@inferagraph/redis-cache-provider

@inferagraph/redis-cache-provider

Redis-backed CacheProvider for @inferagraph/core's AIEngine. Suitable for production deployments where the built-in in-memory lruCache is insufficient — e.g. when you want a cache shared across server processes, persisted across browser reloads, or sized larger than available process memory.

Honors the same maxEntries + ttl semantics as lruCache, so swapping providers is a one-line change.

Installation

pnpm add @inferagraph/redis-cache-provider @inferagraph/core redis

Usage

import { AIEngine } from '@inferagraph/core';
import { redisCache } from '@inferagraph/redis-cache-provider';

const cache = redisCache({
  url: process.env.REDIS_URL ?? 'redis://localhost:6379',
  // Optional bounds. If both are unset, defaults to (500, '24h') — same as lruCache.
  maxEntries: 1000,
  ttl: '12h',
});

const engine = new AIEngine({ /* ... */ });
engine.setCache(cache);

Configuration

Option	Required	Description
url	One of url / client	Redis connection URL (e.g. redis://localhost:6379, rediss://...).
client	One of url / client	Pre-built node-redis client. When provided, url is ignored. Useful for tests and connection-pool reuse.
prefix	No	Key prefix to namespace this cache (default infera:cache:). Lets multiple InferaGraph instances share a Redis instance without collisions.
maxEntries	No	Maximum entries to retain. -1 disables the bound. See defaults note below.
ttl	No	Time-to-live per entry. Number (ms) or duration string (5m, 2h, 7d, 1w). -1 / '-1' disables the bound.

Defaults: when both maxEntries and ttl are unset, the provider defaults to (500, '24h'). When only one is set, the unset bound is treated as no-limit. This matches lruCache's contract.

Notes

• The provider connects lazily on first operation, so constructing it is cheap and won't throw.
• maxEntries is enforced by maintaining a Redis ZSET index at ${prefix}__index keyed by insertion timestamp; on overflow, the oldest key is evicted (insertion-order; not strict LRU).
• clear() uses SCAN (never KEYS) so it is safe on large datasets.

License

MIT