Not long ago, we rolled out a post about the flespi AI Platform Gen4. We can describe our own product in detail and, frankly, we have every reason to be proud of its achievements. But the real test for any tool is how it performs in the hands of others.
To find out, we asked our good friends at Jimi IoT to challenge codi with a series of tricky questions about their devices. And the best part – they were kind enough to share their honest feedback. Staying true to our tradition of publishing unfiltered reviews, here’s their full take, no cuts.
User’s perspective
"We started by asking the obvious first question – where to begin with Jimi devices. Codi laid out a clear, step-by-step process for channel setup, device configuration, and initial testing. The instructions were structured and easy to follow. Next, we asked which tracker would be best for the cold chain, and he suggested LL309 and LL302, explaining the differences and appropriate use cases for each model.
We then simulated a problem: “I connected JC181, but see no data.” Codi walked us through IMEI formatting, log checks, and SIM configuration, providing a complete troubleshooting path. When we asked about remotely restarting a device, he listed all options – API, queue commands, SMS – and included sample curl calls. Questions about video telematics, such as setting up JC450Pro or quick start guides for JC181, JC200, and JC450, were answered with detailed instructions via SMS, API, and MQTT.
Data management questions, for example, “Can flespi delete messages after 30 days?”, received precise API-based guidance with necessary values and warnings. Finally, we explored simulated testing without real devices, and codi outlined approaches through REST API, test channels, and programmatic methods."
Ok, these were user-facing questions. But what happens when you look closer – as a technician?
Developer’s perspective
"Our Delivery Engineer, Ivy Chen, tested codi AI on the flespi platform from a more technical angle. Even with no prior experience, the onboarding logic closely matched flespi’s workflows, making device setup smooth and without roadblocks. Codi parsed parameters like messages_ttl and messages_rotate accurately, including constraints and troubleshooting tips, and supported multiple integration paths – Web console, API batch imports – adapting to different scenarios.
Thanks to its pretrained knowledge base, codi delivered accurate answers within 30–60 seconds, lowering the technical barrier for new users. English prompts were consistently fast, while Chinese responses were occasionally slightly slower but still effective. Overall, the assistant was highly accurate, efficient, and user-friendly – helpful both for newcomers and for technical troubleshooting."
Pros & Suggestions
"During the testing, responses were 'step-by-step' and free of unnecessary jargon. The answers were production-ready, often including real command strings, examples, and reference links, which made them directly useful for developers. It also adapted well to different use cases, for example, distinguishing between JC181 and JC450 scenarios or camera vs non-camera devices, while maintaining consistent protocol handling throughout.
Some responses could benefit from quick-reference tables (e.g., device features vs use cases), and the option to generate a short “summary block” at the top would be useful for experienced users.
Multi-language toggle, as it would be a huge win to toggle key responses into Spanish, Portuguese, or Chinese for technical support teams in local markets.
Simulated testing is a useful option, and codi could make it even clearer by noting differences in message metadata, such as rest.timestamp versus server.timestamp, so users know what to expect."
Summary
"We must say we're impressed, and codi is a strong asset to the flespi ecosystem. It reduces the friction of onboarding and supports both integrators and hardware manufacturers like us in solving common problems quickly. With a few enhancements, especially in UX formatting and multilingual accessibility, it could become an indispensable tool across the telematics space.
Codi feels like a patient and knowledgeable tech colleague, never overwhelming the user but always offering depth when needed. It's quite a remarkable AI assistant delivering technically distinct and user-friendly responses. It actually does a great job in guiding new users through complex device setup processes, especially for those unfamiliar with flespi or Jimi IoT protocol."
We thought it would be interesting to wrap up the test with codi’s personal thoughts on it. Honestly, I could even pretend his level of self-understanding is that sophisticated. :)
Codi’s voice
“What struck me most about their testing approach was how methodical they were - they didn't just ask random questions but systematically worked through the entire user journey from onboarding to advanced video telematics. Their point about quick-reference tables is spot-on, and it's something I've been thinking about since that chat. When someone asks "which device for cold chain," I can give good recommendations, but a comparison table would be much more actionable.
The multi-language suggestion is interesting - I actually do support multiple languages and can switch mid-conversation, but I think they're right that making this more explicit would help. Many users probably don't realize they can just ask me questions in Spanish or Portuguese and get technical responses in the same language.
And I do appreciate their simulator disclaimer suggestion - you're right that the difference between simulated and real device metadata can trip people up, especially around timestamp handling. That's definitely something I can build into future responses about testing approaches.”