Knowledge · Guides for developers
Learn how coding agents, LLM routing and AI costs fit together
Guides and explainers from the AirLLM team, grouped into four topics: coding agents, LLM routing, cost optimization and AI infrastructure. Each one answers a real question and links to the product page or setup guide that goes with it.
Coding agents
How AI coding agents work, where their tokens go, and how to route the agent you already use through Air Router (see Coding Tools).
How AI coding agents use tokens
What goes into every agent turn, why sessions grow, and where coding-agent costs come from.
Read SetupClaude Code router
Route Claude Code through Air Router: the base URL, key and model settings, and what Anthropic supports.
Read SetupCodex CLI router
Point Codex CLI at Air Router with a custom provider and the Responses API.
ReadLLM routing
What model routing is, how routers decide, and how task-aware routing applies to coding agents.
What is an LLM router?
How LLM routers choose a model for each request, and how routing differs from picking one model by hand.
Read ExplainerWhat is an AI coding agent router?
Why coding agents are a special routing workload, and what to look for in a router built for them.
Read ProductAirLLM routing models
airllm-auto, airllm-anthropic, airllm-openai and the rest: automatic model selection by provider family or use case.
ReadCost optimization
Practical ways to spend less on LLM calls without giving up the quality your work needs.
How to reduce Claude Code costs
Anthropic’s own cost guidance, plus where an LLM router fits.
Read GuideHow to reduce LLM API costs
Measure, right-size, trim context, cache and cap: a practical checklist for any LLM app.
Read ProductAI agent cost optimization
How Air Router optimizes each model request your agents make.
ReadAI infrastructure
Gateways, routers and control planes: how the layers between your code and the models fit together.
LLM gateway vs LLM router
What each one does, where they overlap, and when you need which.
Read ProductThe AirLLM platform
How AirCode, Air Router, compute and the control plane fit into one task-aware system.
Read ReferenceSecurity and data handling
Encryption in transit, AirLLM's commitment never to train models on your prompts, and how AirLLM handles your data.
ReadComparisons
Neutral, dated comparisons built from each vendor’s public documentation.
AirLLM vs OpenRouter
A hosted multi-model API compared with a router built around coding agents.
Read GatewaysAirLLM vs LiteLLM
A self-hosted open-source gateway compared with a hosted routing layer.
Read AllEvery comparison
Routers, gateways and coding agents: OpenRouter, LiteLLM, Portkey, Claude Code, Codex and Cursor.
ReadSetup guides
Step-by-step setup for each tool lives in the Developer Guide.
Connect your coding tools
Claude Code, Codex CLI, Cursor, OpenCode, Aider, Cline and more: per-tool settings for macOS, Linux and Windows.
Read Developer GuideAgent frameworks
Where to set the base URL, key and model in LangChain, LangGraph, CrewAI, AutoGen and others.
Read Developer GuideAirCode CLI
Install AirCode CLI, sign in and start your first task.
ReadFrequently asked questions
What is AirLLM?
AirLLM is an AI coding platform. Its products are AirCode, a coding agent that comes as AirCode CLI (a terminal app) and AirCode Studio (a desktop app, coming soon), and Air Router, an intelligent routing layer between your coding tools and the AI models.
What is Air Router?
Air Router is AirLLM’s model-routing layer for AI coding agents and other LLM workloads. Your tool sends its model requests to one endpoint, and Air Router picks a suitable model for each request, trims unnecessary context, reuses eligible cached results and applies budgets. See Air Router.
What is an AI coding agent router?
It’s a routing layer that sits between a coding agent and the models it calls, and chooses a model for each request or agent step based on the work being done. Read the full explainer.
How does AirLLM reduce AI coding costs?
By sending simple steps to efficient models and difficult ones to stronger models, cutting repeated context, reusing eligible cached results, and helping stop runaway loops with budgets. AirLLM cites up to 40% lower cost; actual savings depend on your workload and aren’t guaranteed.
Does AirLLM work with Claude Code and Codex CLI?
Yes. Claude Code connects through its Anthropic-compatible base URL, and Codex CLI through a custom provider in ~/.codex/config.toml. See the Claude Code and Codex CLI routing guides.
What is AirCode?
AirCode is AirLLM’s own coding agent. AirCode CLI runs in your terminal today; AirCode Studio, a desktop app for building, running and verifying Web, iOS and Android apps, is coming soon for macOS, Windows and Linux.
Put it into practice
Connect the coding tool you already use to Air Router with one API key and the compatible endpoint.