Stop losing tokens to rate limits

Your keys fail.
Your app doesn't.

One API key hits a rate limit and your entire pipeline stalls. APIShift pools your keys, rotates to healthy ones instantly, and only falls back when every key is cooling down.

Install

python$ pip install APIShift

npm$ npm install @apishift/core

Get started See the demo ↓

See it break. See it recover.

Rate limited? Already shifted.

Watch APIShift exhaust a Gemini key pool — keys fail, rotation kicks in, and when the whole pool is down, a fallback provider picks up mid-sentence without losing a single token.

apishift terminalv1.0.0

gemini-key-1

gemini-key-2

openrouter

Active

Shifting

Failed

Standby

< 5ms

Failover overhead

Tokens lost on shift

Providers supported

Three ways to start

Copy. Paste. Ship.

Drop the agent prompt into Cursor, paste the Python snippet, or wire up the TypeScript SDK — you're resilient in under a minute.

Coding Agent

// Paste into Cursor, Windsurf, Claude Code // or any AI coding agent: Add APIShift to my project with a Gemini key pool using adaptive rotation, streaming, memory, and fallback only if needed at .apishift/memory.jsonl

Python

from APIShift import Conversation conv = Conversation.from_gemini_key_pool( api_keys=['KEY_1', 'KEY_2', 'KEY_3'], key_strategy='adaptive', memory_path='.apishift/memory.jsonl', ) for chunk in conv.send_message_stream("Hello"): print(chunk, end="", flush=True)

TypeScript

import { APIShift } from '@apishift/core'; import { google } from '@ai-sdk/google'; const shift = new APIShift([ { provider: 'gemini', keyIndex: 0, model: google('gemini-1.5-flash', { apiKey: process.env.GEMINI_KEY_1 }) }, { provider: 'gemini', keyIndex: 1, model: google('gemini-1.5-flash', { apiKey: process.env.GEMINI_KEY_2 }) } ], { keyStrategy: 'adaptive' }); await shift.streamMessage('Hello');

Built for production

Everything your API keys need.

✓
One provider, many keys
Pool 2, 5, or 50 keys from the same provider. APIShift treats them as one elastic capacity — not scattered credentials.
✓
Smart rotation, not round-robin
Skips cooling-down keys, spreads load by usage count, and respects Retry-After headers automatically.
✓
Fallback is the last resort
Cross-provider failover only triggers after every key in your primary pool has been tried or is cooling down.
✓
Zero context loss
Mid-conversation key or provider changes carry the full conversation history. The model never loses track.
✓
Stream-safe failover
Failover happens before tokens reach the client — not after. Works with Python generators and Vercel AI SDK streams.
✓
Know which key handled what
Inspect provider, pool, key label, attempts, and cooldowns after every call — without ever exposing raw API keys.

Your keys fail.Your app doesn't.

Rate limited? Already shifted.

Copy. Paste. Ship.

Everything your API keys need.

Stop babysitting API keys.

Your keys fail.
Your app doesn't.