Lindy AI — Slack-native AI agents, and what a credit actually buys

• Lindy AI

  This page

  PricingRoughly even

  • Three self-serve tiers priced per user, each granting a monthly credit allowance per seat
  • Credit allowances rise steeply between tiers rather than features being withheld
  • Enterprise is sales-led and adds shared usage, compliance paperwork and support
  • A short free trial rather than a permanent free plan

• Zapier

  PricingRoughly even

  • Free plan with a modest monthly task allowance for a single user
  • Paid tiers priced per user with task allowances that scale by tier
  • Higher tiers add more steps per task, multi-user features and premium app access
  • Charges are per task executed, which makes cost predictable per unit of work

• Make

  PricingRoughly even

  • Free plan with an operations allowance rather than a hard task cap
  • Paid tiers priced per user with rising operation and data allowances
  • Additional operations can be bought without moving tier
  • Generally the cheapest of the four at high volume, since pricing tracks operations not seats

• n8n

  PricingRoughly even

  • Free if you self-host it, which is the whole point of the project
  • Cloud tiers priced per user with execution allowances, plus an enterprise option
  • Community edition has no licence cost at all
  • The cheapest at scale for anyone willing to run the infrastructure

1. Lindy AI — official site
2. Lindy AI — pricing
3. Lindy AI — integrations
4. Lindy AI — security
5. Lindy AI — terms
6. Zapier — official site
7. Make — official site
8. n8n — official site

Free tier

No

Pricing summary

Lindy is priced per user, and the tiers are drawn by how many credits each seat receives rather than by which features are switched on — three self-serve levels that climb steeply in allowance, then an enterprise tier sold as a quote with shared usage, compliance paperwork and dedicated support. Starting is a short free trial rather than a free plan, so the honest first step is a bounded experiment rather than an open-ended free tier, and the decision to make during it is narrow: pick two or three of the routines your team would actually run, and find out what they consume. That matters because credits are consumed per unit of work the AI does, with heavier jobs — a long research brief, a meeting processed with notes and follow-ups, a multi-tool sequence — costing more than a small lookup, and because a routine that runs on a schedule spends credits whether or not anyone asked for it that day. A per-user model also means the arithmetic changes with team size in a way that a task-based platform does not: one seat is trivial to justify, and twenty seats is a budget line that has to be defended on work actually removed rather than on capability available. Enterprise adds the controls a security review will ask for, including single sign-on, audit logs and a signed compliance agreement, and it is worth knowing those sit above the self-serve line rather than inside it. Every one of these numbers — allowances, tier names, trial length — is subject to revision, so verify them on the official pricing page before you commit a team to it.

Reviewed on 19 September 2026 · Lindy AI — pricing

• A teammate that lives in Slack

  The primary surface is a Slack app that answers in threads, responds to mentions and can be addressed like a colleague. That choice does a lot of quiet work: adoption does not require anyone to learn a new interface, the work and the conversation about the work sit in the same place, and the output is visible to the team rather than trapped in someone's private chat.

• A connector library that spans the usual stack

  Well over a thousand integrations cover CRM, email, calendar, storage, project management, support desks and data warehouses, which is what lets one routine touch several systems in a single brief. The practical caution is the same as for any tool of this kind: every connected account is a permission you have granted, and the useful discipline is to connect the three systems a routine needs rather than the thirty it might.

• Scheduled routines and triggers

  Beyond answering when asked, routines can run on a schedule or fire on an event — a daily brief, a weekly report, a message that triggers a triage sequence. This is where the product stops being a chatbot and starts being infrastructure, and it is also where credit consumption becomes continuous rather than on-demand, which is the thing to watch during a trial.

• Meeting capture, notes and follow-up

  Meetings can be recorded, transcribed, summarised and turned into follow-up actions without leaving the product, and a shared meeting library keeps them findable. It is a genuinely useful bundle for teams that already run on meetings, and it is one of the clearer differences from a pure automation platform, which would need three integrations to approximate it.

• Inbox and reply drafting

  Inbox management and drafted replies are core use cases rather than add-ons: the teammate reads incoming mail, drafts a response in the house style, and either sends it or queues it for a human. The value is in the drafting rather than the autonomy — most teams keep a person in the loop, and the product is designed to make that review cheap.

• Persistent context about the company and the tools

  The product keeps a working memory of your workspace and the systems it can reach, which is what lets a later request refer back to earlier work without being re-briefed. This is also the capability with the most privacy weight: the better it knows the business, the more of the business it has access to, which is a trade to make deliberately rather than by default.

• A skill set you can extend

  Dozens of built-in skills cover the common jobs, and custom ones can be added for the specifics of a team. The honest reading is that the built-ins get you running quickly and the custom skills are where the value concentrates, because a routine shaped around your actual process needs less supervision than a generic one that keeps being almost right.

• Role-shaped use cases rather than one-size automation

  The vendor organises the product around roles — marketing, sales, support, operations, finance, engineering — with concrete routine examples for each. It is a sensible way to sell the product and a good way to evaluate it: find the two or three examples that match work your team already does by hand, run those during the trial, and ignore the rest of the list.

1. 1

   Start from a job someone already does by hand

   The routines that succeed are the ones replacing a specific recurring chore that a named person currently spends time on — triaging a support queue, drafting the weekly update, prepping for meetings. Starting from a capability list instead produces impressive demos and nothing durable, because there is no existing baseline to compare against and nobody whose week improves.

