OpenLLM

Features

• One key for every model. Subscriptions such as Claude Code, Codex, Kimi Code and Grok, provider API keys, local servers such as Ollama and LM Studio, and included hosted models all sit behind one sk-llm key. OpenLLM lists 293 models across 24 providers. Official source 1 ↗ · Official source 2 ↗
• Fallback chains. Define an ordered chain per model alias. When a hop returns a rate limit, overload or provider error, the request moves to the next hop, for example from an exhausted Claude Pro or Max allowance to another plan or API key. Official source 1 ↗ · Official source 2 ↗
• Works with existing clients. One command per client runs Claude Code, Codex, OpenCode, Grok Build, Hermes and others through OpenLLM without changing their config. The ChatGPT Mac app and Raycast AI can use it too. Official source 1 ↗ · Official source 2 ↗
• Included inference. Every account gets a monthly allowance on hosted models served on OpenLLM's own key, so the first request works before you connect anything. Official source 1 ↗ · Official source 2 ↗
• Shared memory and integrations. An MCP server gives every agent shared memory across sessions. Context.dev web data turns any page into Markdown, JSON or a screenshot, and integrations added once from the marketplace show up in every client. Official source 1 ↗ · Official source 2 ↗
• Encrypted key vault. Provider keys are encrypted in your browser under a recovery phrase only you hold, and OpenLLM stores only ciphertext. There is no master key. Official source 1 ↗
• Usage and cost analytics. See requests, tokens, latency and API-equivalent cost per model and provider, with charts you can export or share. Official source 1 ↗

Pricing & free limits

Billing details

Billing runs through Stripe. Subscriptions renew automatically, and after cancelling Pro stays active until the end of the billing period. Official source 1 ↗ · Official source 2 ↗

Fees are non-refundable except where the law requires it or a separate refund policy applies. Official source 1 ↗

Only fresh input and output count toward the Free budget; cached prompt tokens don't. All monthly counters reset on the 1st at 00:00 UTC. Official source 1 ↗

Free access & limits

Free includes 50M non-cached tokens a month on your own providers and 2 connected devices. Official source 1 ↗

When the Free budget runs out, OpenLLM refuses all inference requests, including included models, until the 1st of next month. Official source 1 ↗

Best for & limitations

Keep a Claude Code session going after hitting a Claude Pro or Max usage limit by falling back to another plan or API key. Official source 1 ↗

Use one key and one usage dashboard across Claude Code, Codex, OpenCode and other agents. Official source 1 ↗ · Official source 2 ↗

Route coding agents to local models running in Ollama or LM Studio. Official source 1 ↗

• For developers. OpenLLM is aimed at developers running coding agents; setup involves a CLI or the Mac app and connecting providers. Official source 1 ↗ · Official source 2 ↗
• Daemon on macOS and Linux only. The local daemon that serves subscriptions and local models runs on Mac or Linux machines; no Windows daemon is listed. Official source 1 ↗ · Official source 2 ↗
• Free cap stops all requests. On Free, hitting the 50M token budget blocks every inference request, including included models, until the monthly reset. Official source 1 ↗
• Lost recovery phrase. OpenLLM cannot recover your vault. If you lose the phrase you reset and paste your keys again. Official source 1 ↗
• Trial needs a card. The 3-day trial needs a card and renews as paid Pro unless you cancel. Official source 1 ↗