LiteLLM alternatives for gateways and coding agents
AutoRouter7 min read
For coding agents in Codex and Claude Code, use AutoRouter, the Mac app for Codex and Claude Code: it needs no gateway or API keys, uses the ChatGPT and Claude plans you already pay for and picks a model for each request. For other jobs, Bifrost is the closest self-hosted swap, Portkey, Requesty and OpenRouter remove the servers, and Vercel AI Gateway and Cloudflare AI Gateway fit apps already on those platforms. Claude Code Router is a free local proxy for one machine.
AutoRouter wrote this post, so weigh its rows accordingly. Every claim about another product comes from that product's own docs or pricing page, linked where it is used and read on November 16, 2026.
Why do teams look for a LiteLLM alternative?
Most teams look because LiteLLM is a server they have to run. It is an open-source gateway that calls 100+ LLMs in the OpenAI format, MIT licensed outside its enterprise directory, and features such as SSO, SCIM and audit logs need an Enterprise license at custom prices (GitHub, Enterprise). Its Claude Code setup with a Claude plan needs LiteLLM v1.81.14 or later with PostgreSQL (tutorial).
Supply-chain risk comes up too: on March 24, 2026, two LiteLLM releases on PyPI, 1.82.7 and 1.82.8, were compromised. LiteLLM says its official proxy Docker image was not affected, removed both releases, and shipped a clean v1.83.0 on March 30 from a rebuilt pipeline (LiteLLM).
Which LiteLLM alternatives fit which job?
Eight options cover what LiteLLM does, each best at one job:
- Bifrost is best for teams that want to keep a self-hosted gateway and make it faster. It is open source under Apache 2.0, connects 23+ providers, and its README cites about 11 microseconds of overhead per request at 5,000 requests a second on a t3.xlarge (GitHub). It forwards a Claude Code subscription sign-in on its
/anthropicroute while applying its own virtual keys (docs). - Portkey is best for organizations that want budgets, rate limits, logs and fallbacks without running servers, though it can be self-hosted too. Developer is free with 10,000 logs a month and Production is $49 a month as of November 16, 2026; its pricing page now says Portkey is Prisma AIRS AI Gateway, part of Palo Alto Networks (pricing, Claude Code docs).
- Requesty is best for one hosted bill across 600+ models, with fallbacks and spend by repo, branch and developer. Pay as you go adds 5% to model prices (pricing, Claude Code docs).
- OpenRouter is best for one key and one balance across hundreds of models. Inference is billed at list price, with a 5.5% fee on credit purchases on Standard (pricing).
- Vercel AI Gateway is best for apps on Vercel. It charges no markup on tokens, sets budgets by team, project, key or member, and
vercel ai-gateway coding-agents setupconnects Claude Code, Codex and Cursor among nine agents (pricing, changelog). - Cloudflare AI Gateway is best for apps on Cloudflare. Analytics, caching and rate limiting are free, and Unified Billing charges 5% on credit purchases (pricing).
- AutoRouter is best for developers who code with agents on ChatGPT, Claude, Grok or Gemini subscriptions. It is a Mac app that adds 20+ models to the model picker in Codex and Claude Code, with no API keys or endpoints to set up, and it also works in Cursor, Claude Desktop, T3 and Terminal. It uses the plans you already pay for first, lets any of those apps use any of your plans (a Claude plan in Codex, or a ChatGPT plan in Claude Code), and picks the model for each request after your own rules. Free costs $0 and routes across two subscriptions; Pro is $50 a month with $20 of included usage and connects any number of plans (AutoRouter).
- Claude Code Router is best for one developer on one machine. It is a free, MIT-licensed local gateway for Claude Code, Codex and other agents, as a desktop app, CLI or Docker image (GitHub).
For hosted APIs in particular, OpenRouter alternatives for coding agents compares fees on the same $100 of usage, and Claude Code Router alternatives covers local proxies.
How do the LiteLLM alternatives compare?
They differ most in who runs them, as of November 16, 2026:
| Tool | Best for | Who runs it | License or price |
|---|---|---|---|
| LiteLLM | Self-hosted team gateway | You, on your servers | MIT outside enterprise; Enterprise custom |
| Bifrost | Faster self-hosted gateway | You, on your servers | Apache 2.0 |
| Portkey | Budgets, logs and fallbacks | Portkey, or you | Developer free; Production $49/month |
| Requesty | One hosted bill | Requesty | Free tier; 5% markup |
| OpenRouter | One key for many models | OpenRouter | List price; 5.5% fee on credits |
| Vercel AI Gateway | Apps on Vercel | Vercel | No token markup |
| Cloudflare AI Gateway | Apps on Cloudflare | Cloudflare | Core features free |
| AutoRouter | Coding agents on subscriptions | A Mac app on each machine | Free $0; Pro $50/month with $20 of usage |
| Claude Code Router | One developer's machine | You, on your machine | MIT, free |
What does running Claude Code through LiteLLM take?
