9Router

Features

• One OpenAI-compatible endpoint. Point every tool at one local URL. 9Router translates between OpenAI, Claude, Gemini and other request formats, so a tool that speaks OpenAI can use a Claude or Gemini model. The same endpoint also serves embeddings, text-to-speech, speech-to-text, image and video generation and web search from supported providers. Official source 1 ↗ · Official source 2 ↗
• Combos with tiered fallback. Routing works through combos: a named, ordered list of models, such as your Claude subscription first, a cheap GLM model second and a free Kiro model last. Your tool calls the combo name, and 9Router moves to the next model when the current one runs out of quota or returns an error. Official source 1 ↗ · Official source 2 ↗
• Multiple accounts per provider. Add several accounts for one provider and let 9Router rotate between them round-robin or by priority, moving to the next account when one hits its quota. Official source 1 ↗
• Quota tracking and usage analytics. The dashboard shows token use per provider and model, reset countdowns for 5-hour, daily and weekly limits, and estimated costs. The estimated cost is what the same usage would cost on paid APIs, not a bill. Official source 1 ↗
• Built-in token savers. RTK compresses tool output such as git diff, grep and ls before it reaches the model and is on by default; 9Router claims 20–40% fewer input tokens. Optional Caveman mode adds a prompt that makes replies terse. One request header turns all token savers off. Official source 1 ↗ · Official source 2 ↗
• OAuth sign-in with automatic token refresh. Subscription providers such as Claude Code, Codex and GitHub Copilot are connected by signing in through OAuth in the dashboard; 9Router refreshes the tokens before they expire and tracks each account's quota resets. Official source 1 ↗
• Cloud Sync and remote access. Optional Cloud Sync copies providers, combos and settings across devices, and a Cloudflare tunnel lets you reach your router from elsewhere. Official source 1 ↗ · Official source 2 ↗

Pricing & free limits

Billing details

The dashboard's cost figures estimate what your usage would cost on paid APIs; they are not charges. Your real bill is whatever your upstream providers charge. Official source 1 ↗

Routing through 9Router does not lift a subscription's own limits: Claude Code and Codex quotas still reset on their 5-hour and weekly windows, and 9Router falls back to another model when they run out. Official source 1 ↗

Free access & limits

Free upstream tiers are capped and set by those companies: Kiro gives about 50 credits a month (plus 500 trial credits in a new account's first 30 days), OpenCode Free's model list changes without notice, and Vertex AI's $300 credit applies only to new Google Cloud accounts for 90 days. Official source 1 ↗

Best for & limitations

A Claude Pro or Max subscriber builds a combo that uses Claude first, then a cheap GLM model, then a free Kiro model, so Claude Code keeps answering after the subscription's 5-hour limit is hit. The README's example puts the extra cost at about $5 a month on top of the $20 subscription. Official source 1 ↗

Set ANTHROPIC_BASE_URL to http://localhost:20128/v1 and optionally map Claude Code's opus, sonnet and haiku aliases to 9Router model names with ANTHROPIC_DEFAULT_*_MODEL variables, then run claude --model as usual. Official source 1 ↗

Without a subscription, connect Kiro or OpenCode Free (no sign-up for OpenCode Free) in the dashboard and point Codex, Cursor or Cline at the local endpoint, within those providers' free-tier caps. Official source 1 ↗

• Your subscription accounts run through a third-party proxy. Connecting Claude Code, Codex, Copilot or Cursor means signing those accounts in to 9Router, which keeps and refreshes their OAuth tokens and sends your requests on to them. Its MITM Bridge goes further, intercepting Antigravity, Copilot and Kiro IDE traffic so those subscriptions can call other backends, and the site's only guidance is to “mind each tool's policy”. The official pages we checked say nothing more about provider terms, so read each provider's terms before connecting a paid account. Official source 1 ↗ · Official source 2 ↗
• Designed for multi-account and IP rotation. 9Router round-robins requests across several accounts per provider and offers proxy pools that rotate IPs for region-restricted providers. Providers set their own rules on accounts and access; 9Router's pages do not address them. Official source 1 ↗ · Official source 2 ↗
• Free tiers it lists have changed or ended. The README says the iFlow, Qwen Code and Gemini CLI free tiers ended in 2026 and Kiro moved to a capped free tier, while the homepage still lists iFlow, Qwen and Gemini among its free providers. Zero-cost routing depends on third-party free tiers that can change without notice. Official source 1 ↗ · Official source 2 ↗
• No unified billing or model marketplace. Unlike a hosted API marketplace, 9Router does not sell model access. You open and pay an account with each provider, and 9Router only routes between them; OpenRouter itself can be connected as one of its API-key providers. Official source 1 ↗ · Official source 2 ↗
• Token savers change requests by default. RTK compression of tool output is on by default, and Caveman mode rewrites the system prompt when enabled. Turn them off in the dashboard or per request if you need unmodified prompts. Official source 1 ↗
• You secure the stored keys yourself. Provider connections, API keys and usage history are stored in a local SQLite database. If no INITIAL_PASSWORD is set, the dashboard falls back to the password 123456, so set your own, especially on a VPS or shared machine. Official source 1 ↗