Open-source AI agents for business, priced per finished task

The four ways agents are billed

Four billing shapes sit on vendor pricing pages today, and each measures a different quantity. Per outcome charges for a result. Per credit charges for metered work. Per task charges for steps. Per seat charges for people. One month of the same work produces four bills, and the gap between the cheapest and the dearest runs to ten times.

Per outcome. Intercom's Fin AI Agent costs $0.99 per outcome on all plans, and one conversation produces one outcome at most. Intercom charges nothing when Fin cannot answer. Voice runs at $1.99 per voice outcome, sold through sales. A support queue of short chats pays by the conversation, so ten seconds and ten minutes cost the same.

Per credit. Lindy opens with $50 in free credits, no credit card required, and those credits last 7 days. Past the trial, every active person in the workspace is billed from $29.99 a month for 3,000 credits each (its pricing page, 27 October 2026), pooled across the workspace. Top-ups run $10 per 1,000 credits, and monthly credits do not roll over. Kortix sells credits at 100 to the dollar: free is $0 with 500 credits per month, every feature and 1 project, and the paid plan is credits: 25 tiers from $50 a month (5,000 credits) to $50,000 a month (5,000,000 credits). Neither page calls a credit a finished task, and a credit buys different work on each.

Per task. Zapier bills Copilot, AI steps, code and SDK calls on one task-based model, and its free plan includes 100 tasks a month. The unit is a step. A job that takes five steps bills five tasks, so a plan's task allowance is not a count of finished work.

Per seat. Salesforce Agentforce sells Flex Credits, conversations or per-person licensing, and PayGo bills monthly in arrears with no upfront commitment. Intercom prices its helpdesk per seat, separately from the outcome rate. Lindy bills every active person in the workspace. Headcount becomes the meter, and one shared pool spreads a single person's credits across the team.

Converting them to one unit

A finished piece of work is the only unit that survives a renewal: a resolved conversation, a filed report, a merged change. Converting a published rate into that unit takes one number from your own work, the amount the vendor meters behind a task.

Outcome billing converts without arithmetic. One outcome per conversation at 99 cents makes a resolved conversation cost 99 cents.

Credit billing converts with a division. Kortix at 100 credits per dollar makes one credit worth one cent. Lindy's top-up at $10 per 1,000 credits lands on the same cent. Multiply the credits your own task consumes by that cent, and a ten-credit task costs ten cents.

Task billing converts with a second division, on the plan you hold. The first 100 tasks are free on Zapier. Past that, the price of a task is the plan's price divided by the tasks the plan includes, and the count to track is steps per finished job.

Seat billing runs the same division backwards. A $29.99 active person carries 3,000 credits, which is 300 tasks at ten credits each, so a person who uses the whole pool pays about a dime a task: $29.99 ÷ 300. The same person finishing 50 costs $29.99 ÷ 50, about 60 cents a task, and finishing five costs about six dollars.

None of the four needs a forecast or an analyst's range. Each uses a rate the vendor publishes and one number you can measure in a week.

What a 500-task month costs on each

Take 500 finished tasks in a month, and 10 credits for each task on the two credit vendors. The workload is the same on every line; only the meter changes. Rates here were read on 27 October 2026, and all four vendors revise them.

• Per outcome, Intercom at $0.99 per outcome: $495, with each conversation counted as one finished task.
• Per credit, Kortix at 100 credits per dollar, 10 credits a task: 5,000 credits, $50.
• Per credit, Lindy top-ups at $10 per 1,000 credits, 10 credits a task: 5,000 credits × $0.01, $50.
• Per task, Zapier: 400 tasks billable after the free 100, at the rate on the plan you buy.
• Per seat, Lindy at $29.99 for 3,000 credits: $29.99 for the first active person, then 2,000 top-up credits at $10 per 1,000, so $20 more and $49.99 for 5,000 credits. Two active people cost $59.98 and carry 6,000.

The two credit vendors land within a cent of each other: Kortix sells 100 credits for a dollar and Lindy's top-ups sell 100 for a dollar, while the seat rate divides to $29.99 ÷ 3,000, a hair under a cent a credit. That agreement is arithmetic, and it says nothing about how much work a credit buys on either one. The per-outcome bill runs ten times higher, and it stays higher only while a conversation and a finished task are the same object. For a support queue they are close. For a research report they are not.

Where the bill surprises a team

Voice. Moving a fifth of conversations to Fin's voice outcome at $1.99 takes a 500-conversation month from $495 to about $595. The plan page quotes the text price, and the voice rate arrives through sales.

Expiry. Lindy's monthly credits refresh at the start of each cycle and do not roll over. A quiet month loses the credits it did not use, and the next busy month starts from zero.

Steps. A workflow that gains a verification step raises the task count for the same finished job. Nobody changed the plan, and the bill moves.

Loops. A retry loop on a metered plan bills every attempt. Kortix's brake is a gate you set: "Approval gates you set. Off until you set them."

The meter that leaves. "Run it on Kortix Cloud, in your VPC, or on your own on-prem network. Self-host is free." Self-host adds no line to the bill; the machines and the model keys are yours.

Before signing, ask for the rate and the unit that rate counts, then measure how many units your own work takes. A pricing page that answers both can be recomputed by a buyer. A range quote cannot.