Evenfire is an intelligence layer platform for running LLM agents that take real actions, on your infrastructure
TypeScript
import type { NextFunction, Request, Response } from 'express'
import { PluginWorkloadError } from '../../domain/errors'
/**
* Per-workload rate limiting (plan §3.2): sliding one-minute request window
* plus a concurrent-request ceiling, both keyed by the caller workload id.
* In-memory is correct here — the SDK server is a single pod-local process.
*/
export interface RateLimitOptions {
maxRequestsPerMinute: number
maxConcurrent: number
now?: () => number
}
interface CallerState {
timestamps: number[]
inFlight: number
}
export function createRateLimitMiddleware(opts: RateLimitOptions) {
const callers = new Map<string, CallerState>()
const now = opts.now ?? (() => Date.now())
return (req: Request, res: Response, next: NextFunction): void => {
const callerRef = req.pluginWorkloadCallerRef ?? 'unknown'
let state = callers.get(callerRef)
if (!state) {
state = { timestamps: [], inFlight: 0 }
callers.set(callerRef, state)
}
const t = now()
state.timestamps = state.timestamps.filter(ts => t - ts < 60_000)
if (state.timestamps.length >= opts.maxRequestsPerMinute) {
const err = new PluginWorkloadError(
'quota_exceeded',
`rate limit of ${opts.maxRequestsPerMinute} requests/minute reached`,
false
)
res.status(err.httpStatus).json(err.toBody())
return
}
if (state.inFlight >= opts.maxConcurrent) {
// 503 + retryable: the caller should back off and retry (plan §5.1).
const err = new PluginWorkloadError(
'provider_unavailable',
`concurrent request limit of ${opts.maxConcurrent} reached`,
true
)
res.status(err.httpStatus).json(err.toBody())
return
}
state.timestamps.push(t)
state.inFlight += 1
res.once('close', () => {
state.inFlight = Math.max(0, state.inFlight - 1)
})
next()
}
}