AI gateways
Bonito vs Kong AI Gateway
API gateway for AI traffic vs. AI control plane with agents
Kong AI Gateway 2.0 went GA on September 1, 2026
Kong AI Gateway is the AI traffic layer (LLM, MCP and A2A) from the Kong API management company, now a dedicated runtime in version 2.0. It suits teams that already run Kong for their APIs. Bonito adds agents, managed knowledge bases, cost routing and residency pinning on top of the gateway, as a managed service.
Kong AI Gateway pricing
Konnect Plus: $100 per month per LLM model (up to 5), plus per-gateway and per-request charges. Enterprise: custom, annual.
Model costs
No markup. You use your own provider accounts.
What Kong AI Gateway does well
- A deep plugin set: multi-provider load balancing and failover, semantic cache, prompt guards, PII sanitizer and prompt compression
- Dollar budgets per consumer tier that return 429 when exhausted
- MCP proxy that turns existing APIs into MCP tools, with OAuth2 and tool-level ACLs
- Image generation and a Sora video route through the same gateway
- Compliance: ISO 27001, SOC 2 Type II and PCI DSS
Where Bonito goes further
- Builds and runs agents: visual canvas, chat builder, approval queue, schedules and persistent memory
- Managed knowledge bases with upload, embedding and search; Kong's RAG Injector reads a vector database you run
- Cost routing to cheaper models for simple work
- Automatic Bedrock cross-region profiles and equivalent-model failover without writing plugin config
- Data residency pinning to CA, US or EU in one setting
- Agent trace tree with dollar cost per request, agent, team and provider
Where Kong AI Gateway is ahead of Bonito
- Semantic caching; Bonito has none
- Open-source gateway core and self-hosting
- One gateway for AI traffic and the rest of your APIs
- PCI DSS and a long API management track record
Capability by capability
| Capability | Bonito | Kong AI Gateway |
|---|---|---|
| Agents | ||
| Build and run agents | Yes | Nogoverns A2A traffic; does not build or host agents |
| Approval queue and scheduled runs | Yes | Nonot found in their docs (Sep 2026) |
| Dollar budgets per agent | Yes | Partialdollar budgets per consumer tier, not per agent |
| MCP support | Yes | Yes |
| Routing | ||
| OpenAI-compatible gateway | Yes | Yes |
| Automatic failover across providers | Yes | Yes |
| Cheaper models for simple work (cost routing) | Yes | Nonot found in their docs (Sep 2026); cost budgets exist |
| Bedrock, Vertex, Azure and direct providers in one place | Yes | Yes |
| Semantic caching | No | Yes |
| Governance | ||
| SSO, RBAC and audit log | Yes | YesSSO and audit logging on Enterprise |
| Data residency pinning | Yes | Partialyou choose where the gateway runs; no residency pinning found |
| Observability and knowledge | ||
| Agent traces with cost per request and agent | Yes | Partialtoken and cost analytics, OpenTelemetry; no agent trace tree |
| Knowledge bases (RAG) | Yes | PartialRAG Injector over your own vector database |
| How you run it | ||
| Whole stack from one config file | Yes | YesdecK and kongctl |
| Self-hosted or open source | Partial | PartialKong Gateway core is open source; AI plugins and 2.0 are Enterprise |
| No markup on inference | Yes | Yes |
Checked against Kong AI Gateway's own documentation on 2026-09-27. Products change quickly; if something here is out of date, tell us at hello@trybonito.com and we will fix it.
Sources
See Bonito on your own stack
Start free, or get a report on where agents fit in your company first.