LinkedIn API Rate Limits 2026: How Each API Enforces Them

LinkedIn API rate limits in 2026: the official daily quotas, and how Unipile, Edges, Linked API, LinkupAPI and Swarmhit count, refuse and reset calls.

Published 10 min read
LinkedIn API Rate Limits 2026: How Each API Enforces Them

LinkedIn's official API limits every app per day: a cap on the calls the whole application makes and a cap on the calls made with each member's token, both reset at midnight UTC, and both answered with HTTP 429 once spent. LinkedIn does not publish the numbers. Each endpoint has its own, and you read them in your app's Analytics tab in the Developer Portal. Unofficial LinkedIn APIs, the ones that drive a member account you connect, add their own layer on top, and each provider counts something different: requests, actions, or nothing at all.

We read the rate-limit documentation of the official API and of five account-based providers (Unipile, Edges, Linked API, LinkupAPI and Swarmhit) on 14 November 2026. The comparison below is what each one counts, who sets the cap, what your code receives when it is crossed, and when the window frees up again.

What are the rate limits on LinkedIn's official API?

The LinkedIn API rate limiting page on Microsoft Learn defines two daily limits. The application limit is the total number of calls your app can make in a day. The member limit is the number of calls one member's token can make through your app in a day. Both reset at midnight UTC, and a call over either one gets a 429. LinkedIn adds that it may also return a 429 to protect its infrastructure, in which case service returns to normal by itself.

The numbers vary by endpoint and are not published. The Developer Portal shows them in the app's Analytics tab, but only for endpoints the app has called at least once that UTC day, so the documented way to discover a limit is to make one test call and refresh the page.

That last detail matters for multi-tenant products. If one customer's token burns its member limit at 10:00 UTC, no email is sent, and the first signal is a 429 on that customer's calls until midnight UTC. The official API also only covers what LinkedIn approves for your app, which leaves out the messaging and invitation actions most outreach products need. That gap is why the providers below exist, and our LinkedIn API guide covers which official products are open to whom.

Why do unofficial LinkedIn APIs have two layers of limits?

An account-based API signs in as a real LinkedIn member, so two different systems can refuse a call. The provider can refuse it before it leaves its infrastructure, by its own counting rules. LinkedIn can refuse it, or worse, accept it and later restrict the account, by the behavioral rules it applies to any member: invitations per week, profile views, messages to strangers. Those member-side rules are the ones our LinkedIn connection limits page tracks against LinkedIn's help center. The sign-in that comes before any of these limits, with its 2FA codes and CAPTCHAs, is covered in our guide to the LinkedIn account connection API.

The two layers show up in the error itself when the provider is careful about it. Unipile returns its own refusal as api/too_many_requests and LinkedIn's as provider/too_many_requests, and warns that its rate-limit headers describe its own window, not LinkedIn's. Edges does the same with LIMIT_REACHED for its quotas and STATUS_429 for a raw LinkedIn 429. A retry policy that treats both the same way will either wait too long on a short provider window or hammer LinkedIn on a long one.

How does each LinkedIn API provider enforce rate limits?

APIWhat is countedWho sets the capWhat your code gets when it is crossedWhen it frees up
LinkedIn official APICalls per app and per member token, per dayLinkedIn, per endpoint, not published429Midnight UTC
Unipile (v2)Requests per connected account and per route, in up to two windows per ruleUnipile429 api/too_many_requests, x-ratelimit headers and retry-afterWhen the window that started at the first request expires
EdgesAllowances per identity and action type, a throughput ceiling per identity, and requests per workspaceEdges (Smart Limits)LIMIT_REACHED with the window that tripped and postponed_until in live mode; async and scheduled runs are postponed insteadRolling windows, from one minute to about 31 days for InMail
Linked APIActions per account in 12 categories, daily, weekly or monthlyYou, starting from defaults such as 50 messages and 10 connection requests a dayThe action returns limitExceeded instead of runningAt the end of the period you configured
LinkupAPINothing on LinkedIn: "We don't enforce any limits on our side"LinkedInLinkedIn's own 429 or 500Whenever LinkedIn lifts it
SwarmhitActions per sender and per kind of action, daily, rolling weekly and pending invitationsSwarmhit defaults by LinkedIn plan, adjustable within a range429 with limit, used, scope and resetAt00:00 UTC for daily caps, when the oldest invitation leaves the 7-day window for the weekly one

Three designs sit in that table. Unipile counts requests, the way a classic API gateway does, and its own documentation draws the line: the rate limiter "does not replace good usage practices", because LinkedIn also watches account age, activity patterns and reply rates, which a request counter cannot see. Edges and Swarmhit count LinkedIn actions per account, which maps to what LinkedIn actually judges. Linked API counts actions too, but the numbers are yours to choose, and its guide ends on "experiment carefully with your own limits". LinkupAPI publishes recommendations (80 to 100 invitations a day on a paid account, 30 to 50 messages) and leaves enforcement to LinkedIn. Our Unipile review, Edges review, Linked API review and LinkupAPI review cover the rest of each product.

