Blog
Notes from the gateway
Product updates, routing math, and the occasional deep dive into what an LLM gateway actually does under the hood. RSS.
- 6 min read
Announcing opa-sidecar-a2a: A Reference Implementation for Chain-Aware Agent Authorization
Agent-to-agent traffic is going into production faster than the authorization models are catching up. We built and open-sourced a reference implementation that shows what chain-aware authorization looks like, using Open Policy Agent as the engine. MIT licensed, Go, single binary.
A2AAgent AuthorizationOPAOpen Policy AgentChain-aware AuthorizationOpen SourceMCP - 10 min read
Portkey Alternatives (2026): 6 Gateways Compared After the Palo Alto Acquisition
Palo Alto Networks acquired Portkey in May 2026, and log-metered pricing is squeezing high-traffic teams. Honest comparison of 6 alternatives (Leanroute, LiteLLM, Bifrost, Kong AI Gateway, Helicone, OpenRouter) with a decision tree.
PortkeyPortkey alternativeAI GatewayLLM GatewayPalo Alto NetworksBYOKComparison - 9 min read
Kong AI Gateway Alternative (2026): When a Managed BYOK Gateway Wins
Kong AI Gateway is the enterprise procurement-safe pick, with $30-50K+ annual contracts and a real setup surface. Honest comparison of alternatives (Leanroute, LiteLLM, Bifrost, Portkey) and when Kong is genuinely the right call.
Kong AI GatewayKong alternativeAI GatewayLLM GatewayEnterpriseBYOKComparison - 10 min read
Bifrost Alternative: When a Managed BYOK Gateway Beats Self-Hosting Go
Bifrost is a legitimately fast open-source LLM gateway from Maxim AI, a Go rewrite of the LiteLLM pattern benchmarked at 40-50x LiteLLM's throughput. If you already have a platform team, it's a defensible pick. Here's when a managed BYOK gateway like Leanroute is the better call, and when you should stay on Bifrost.
BifrostBifrost alternativeAI GatewayLLM GatewayMaxim AILiteLLMBYOKComparison - 8 min read
AWS Bedrock via Leanroute: BYOK, Unified Logs, and Multi-Provider Routing
AWS Bedrock is a great managed model service. It is not a gateway. Here is how teams use Bedrock through Leanroute to get unified BYOK across Bedrock plus 13 non-AWS providers, gateway-edge spend caps, and OpenAI-compatible wire format on top of Bedrock's Anthropic and Meta models.
AWS BedrockBedrockAI GatewayLLM GatewayBYOKIAMMulti-cloud - 9 min read
FinOps for LLM Spend: Three Reasons Your Bill Changed
When your LLM bill moves, three separate things could have caused it. Your usage changed, your pricing changed, or the way your app talks to the model changed. Mixing them up is how teams end up thanking the wrong lever. A simple framework for pulling them apart.
FinOpsLLM CostAI EconomicsCost AttributionAI GatewayLLM GatewayBYOK - 8 min read
OpenRouter Alternative: When BYOK Beats Per-Request Markup (2026)
OpenRouter's per-token markup starts hurting around $200-500/mo. Break-even math, three real switch scenarios, and when OpenRouter is still the right pick — honest BYOK alternative comparison.
OpenRouterOpenRouter alternativeAI GatewayLLM GatewayBYOKCost OptimizationLLM Routing - 8 min read
LiteLLM Alternatives 2026: 5 Gateways Compared (Post-CVE-2026-35029)
After CVE-2026-35029, teams are re-evaluating LiteLLM. Honest comparison of 5 alternatives — Leanroute, Portkey, Kong AI Gateway, Helicone, OpenRouter — with a decision tree and when to stay on LiteLLM.
LiteLLMLiteLLM alternativeAI GatewayLLM GatewayPortkeyKong AI GatewayHeliconeOpenRouterComparison - 8 min read
Self-Hosting an LLM Gateway: When It's Right, When It's Not
LLM middleware holds every provider key in your stack, sees every prompt, and often has admin access to spend caps. Self-hosting it means owning the full operational surface. A recent LiteLLM CVE is a reminder to make that choice deliberately, not by default.
AI GatewayLiteLLMSelf-HostingManaged ServicesLLM InfrastructureSecurityDevOps - 10 min read
Chain-aware Authorization: OPA + MCP for the A2A Auth Problem
AI agents are nothing but APIs on cron with autonomy and reasoning. Everything we learned about API authorization still applies. Here is the pattern I think should become standard, and the reason it works.
AI AgentsAuthorizationMCPOPAA2AAI InfrastructureLLM - 12 min read
Cut Claude Code Bills 60% by Routing to DeepSeek and Groq
Claude Code defaults to Sonnet. Sonnet is great. Sonnet is also expensive. Here is a practical walkthrough for routing your Claude Code sessions across 14 providers, keeping the same tools and prompts, and cutting your bill by more than half.
Claude CodeLLM RoutingAI GatewayCost OptimizationDeepSeekGroqMCP - 5 min read
Migrating from OpenRouter to Leanroute
Already using OpenRouter? Learn how to migrate to Leanroute while keeping your OpenAI-compatible integration and adding native MCP forwarding.
OpenRouterAI GatewayLLM GatewayMCPAI InfrastructureLLM Routing - 3 min read
Deepseek Price Hike, Routing Arbitrage Saves Cost
DeepSeek Raised Prices and Added Peak Hours. Here’s How Routing Arbitrage Keeps Your LLM Bill Under Control
AI GatewayLLM GatewayMCPAI InfrastructureLLM Routing - 3 min read
Stop Building Your Own LLM Router
Model routing looks simple until you have to maintain it. Here's why most AI teams should use routing infrastructure instead of building their own.
AI GatewayLLM RoutingAI InfrastructureLLMDeveloper Tools - 5 min read
Why Every AI Gateway Will Need to Speak MCP
AI Gateways solved model fragmentation. The next challenge is tool fragmentation. Here's why the future of AI infrastructure is one gateway for models and tools.
AI GatewayMCPModel Context ProtocolAI InfrastructureProduction AI - 12 min read
What Is an AI Gateway? The Complete Guide for Production AI Applications (2026)
Learn what an AI Gateway is, why production AI applications need one, and how model routing, failover, and MCP are changing AI infrastructure.
AI GatewayLLM GatewayAI InfrastructureModel RoutingMCP - 3 min read
One endpoint for chat + MCP: why we built MCP passthrough
Every LLM gateway routes chat completions. Leanroute is the only one that also carries MCP tool servers end-to-end — no separate MCP client, no SDK swap, no protocol boundary in your stack.
MCPModel Context ProtocolLLM GatewayAnthropicproduct