specific regulatory evidence generation is not a documented feature. |
| API management | Not an API management platform. | Established API lifecycle management with a large plugin ecosystem. |
Purpose-built versus extended
"Extended from API management" is often used as a criticism. It should not be. Kong's AI plugins inherit two decades of work on rate limiting, authentication, observability and operational hardening, and an AI gateway that inherits a mature data plane starts from a better place than one that reimplements it.
The question is what the platform optimizes for when the two goals pull apart. An API gateway's model of the world is the request: a method, a path, headers, a body, a consumer. Kong's AI plugins extend that model intelligently — token metering reads the provider's usage response, model-aware routing reads the model field from the body — but the unit of governance is still the request.
Smartflow's unit of governance is the interaction: which principal, holding which authority, asked what of which model, under which policy version, and what came back. That framing is what makes the evidence question tractable, and it is a different shape from an API request even when the bytes on the wire are identical.
The licensing line
Kong publishes an unusually clear free-versus-enterprise split, which is worth understanding before a proof of concept. Six AI plugins are free and open source: AI Proxy, AI Prompt Guard, AI Prompt Template, AI Prompt Decorator, AI Request Transformer and AI Response Transformer. Everything semantic, agentic, MCP-related or vendor-integrated carries an "AI License Required" label — semantic cache, semantic prompt and response guard, AI rate limiting advanced, PII sanitizer, RAG injector, MCP proxy, MCP OAuth2 and A2A proxy among them.
A Kong evaluation that starts with the open-source plugins and a Kong deployment that ends with governance are therefore two different commercial conversations. That is not a hidden cost — Kong labels it plainly in the Plugin Hub — but it is worth pricing at the start rather than at renewal.
Retention and the examination window
Kong Konnect retains audit logs for seven days and then permanently deletes them, with no documented extension. Self-managed Kong Gateway makes retention configurable and can RSA-sign each entry, which is a genuinely strong integrity feature and one Kong does not get enough credit for.
The gap matters because regulatory retention windows are measured in years: six under HIPAA, seven under SOX, and longer in several supervision contexts. Kong's own documentation offers the right answer — pipe audit entries into a SIEM or logging warehouse, which the Admin API supports. That works. It also means the evidence a regulator asks for lives in a third system and has to be correlated back. Smartflow's examination suite is the attempt to remove that correlation step.
Where Kong excels
Kong is the proven choice for API management at scale, and in 2026 it has been a fast mover in AI. AI Gateway 2.0, generally available on 1 September 2026, decoupled the AI release cycle from the API gateway release train so the AI product can ship on its own cadence — a structural decision that says something about intent. Gateway 3.16 added runtime log-level control, CEL-driven dynamic plugin configuration and a FIPS 140-3 package.
If the requirement is one governance layer across APIs, events, LLM calls, MCP tool access and A2A traffic, Kong is making that case credibly and no purpose-built AI gateway covers the first two.
When to choose Smartflow
- AI governance is the primary requirement, not one traffic class inside a broader API estate
- Evidence generation: examination packages and model inventory produced from runtime records rather than reconstructed from a SIEM
- Agent identity: a cryptographic credential held by the agent, not a key borrowed from a consumer record
- Regulated-industry controls: information barriers and supervision evidence, which have no direct equivalent in the Kong plugin family
Running both
These two are more often complementary than exclusive. A common shape is Kong at the edge for API lifecycle management, authentication and general traffic policy, with Smartflow governing the AI path behind it. The cost is an extra hop; the benefit is that neither platform is asked to do the thing it was not designed for. Where an organization has already standardized on Kong, this is usually a better first move than a replacement.
Common questions
Does Kong support semantic caching? Yes. The AI Semantic Cache plugin shipped in Kong Gateway 3.8 on 11 September 2024 and is actively maintained. It requires an AI Gateway enterprise license.
Does Kong support agent-to-agent traffic? Yes. The AI A2A Proxy plugin reached general availability with AI Gateway 3.14 in April 2026.
How long does Kong retain audit logs? Konnect retains them for seven days and then permanently deletes them. Self-managed Gateway retention is configurable.
Which Kong AI plugins are free? AI Proxy, AI Prompt Guard, AI Prompt Template, AI Prompt Decorator, AI Request Transformer and AI Response Transformer. The semantic, agentic and MCP plugins require an AI license.
Related reading
Enterprise AI gateway guide · Model Context Protocol · Semantic caching · Agent governance · All comparisons · MCP security · Model allowlist and denylist · Agentic AI · AI guardrails
Verified against each vendor’s published documentation in September 2026. Vendor capabilities change. If something here is out of date, tell us and we will correct it.
Kong sources: konghq.com/products/kong-ai-gateway, konghq.com/pricing, developer.konghq.com (ai-gateway, plugins, plugins/ai-semantic-cache and its changelog, plugins/ai-rate-limiting-advanced, plugins/ai-mcp-proxy, gateway/audit-logs, konnect-platform/audit-logs, gateway/deployment-topologies), konghq.com/blog product releases for AI Gateway 3.8, 3.14, 2.0 GA, Agent Gateway and Gateway 3.16, github.com/Kong/kong LICENSE.
Put this in the path of your own agents.
Policy enforced inline between your agents and every model and tool they reach, with a record bound to the human who owns it.
Request a Demo Read the docs