How long does a burst take to clear on each design?

Window semantics decide how long a blocked account stays blocked, and they differ more than the caps do. Unipile's windows are not calendar windows: one starts at the first request when none is active and keeps a fixed length, so a one-minute window opened at 14:23:17 expires at 14:24:17, not at 14:24:00, and later requests do not extend it. Its documented 429 looks like this:

Unipile 429 (from Unipile's rate-limit documentation)
HTTP/1.1 429 Too Many Requests
x-ratelimit-limit: 10
x-ratelimit-remaining: 0
x-ratelimit-reset: 34
retry-after: 34

Edges publishes the throughput ceiling it applies to every identity, whatever the mix of actions: 5 actions a second, 50 a minute, 500 per 15 minutes, 2,500 per 4 hours and 15,000 per 24 hours. Those windows roll. At 50 a minute, an identity needs at least 10 minutes to fire 500 actions, so one that does it between 09:00 and 09:10 cannot act again until its first actions leave the 15-minute window at 09:15. Even a perfectly spread day tops out at 15,000 actions, roughly one every 5.8 seconds. These ceilings apply to any mix of actions, on top of the separate allowance each action type has.

The official API is the bluntest: a member who exhausts a quota at 00:05 UTC waits almost 24 hours. Swarmhit's daily caps reset at 00:00 UTC as well, while its weekly invitation cap rolls, so a sender at the weekly cap gets one slot back each time the oldest invitation of the last 7 days ages out.

What should your code do when it gets a 429?

Read the error before reading the clock. Every provider above returns enough to tell a provider window from a LinkedIn refusal, and the right wait is different for each. On Swarmhit, a 429 carries details when the sender hit one of Swarmhit's caps, and none when LinkedIn itself throttled the sender. A minimal handler, written by us against the fields in Swarmhit's LinkedIn safety documentation:

Handling a Swarmhit 429 (our code)
type LimitDetails = { limit: number; used: number; scope: "day" | "week" | "pending"; resetAt?: string };

async function onTooManyRequests(senderId: string, action: string, body: { error: { details?: LimitDetails } }) {
  const d = body.error.details;
  if (!d) {
    // No details: LinkedIn throttled the sender itself. Back off hard, widen the gaps.
    await pauseSender(senderId, { hours: 24 });
    return;
  }
  if (d.scope === "pending") {
    // No resetAt: the invitation backlog clears as invites are accepted or withdrawn.
    await withdrawStaleInvitations(senderId);
    return;
  }
  // day or week: store the reset time and gate future calls on it locally.
  await saveGate(senderId, action, new Date(d.resetAt!));
}

The 24-hour pause is our own choice for the example, not a documented value. What is documented is the rule behind it: never sit in a retry loop on a 429, because retrying does not free the window and the burst itself is a pattern LinkedIn notices.

What breaks in production?

A retry loop on 429. The symptom is an account that gets restricted while every individual call looked fine. The cause is a generic HTTP client that retries 429s with short backoff, turning one refused invitation into dozens of attempts in a minute. The fix is to treat a 429 on a LinkedIn action as a gate, stored per account and per action, and to retry only after the reset time the API gave you.

Provider headers read as LinkedIn's. A worker sees x-ratelimit-reset: 34 on a Unipile provider/too_many_requests and retries in 34 seconds. Unipile's documentation says those headers do not describe the provider's limit, so the retry hits LinkedIn again while it is still refusing. Branch on the error type first.

A scheduler that wakes at midnight. A job queue that releases everything at 00:00 UTC works on the official API and on Swarmhit's daily caps, and fails on rolling windows. On Edges or on Swarmhit's weekly invitation cap, the slots come back one by one as old actions age out, so a midnight release runs straight into the limit again. Use the reset time each response gives, per account, instead of a global clock.

