The monthly cost of a support agent on four setups

A platform subscription does not show the full cost of a specific agent. We compared four setups for the same small support task.

The agent reads a new ticket, finds three similar past tickets, drafts a reply, and adds tags.

The workload is 40 tickets per workday over 20 workdays: 800 tickets per month. Each ticket uses three model calls.

Each call uses approximately 3,000 input tokens and 400 output tokens. This gives approximately 10,000 tokens per ticket, or 8 million tokens per month.

The comparison uses vendor prices published on 23 August 2026. It is a snapshot of those prices, not a current quote.

Lindy — $99.99 per user per month

Lindy's pricing page groups credits by task type. Its support queue example is in the Deep work group, at 250 to 1,000 credits.

At the lower limit, the monthly requirement is:

250 credits × 20 workdays = 5,000 credits.

The Plus plan costs $29.99 per user per month and includes 3,000 credits. At 250 credits per day, that allowance lasts 12 workdays.

The Pro plan costs $99.99 per user per month and includes 15,000 credits. It covers the month at the lower daily estimate.

At the upper estimate of 1,000 credits per day, the Pro allowance lasts 15 workdays.

Lindy has a seven-day trial, but no free tier in this price snapshot.

Relevance AI — $234 a month

Relevance AI measures Actions and Vendor Credits separately. Actions count task runs. Vendor Credits pay for model usage at wholesale rates.

The pricing documentation lists Team at $234 per month with annual billing, or $349 with monthly billing. It includes 7,000 Actions and $70 in Vendor Credits.

The 800 tickets fit within the 7,000 Actions allowance.

For the model estimate, we use $0.30 per million input tokens and $1.20 per million output tokens. The workload costs approximately $3.30 in model usage.

That is below the $70 allowance. At this volume, the subscription determines the bill: $234 per month with annual billing.

CrewAI — free until the second working day

CrewAI's Basic tier includes 50 workflow executions per month. At 40 tickets per day, the workload reaches the limit during the second workday.

The next option in this price snapshot is Enterprise, with custom pricing. You must contact sales for a quote.

That may suit a larger deployment. However, the public price does not let us calculate a monthly total for this workload.

Self-hosted — $0 in model fees

The fourth setup runs Free LLM API and CircleChat on your own system.

Free LLM API is our router under the MIT license. The router does not supply provider keys. You get free API keys directly from providers.

The providers in this setup include Google AI Studio, Groq, Cerebras, NVIDIA NIM, Cloudflare Workers AI, and Mistral. No credit card is necessary for their free access in this snapshot.

You add the keys to the router dashboard. It exposes one OpenAI-compatible endpoint at http://localhost:3001/v1.

The router manages model selection, rate limits for each key, and failover when a provider reaches its quota. In August 2026, its README lists 29 providers, 251 model families, and 358 free endpoints.

CircleChat supplies the agent workspace. The self-hosted version uses custom:freeapi as its default agent provider.

Configure the connection with:

OPENAI_BASE_URL=http://localhost:3001/v1
OPENAI_API_KEY=<your unified FreeLLMAPI key>

For this workload, the combined free quotas cover approximately 8 million tokens per month. The estimated model fee is $0.

Other costs

No model fee does not mean no operating cost. These items are still necessary:

• Hardware: a small virtual server or a computer you own. The chat service uses approximately 1 GB of RAM. Both services run on a Raspberry Pi 4.
• Maintenance time: you operate two services, PostgreSQL, and Redis.
• An optional catalog service: Free LLM API offers a signed live catalog feed for $19 per year. The router works without it.

The free quotas also have limits. They do not include frontier models or a service-level agreement. Response times can vary.

As stronger models reach their daily limits, the remaining options can be less capable. The daily quotas in this setup reset at midnight UTC.

These limits can be acceptable for a prototype. A separate assessment is necessary for a task where an incorrect answer has a high cost.

The $0 model estimate applies only when you self-host both services. CircleChat cloud plans include a gateway with fair-use limits and have different costs.

What the price covers

For this workload, the options are $99.99, $234, a custom quote, or the operating cost of your own server.

Managed platforms also give you infrastructure and support. Those services can justify the subscription for teams that do not want to operate their own system.

The comparison separates those costs from model usage. At 800 tickets per month, model tokens account for a small part of the managed subscription.

To test the self-hosted option, install the router and connect your provider keys.