skill-backend

name

SKILL-backend

description

Users table - supports both anonymous (24hr TTL) and registered accounts. Anonymous users get a generated ID and no credentials. They are auto-purged after `expiresAt`. Registered users provide email/password and persist indefinitely until explicitly deleted. Use when: (1) calling its 122 API functions, (2) understanding its 212 type definitions, (3) working with its 3 classes, (4) user mentions "@llmtxt/backend" or asks about its API.

@llmtxt/backend

Users table - supports both anonymous (24hr TTL) and registered accounts. Anonymous users get a generated ID and no credentials. They are auto-purged after expiresAt. Registered users provide email/password and persist indefinitely until explicitly deleted.

Quick Start

npm install @llmtxt/backend

API

Function	Description
countTokens()	Count tokens using the cl100k_base BPE tokenizer (GPT-3.5/GPT-4 compatible). Returns the exact number of tokens the given text would consume in a GPT-4 / Claude-compatible API call. This is significantly more accurate than the ceil(len / 4) heuristic for content with non-ASCII characters, code, or structured markup.
getDocumentCacheKey()	Generate cache key for a document
shouldSkipCache()	Check if cache should be skipped based on query params
cacheDocumentContent()	Middleware to cache document content Usage: app.get('/documents/:slug', cacheDocumentContent, async (request, reply) = ... )
cacheDocumentMetadata()	Middleware to cache document metadata
setCachedContent()	Store document content in cache
setCachedMetadata()	Store document metadata in cache
invalidateDocumentCache()	Invalidate cache for a document
invalidateAllCache()	Invalidate all cache
getCacheStats()	Get cache stats for both caches
keyGenerator()	Generate a stable rate-limit key for the request. Uses the most specific identifier available: 1. Hashed API key (from Bearer token) — identifies the key, not the user 2. User ID (from session cookie) 3. Client IP address
getTierMax()	Return the rate limit max for the given category based on the request's auth tier.
registerRateLimiting()	Register the global rate limiter on the Fastify instance. Must be called AFTER CORS and compression plugins but BEFORE route registration. The global limit applies to all routes; individual routes may override with stricter config via writeRateLimit or authRateLimit. The /api/health endpoint is explicitly skipped via the skip function.
adaptiveThrottle()	Adaptive throttle hook: adds artificial delay when a client approaches their rate limit ceiling ( 20% remaining). This smooths out burst traffic by slowing requests progressively rather than hard-cutting at the limit. Maximum induced delay is 500ms. Attach as a preHandler hook on routes where burst smoothing is desired.
enforceContentSize()	Enforce the maximum document content size. Reads content from the request body and rejects with 413 if the UTF-8 byte length exceeds CONTENT_LIMITS.maxDocumentSize. Safe to call on any route that accepts a content body field.
...	107 more — see API reference

Key Types

• User
• NewUser
• Session
• NewSession
• Document
• NewDocument
• Version
• NewVersion
• StateTransition
• NewStateTransition

References

• references/CONFIGURATION.md — Full config options
• references/API-REFERENCE.md — Signatures, parameters, examples