One tenant draining a shared app quota. On the official API, the application limit is shared by every member using your app. A single customer running a heavy sync can push the whole app over its daily quota, and the 75% email arrives an hour or two later. Meter calls per tenant in your own code before LinkedIn does it for you.

Caps nobody chose. On Linked API the caps are yours, and an account added in a hurry keeps whatever defaults it got, which may be too high for a new or inactive account. Set limits when the account is connected, not after the first warning.

Is a rate limiter the same as account safety?

No, and the providers that enforce limits say so themselves. A limiter answers "how many", while LinkedIn also judges "how": the gaps between actions, the time of day, the share of invitations accepted, how new the account is. Edges notes that its quotas alone will not pace live calls. Unipile says its limiter does not replace good usage practices. Our page on whether LinkedIn automation is safe covers the behavioral side, and what to do after a restriction covers the aftermath.

Swarmhit splits the two jobs by API. On the Smart API, campaigns are scheduled and paced by the engine inside send windows, with per-sender caps enforced server-side on every action, direct calls included, and new accounts warming up before they reach full volume. On the Raw API, in early access, your code decides the pace, and Swarmhit refuses what a runaway client would do before anything reaches LinkedIn: one write at a time per account, a gap between writes, daily caps per kind of action, and a rest after LinkedIn pushes back. Prices for both are on the pricing page.

Let the engine own the limits

Swarmhit's Smart API paces every LinkedIn sender, enforces per-sender caps server-side and tells your code exactly which window it hit, with limit, used, scope and resetAt on every 429.

LinkedIn from $16.90 per sender a month, tiered down to $6.90; cold email from $59 a month for 125,000 sends.

FAQ

Does LinkedIn publish its API rate limits?

No. Microsoft Learn states that standard rate limits are not published in the documentation and vary by endpoint. You see them in the Developer Portal: select your app, open its Analytics tab, and the page lists usage and limits for each endpoint your app called at least once that UTC day. To see the limit of an endpoint you have not called yet, LinkedIn's documented method is to make one test call to it, then refresh the Analytics page.

When do LinkedIn API rate limits reset?

The official API's application and member limits reset every day at midnight UTC. Account-based providers reset on their own schedules. Unipile's windows start at the first request and keep a fixed length, Edges uses rolling windows from one minute to about 31 days for InMail, and Linked API resets at the end of the daily, weekly or monthly period you set. Plan retries on the reset time each API returns, not on midnight.

Does LinkedIn warn you before you hit the limit?

Only for application limits. LinkedIn emails your app's developer admins when the app passes 75% of an assigned quota, with a delay of about 1 to 2 hours, and sends a new alert if another endpoint crosses the threshold later the same day. Nothing is sent when a single member's limit, or a combined member and app limit, is reached, so a per-member 429 is the first signal you get.

Do requests served from cache count against Unipile's rate limit?

No. Unipile's rate-limit documentation says requests served from cache do not count, that counters are kept separately for each connected account, and that a rule can have up to two windows that a request must fit at the same time. On accepted requests the x-ratelimit headers describe the last window checked, and on a Unipile 429 they describe the first window that refused the call, with retry-after added.

What happens to an Edges run when it hits a limit?

It depends on the run mode. In live mode the request fails with a limit error, and Edges recommends waiting until the postponed_until time it returns rather than computing the wait yourself. In async and scheduled modes the work is postponed and resumes automatically once the window frees up, so a batch keeps going without your code retrying anything. Raw LinkedIn 429s come back separately as STATUS_429, with exponential backoff advised.

Do Sales Navigator accounts get higher limits?

On LinkedIn's side, often yes, and providers reflect it. Unipile recommends about 1,000 profiles a day from standard search and 2,500 from Sales Navigator or Recruiter, and extra profile-visit budget for those accounts. Swarmhit enforces the same 2,500 and 1,000 daily ceilings on search imports per sender. Edges tracks allowances per identity and per action type, and counts InMail on its own long window of about 31 days.

Alexandre Risser

Written by

Alexandre Risser

Swarmhit

Building Swarmhit. Writes about LinkedIn outreach, multi-sender infrastructure, and outbound that books meetings.

Ready to ship LinkedIn, email and WhatsApp outreach in your product?

The Swarmhit Smart API runs sequences, senders, mailboxes and the inbox behind one API, with the whole logic handled. Public pricing per channel, from $16.90 an account.

Keep reading