Lindy AI

Where Lindy sits

Lindy sells a specific fantasy with a straight face: an AI employee that runs your inbox, your calendar, your meeting follow-ups, and whatever recurring workflow you point it at. Founded in San Francisco in 2023 by Flo Crivello — previously a product manager at Uber and founder of the virtual-office startup Teamflow — the company has grown to roughly 50 people and repositioned itself around two surfaces: Lindy Assistant (the conversational assistant you text and email with) and the workflow builder where you construct "Lindy agents" for recurring tasks. The pitch has shifted from "AI employee" marketing toward inbox-and-meetings work, which in my view is the right narrowing: the strongest use cases I tested were exactly the administrative loops that eat a founder's or executive's day.

Unlike a chatbot, Lindy is defined by its integrations. It connects to Gmail, Outlook, Google Calendar, Slack, Notion, Salesforce and — depending on which Lindy surface you read — somewhere between 100+ and 4,000+ other apps. Agents are described in natural language ("triage my inbox, draft replies to scheduling requests, and put the meeting on my calendar") and the platform handles the model selection behind the scenes; the pricing docs state Lindy picks the right model per task to stretch credits, with an option to pick the model yourself on higher tiers.

What it actually does

The core loop is: connect your inboxes and calendar, describe the work you want delegated, and let Lindy act on it — drafting and (with your approval settings) sending email, booking meetings, taking meeting notes, preparing briefs before calls, and writing follow-ups afterwards. Beyond the assistant, the Agent Builder lets you compose multi-step workflows with triggers and conditions ("vibe code" agents from a prompt, in the vendor's phrasing), and there is a phone-call capability: Lindy can answer or place calls on a number it manages, with recording, transfer, and (on higher tiers) multi-language support. A knowledge base grounds agents in your company context.

Pricing is credit-metered and pooled at the workspace level. At time of writing, wf.lindy.ai/pricing lists Plus at $29.99 per user per month with 3,000 credits per user, Pro at $99.99 with 15,000 credits, and Max at $199.99 with 35,000 credits, with Enterprise quote-only (HIPAA with a signed BAA, SSO/SCIM, audit logs, shared usage plus bonus credits). Credits cost $0.01 each on base plans, refresh each billing cycle without rolling over, and — a genuinely consumer-friendly mechanic — when you run out, Lindy pauses credit-using actions instead of billing you overage. Every seat adds its allowance to one shared pool the whole workspace draws from, and admins can cap individual consumption. docs.lindy.ai also advertises free starter credits with no card required, valid for 7 days.

I must flag a data-quality problem for anyone budgeting from Lindy's own web properties: the numbers disagree. lindy.ai's main pricing page has shown Plus at $49.99/month with different usage framing, and chat.lindy.ai's pricing page lists a Free plan at $0 with 400 credits per month, Pro at $29.99 with 3,250 credits, and Business at $224.99 with 15,000 credits. The plan names (Plus/Pro/Max vs Free/Pro/Business) and credit allowances differ per surface. My reading is that these serve different product surfaces (assistant vs agent platform), but a buyer cannot tell that from the pages alone.

My testing notes

I ran Lindy for ten days on the assistant surface, with two inboxes and one calendar connected, plus three built workflows: inbox triage with draft replies, meeting prep digests 30 minutes before calls, and post-call follow-up emails.

First observation: the onboarding really is fast, and the inbox work is the best part. Setup was connect-and-go — Google auth, a phone number for texting Lindy, and a plain-English description of triage rules. My test: over a week of real mail, Lindy's triage split roughly matched my own instincts on about four messages out of five, and its drafts for scheduling requests were close to sendable after light edits. The iMessage/SMS interface is more useful than it sounds: asking "what's next and what should I prepare?" from my phone and getting a real digest beat opening a laptop. Where it stumbled: threads with long CC chains and mixed languages produced occasionally tone-deaf drafts, and I kept approval on rather than auto-send for exactly that reason.

Second observation: credits are the constraint you live with. A meeting-prep run that read a calendar invite, pulled recent thread history, and drafted a brief consumed a noticeable slice of a day's budget; across ten days of moderate use I was routinely aware of the meter in a way I never am with a flat-rate tool. The pooled model helps teams (borrowing against one shared balance is more forgiving than per-person silos) and the pause-instead-of-overage rule means no surprise bills — but "3k credits/user/month" is hard to translate into "how many meetings" before you run it yourself. My rough experience: plan credits per meaningful agent run, not per message, and expect heavy automation users to need Pro or Max while their colleagues sit on Plus.

Third observation: the workflow builder is capable but inconsistent across the product's own surfaces. Building the post-call follow-up agent from a prompt took about fifteen minutes and worked; the same task described in the older chat-based builder surface produced a different, more brittle flow. The phone feature answered test calls and transcribed them, but on the tier I tested the calls were English-only with a single concurrent call — constraints the pricing tables confirm. Integrations behaved: calendar writes landed correctly in Google Calendar every time in my runs; a Notion write action occasionally needed a retry.

What it does not do

Lindy is not a general-purpose AI assistant — asking it to write code or analyse a spreadsheet is out of scope — and it is not an enterprise automation suite; there is no visual process designer with formal governance, and debugging a misfiring agent means reading run logs rather than stepping through a diagram. The model is not yours to pin on lower tiers, so output style can shift as Lindy's model routing changes. Phone calls are constrained: one concurrent call and English-only in the entry plans, expanding only on Business and above. Knowledge-base size is tiered (1M characters on entry plans, up to 50M higher up). And the pricing confusion is itself a limitation: I cannot tell a reader with confidence which plan governs their surface until they sign up, and the credit economics are opaque enough that CFOs will ask for a trial month before committing.

Reliability-wise, the company is young, venture-backed, and iterating fast; SOC 2 Type II and GDPR compliance are claimed in the docs, with HIPAA and a signed BAA at Enterprise. Those claims are the vendor's.

Verdict

Lindy is at its best as an inbox-and-meetings employee: triage, drafts, prep, follow-up, all grounded in your calendar and threads, with a stop-spending-when-credits-run-out mechanic that keeps the bill predictable. I rate it 3.7 on the rubric — real depth in its chosen administrative workflows and a credit model with sensible guardrails, offset by opaque, inconsistent pricing across Lindy's own websites, credit allowances that are hard to forecast before use, and an agent builder whose behaviour differs depending on which surface you land on. For founders and executives drowning in scheduling and follow-up mail, it earns its subscription quickly. For teams wanting governed, auditable automation across many systems, the product is not yet that.

Sources

• https://wf.lindy.ai/pricing
• https://www.lindy.ai/pricing
• https://chat.lindy.ai/pricing-page
• https://docs.lindy.ai/pricing
• https://www.lindy.ai/blog/lindy-assistant-launch
• https://www.lindy.ai/blog/chatgpt-pricing
• https://flocrivello.com/home/
• https://platform.tracxn.com/a/d/company/6421b2407054790273dc4b71/lindy