AirLLM vs Portkey

At a glance

Portkey describes itself as a production stack for generative AI. It combines an AI gateway, observability, guardrails, governance, prompt management and an MCP gateway. AirLLM is narrower: Air Router is a routing layer for AI coding agents that chooses the model for each step of the work.

One recent change matters if you're evaluating Portkey. Palo Alto Networks completed its acquisition of Portkey on May 29, 2026, and portkey.ai now says "Portkey is now PRISMA AIRS AI Gateway." Product names and plans may change, so check Portkey's site for the current offering.

Routing approach

Portkey's routing is rule-based and explicit. You write a config that sets a strategy: fallback, loadbalance or conditional. You apply it with a header, a saved config ID or a default on an API key. Fallbacks trigger on any non-2xx response by default, or on the status codes you list. Conditional routing matches on request metadata, parameters such as the model name, or the URL path. Retries (up to five), request timeouts, circuit breakers and canary splits are documented on all plans.

AirLLM makes the model choice for you. You select a routing model that expresses a preference: airllm-auto across supported providers, airllm-anthropic for Anthropic's Claude family, airllm-openai for OpenAI's models, or airllm-agent for your own agents. Air Router then picks the underlying model for each request. It is built around coding-agent sessions: it routes agent steps, weighs existing prompt caches before switching models, escalates to stronger models when needed and falls back across providers.

In short, Portkey gives you a policy engine to express the routing you want. AirLLM gives you routing decisions tuned to coding-agent work, with less to configure and less to control.

Coding-agent integration

Both document the main coding agents:

• Claude Code: Portkey sets ANTHROPIC_BASE_URL=https://api.portkey.ai with Portkey headers in ANTHROPIC_CUSTOM_HEADERS. AirLLM uses ANTHROPIC_BASE_URL="https://api.airllm.ai/" with your AirLLM key as ANTHROPIC_AUTH_TOKEN.
• Codex CLI: both use a provider in ~/.codex/config.toml. AirLLM's base URL is https://api.airllm.ai/v1 with wire_api = "responses".
• Cursor and Cline: both use the tool's OpenAI-compatible base URL setting.

Anthropic's documentation says it doesn't endorse third-party gateways and doesn't support routing Claude Code to non-Claude models through any gateway. On AirLLM, airllm-anthropic keeps Claude Code on Claude models; see the Claude Code router guide. AirLLM also ships its own coding agent, AirCode CLI, which uses Air Router out of the box.

Cost controls and caching

Portkey tracks cost on every plan through a central pricing table. Its docs reserve budget limits (in dollars or tokens, with optional weekly or monthly resets) and request or token rate limits for Enterprise and select Pro customers. Simple caching is available on all plans. Portkey's docs and pricing page describe semantic-cache availability differently, so check the current plan details.

AirLLM's controls come with every plan. Monthly plans show a weekly usage budget for each key, the Token Plan draws down prepaid credits, and Air Router is designed to interrupt unproductive agent loops before they consume more tokens. These controls are best-effort, not a guarantee that every charge is stopped. Air Router also reuses eligible cached responses and trims repeated context, such as duplicated files and completed tool history, before a request reaches the model.

Deployment and data

Portkey offers the widest range of deployment choices of the two. You can use the hosted service, run the open-source gateway yourself, or use Enterprise hybrid deployments that keep request and response logs in your own VPC. Its observability is built on logging requests. Enterprise customers can switch to metrics-only logging, and the x-portkey-debug: false header skips storing request and response content.

AirLLM is hosted only, and model availability depends on its upstream providers. Requests pass through short-lived, account-scoped caches, which generally expire within 24 hours of last use. When a request fails, AirLLM may keep its content, with secrets removed, to diagnose the failure, and it keeps samples of responses to check routing quality. AirLLM never uses any prompt to train models. The model provider that answers a request processes it under its own terms: on paid plans AirLLM avoids providers that train on prompts, while on the Free plan providers may train on and retain them.

When to choose which

Choose Portkey if

• You need an enterprise control point for all AI traffic, with observability, guardrails, prompt management and governance in one place.
• You want to write explicit routing rules (fallbacks, weighted load balancing, conditions on metadata) across many providers.
• You need self-hosted or hybrid deployment, or you're standardising on Palo Alto Networks' Prisma AIRS.

Choose AirLLM if

• Your main workload is coding agents or multi-step agents, and you want the model chosen per step without writing routing rules.
• You want routing that accounts for prompt caches, repeated context and agent loops.
• You prefer a hosted service with nothing to deploy.

AirLLM doesn't offer a guardrails product, prompt management or self-hosting. If those are requirements, Portkey covers more ground.

Using both

Some teams use more than one gateway: a central gateway such as Portkey for company applications, and Air Router for developers' coding agents. Each coding tool points at one base URL at a time, so decide per tool which service it uses. Usage, budgets and logs are then tracked separately in each service.

Note. Product details change often. This comparison was compiled from each vendor's public documentation on October 20, 2026. Portkey, Prisma AIRS and other product names are trademarks of their respective owners.

Frequently asked questions

Related

Sources

• Palo Alto Networks — Completes acquisition of Portkey
• Portkey website
• Portkey pricing
• Portkey AI Gateway on GitHub
• Portkey docs — Universal API
• Portkey docs — Configs
• Portkey docs — Fallbacks
• Portkey docs — Conditional routing
• Portkey docs — Budget limits
• Portkey docs — Simple and semantic cache
• Portkey docs — Hybrid deployment architecture
• Portkey docs — Configuring request logging
• Portkey docs — Coding agents
• Portkey docs — Claude Code
• Portkey docs — Codex
• Claude Code docs — LLM gateway configuration