Running Claude Code on a Claude plan through LiteLLM takes four steps, from LiteLLM's tutorial (tutorial):
- Run the LiteLLM proxy, v1.81.14 or later, with PostgreSQL.
- Add an
anthropic/*entry tomodel_listinconfig.yamland restart the proxy. - Create a virtual key in the dashboard for all Anthropic models.
- Point Claude Code at the proxy with
ANTHROPIC_BASE_URL, send the key inANTHROPIC_CUSTOM_HEADERS, then sign in with your Claude plan.
export ANTHROPIC_BASE_URL="https://litellm.example.com"
export ANTHROPIC_CUSTOM_HEADERS="x-litellm-api-key: Bearer <your-virtual-key>"
claude
LiteLLM then logs each session and applies budgets and guardrails, and it reports costs at Anthropic API list prices even for requests your plan pays for. You still choose each model yourself.
AutoRouter takes three steps with nothing to host: download it, sign in to your plans and click Enable for each app, then choose AutoRouter as the model in a new session. It picks the model for each request after your own rules, and when one plan reaches its limit it moves on to the plan with the most room left before it spends any credit.
Which alternative suits a team rolling out coding agents?
For a team whose developers use Codex and Claude Code on subscriptions, AutoRouter Enterprise rolls out to every Mac through Jamf or another MDM, with nothing for anyone to set up. It sets allowed models, team budgets and zero data retention in one place, reports spend by team, project or ticket, and includes SSO, SCIM and an audit log. It is priced by seats; contact sales@autorouter.com.
For a team that needs one gateway for every app it ships, LiteLLM Enterprise, Bifrost and Portkey keep that job in one place, with virtual keys and budgets (LiteLLM, Bifrost, Portkey). How to roll out AI coding tools to an engineering team covers the plans and seats side of the same decision.
When should you stay on LiteLLM?
Stay on LiteLLM when prompts must stay in your own infrastructure, when you need one OpenAI-format endpoint for the apps you ship, or when you have already built budgets and routing around it. Self-hosted LiteLLM can run in your cloud or an air-gapped network (Enterprise).
AutoRouter needs macOS 14 or later, and requests pass through its servers on their way to each provider. It doesn't store transcripts; it keeps the message that opens each request, up to 4,000 characters, to review routing (privacy). For one tool at a time, see AutoRouter vs LiteLLM, AutoRouter vs Portkey, AutoRouter vs Requesty and AutoRouter vs Vercel AI Gateway. What is an LLM router? explains the kinds of router behind all of these.
Questions
Is LiteLLM's auto router the same as AutoRouter?
No. LiteLLM's auto router is a feature of the open-source LiteLLM gateway that maps requests to models you configure (docs). AutoRouter is a separate Mac app for Codex and Claude Code by Fractal Works, Inc.
What is the best open-source LiteLLM alternative?
Bifrost is the closest open-source swap for a team gateway, under Apache 2.0 (GitHub). Portkey also publishes an open-source gateway to host yourself (pricing), and Claude Code Router covers a single developer's machine under MIT (GitHub).
Is there a LiteLLM alternative that needs no API keys?
Yes. AutoRouter signs in to your ChatGPT, Claude, Grok and Gemini plans and adds its models to the picker in Codex and Claude Code, with no keys or endpoints to set up. Pro adds every other model, paid from $20 of included usage each month and then credits.
Does a LiteLLM alternative work with OpenAI's Codex?
Yes. AutoRouter works in Codex, the app and the CLI, and adds itself to Codex's config.toml when you click Enable. Vercel AI Gateway's setup command and Claude Code Router also support Codex (Vercel, Claude Code Router).
Sources
- AutoRouter
- AutoRouter privacy
- LiteLLM on GitHub
- LiteLLM Enterprise
- LiteLLM, Claude Code with a Claude subscription
- LiteLLM, security update, March 2026
- LiteLLM, auto routing
- Bifrost on GitHub
- Bifrost, Claude Code
- Portkey pricing
- Portkey, Claude Code
- Requesty pricing
- Requesty, Claude Code
- OpenRouter pricing
- Vercel AI Gateway pricing
- Vercel, coding agents in one command
- Cloudflare AI Gateway pricing
- Claude Code Router on GitHub