flespi AI agents: three intelligence levels for different tasks

In our previous article, we introduced flespi AI agents and the idea behind them. Since then, we’ve added three intelligence levels, giving users more flexibility in how they delegate tasks. Now let’s take a closer look at what makes these autonomous workers tick, what tasks they are designed to handle, and how they can be used in practice.

More than just agents

We call flespi autonomous AI workers "agents" for simplicity. But compared to the flespi AI Assistant in our support chat, they have agency: they can actually do things. 

And I would argue that they are now more than just "agents" in the common understanding of the word. They were released less than three weeks ago, yet our users are already putting them to work extensively. Let's set aside the administrative tasks in flespi accounts –  getting device lists in a specific format, bulk device renaming or storage policy changes, moving a subset of items to a group or subaccount, finding problematic instances, or creating and testing sophisticated webhooks and calculators. These are the tasks agents were explicitly built for.

What’s especially interesting is the variety of tasks they are already being used for. Here are some of the use cases we’re seeing in practice:

• Health check. Not just “find out if there are any problems with my account” or “which devices have been offline for more than a week.” Users are also setting up active monitoring of critical account instances, with agents notifying responsible people in Telegram before a problem becomes critical. For example, an agent can monitor channel connection counts approaching their limits or detect when stream buffer size or delay grows beyond a defined threshold.
• Device configuration. Real investigations involving several thousand devices. Agents analyze flespi logs and, when needed, manufacturer-specific device logs. They aggregate statistics, identify the problem, and suggest the best way to resolve it. 
• Cross-system work. We already have native connectors to Wialon and Teltonika FOTA WEB (if you want your platform to be the next one to connect, we'd be glad to hear from you). Some users ask agents to analyze data across systems, enriching what flespi knows with information from a third-party platform, and vice versa. Others want to synchronize data between systems or use agents to assist with migration.
• Work with external data sources. Users upload a PDF or CSV file, ask the agent what to do with it, and get the result. 
• Business logic check. Some tasks can be solved with flespi calculators. For example, “find the devices that had a trip start signal more than three times during the previous hour.” No problem – our Analytics can handle that. But if it's a one-time task, and the result will be used and forgotten, why not delegate it to an agent?

Of course, not all tasks are created equal. Some are straightforward and require little context, while others demand broader knowledge, deeper analysis, and more preliminary investigation. That's why flespi AI agents now come with three intelligence levels. 

From junior intern to senior architect

Economy – a junior intern. You make the decisions, and the agent does the work. It's efficient at analysis and clearly defined tasks, but relies on you to provide direction and decide what matters.

Standard – an experienced freelancer. Takes initiative on the task it's given. It can investigate independently, interpret what it finds, and adapt its approach without needing every step spelled out.

Advanced – a senior architect. Sees the bigger picture, plans for the future, and spots what you might miss. It weighs options, identifies hidden dependencies, and gives advice when the task requires judgment rather than just execution.

How much does an AI worker cost?

It is not always easy to make sense of this pricing, because the concept of a step is rather vague and surprisingly broad. The pricing is in credits per step: 25, 100, and 250. One credit is one cent. Which leaves us with the cost per step of 0.25 €/step for Economy, 1 €/step for Standard and 2.5 €/step for Advanced. But keep in mind that in one step an autonomous worker can look through device documentation, find suitable technology description in flespi Knowledge base, investigate device history messages/telemetry/traffic, and equipped with built-in flespi knowledge perform a thoughtful internal reasoning process to solve the task. Knowing all that the pricing sounds reasonable, especially if we compare it to the time/wage of real human engineers doing the same job. But using flespi agent has some advantages:

• No human cons: no need to find/negotiate/discuss the scope/sign NDA, etc.

• Pay only for the time it actually works

• Security built-in: you can give full access to all the needed resources without the fear of leakage/revoking/access management, etc.

• Instantly change between intelligence levels for different tasks: craft the implementation plan with an architect, switch to a newbie to execute, and ask an experienced engineer to check after the newbie. I wish I could have such a toggle in my brain: life would have become much easier in certain aspects 🙂

Bonus for real telematics work (UPDATED: September 21, 2026)

Our business is telematics at its core. And there's a lot of real work there that can be automated.

We are developing flespi AI agents not just to boast about a new AI feature, but because we already see the real value they bring to our customers' business processes. This can give projects built on flespi a competitive advantage.

That's why we encourage our customers to use this feature by extending the number of AI credits included in every pricing plan. Here's how many credits each plan includes:

To put that into perspective: a well-defined task usually takes one step, so set about five steps a month aside for clarifications and coordination, and the rest is finished work. On the Free plan that means about 75 tasks for a junior intern, 15 for an experienced freelancer, or 3 for a senior architect every month. Start raises the architect's share to 7 tasks and Pro to 15, or 45 tasks for an experienced freelancer. Enterprise customers get close to 100 freelancer tasks or 35 architect ones, and Ultimate doubles that once more. At one cent per credit, that is from €20 to €200 worth of AI work included in your plan every month at no additional cost.