2. 2

   Brief narrowly, then widen

   A first routine should have one job, one trigger and one destination, so that when it misbehaves you can see exactly where. Widening a working narrow routine is easy; debugging a broad one that half-works is not, and a teammate that occasionally does the wrong thing across four systems will lose the team's trust faster than one that reliably does something small.

3. 3

   Keep a human in the loop for anything irreversible

   Sending mail to a customer, changing a record, posting publicly, moving money — the value of AI drafting is that review is cheap, and the value of review is that the failure modes are contained. Teams that start with send-on-approval and graduate to autonomy for specific low-risk routines end up with something durable; teams that start autonomous usually end up turning it off.

4. 4

   Measure credit consumption per routine, not per month

   Because the tiers are drawn by allowance rather than by features, the number that decides your plan is what a single routine costs to run. Instrument that during the trial: a scheduled brief that reads a lot of context is a different order of expense from a lookup that fires occasionally, and a routine nobody values but which runs daily will quietly define your bill.

5. 5

   Scope the permissions deliberately

   Every connected account is standing access to a system, so the right default is a dedicated account or app password with the narrowest scope that lets the routine work — not an admin login because it was quicker. This is also worth revisiting after a month, when it becomes clear which connections were actually used and which were granted in an optimistic first afternoon.

6. 6

   Give the routines an owner

   An AI teammate is not a set-and-forget appliance: systems change, schemas move, people switch jobs, and a routine that quietly stopped producing useful output is worse than no routine at all. Name the person who reviews the logs, owns the briefs and retires what no longer earns its credits — the teams that get lasting value from this category are the ones where somebody is accountable for it.

• Small teams without an operations hire

  The clearest fit is a company of perhaps five to fifty people where nobody owns internal automation, and where the admin — chasing invoices, updating the CRM, compiling updates, scheduling — is being absorbed by people who should be doing something else. A per-seat AI teammate is a way to buy back those hours without hiring for a role the company cannot yet justify.

• Sales and marketing teams living in email and Slack

  The vendor's own use cases cluster here for a reason: research before a call, enriched CRM updates, sequence follow-ups, competitor tracking and campaign reporting are all multi-system, context-heavy and repetitive. They are also, critically, things where being approximately right is useful — which is exactly the kind of work an AI teammate handles better than a rigid automation.

• Support teams triaging an unpredictable queue

  Triage is the textbook case: the input is unstructured, the correct action depends on context spread across systems, and the volume varies wildly. A conversational agent that reads the message, finds the account, drafts a reply and routes the ticket is a better fit for that than a branching flow, because it does not need a new branch every time a customer phrases something differently.

• Operators already comfortable delegating to AI

  The product rewards people who have internalised how to brief a model: state the outcome, give the context, define what good looks like. Teams that are still uneasy about AI making judgement calls will spend more time supervising than the routine saves, and they would be better served by a deterministic platform where every step is explicit and inspectable.

• Founders and operators who need a routine, not a platform

  There is a real distinction between wanting to automate a business and wanting three recurring jobs handled. Someone who wants a weekly competitive brief and a chaser for unpaid invoices does not want to learn a scenario builder or run infrastructure — they want to describe the job once. That is precisely the shape of problem this product is built for.

A hosted service with Slack as the primary interface

There is no self-hosted edition and no local model option: routines run on Lindy's infrastructure, with the Slack app as the main surface and a browser extension alongside it. That is what makes the product quick to adopt and is also the constraint — a workflow touching data that cannot leave your network is out of scope, not a configuration question.

Metered in credits per user rather than tasks per account

Each seat carries a monthly credit allowance, and credit cost tracks the complexity of the work rather than a simple per-run count, which means the same routine can be cheap or expensive depending on how much context it reads. Allowances rise steeply between tiers and the balance resets rather than accumulating, so the arithmetic that matters is the cost of your routines against your seats.

A large permission surface by design

The connector library is what makes cross-tool routines possible, and it is also the main risk: a teammate with standing access to mail, CRM and calendars is a significant grant. The vendor's stated posture includes encryption by default, scoped permissions that can be revoked, and an audit trail of actions; the operational discipline of narrow scopes and periodic review is yours to supply.

Compliance controls sit on the enterprise tier

Single sign-on, audit logs, a signed compliance agreement and the ability to handle regulated data are documented as enterprise capabilities rather than self-serve ones. If a security review is part of your buying process, the useful step is to establish early which controls you need and which tier actually provides them, rather than discovering the gap after a trial.

What it does not do

It is not a deterministic integration engine: if a step must execute identically every time, an LLM choosing that step is a liability rather than a feature. It is not a development platform — there is no self-hosting, no code-level control and no versioned workflow artefact to review. And it is not a document or knowledge-base product in the way a company-specific assistant is; the memory is in service of the routines, not a searchable corporate brain.

A funded early-stage vendor with a moving price list

Lindy is a venture-backed young company, which is worth weighing for a product you would be embedding into daily operations: features, allowances and tier names move quickly, and enterprise terms are negotiated rather than published. Treat the architecture as durable — Slack-native, credit-metered, connector-rich — and the specific numbers as subject to change.