please contact [email protected]
Hello @sinovoip @Judy_Huang1 @Judy_Huang , I did sent email, twice in fact at Jul 23 and now at Nov 17, but i did not get any answer. I sent it from email registered here on forum, could You please verify that my email was not moved to the spam folder? Thanks
Hello @sinovoip @sinovoip1 @Judy_Huang1 @Judy_Huang,
it really is NOT POSSIBLE to work on and develop products when You are not able to communicate with us. We, developers, were invested our own time and money already, and still, we are loosing time because of You just do not communicate.
It is really too much hard to reply an email or reply here on forum thread or even PM? Just only other developers are communicating, but the lack of support from @sinovoip directly is just THE BIG STOP for me and my project.
My sfp+ NAS, is working already, but there is still problem with the “random restarts”, and I just decided to stop waiting, and rather find and order some other suitable non-bpi board instead, and adapt my custom made case. I am quite sure, that there will be enough space even for bigger board in my custom created case for 20x 3.5" hard drives.
I am really sorry, but needs to figure out other things to be able to start small-series production than just “the core”. Therefore I just get rid of my patience and trust to BPI company, and decided, to step back and try another solution for the core board.
With this post, I just wanted let know You and even other developers.
Have a nice day
… ![]()
→ Please do not go!
I’m sorry to hear about your situation, but this is precisely where BPI and our users should be striving. We hope to achieve better development of the product through continuous communication and efforts.
Could I ask how the better co-operation should work?
I would awaits at least 1 person who should be in touch with us and filter information related or important to @sinovoip.
As the account @sinovoip looks like a group / shared account for me, I was thinking, that @Judy_Huang @Judy_Huang1 is the right person. This does not looks like functional model obviously.
Now, I am little confused about You @hyf , as You are writing:
this is precisely where BPI and our users should be striving
Do You are able to help with it somehow?
Hello Banana Pi Community and SinoVoip Team,
I am an Embedded Systems Engineer and Hardware Innovator specializing in Edge AI deployment, firmware optimization, and hardware-software co-design. My work focuses heavily on bridging low-power microcontrollers with high-performance single-board computers to build real-time, local processing architectures.
As a developer, I am deeply committed to open-source hardware and maintaining clean, highly structured technical documentation that helps the community replicate complex engineering workflows.
I am applying for the Banana Pi BPI-M3 to serve as the high-throughput central compute core for an intelligent, dual-brain industrial safety initiative:
Project Omni Guard.
1. Project Overview: Omni Guard (Structural Health Monitoring)
Omni Guard is an open-source, dual-brain structural health monitoring system engineered for real-time vibration analysis, mechanical stress tracking, and predictive maintenance in industrial infrastructure. The Architecture: The system utilizes a multi-node edge topology. Low-power microcontrollers handle high-frequency, real-time data collection from multi-axis inertial measurement units (IMU arrays) and structural strain gauges.
The Role of the BPI-M3: The Banana Pi BPI-M3 will act as the Central Compute & Aggregation Gateway. It will ingest parallel data streams from the edge nodes, execute local mathematical models for real-time frequency analysis (FFT), and run optimized, quantized machine learning inference pipelines to detect structural anomalies locally without cloud dependence.
Hardware Realignment: The BPI-M3’s Allwinner A83T octa-core processor provides the parallel multi-threading performance necessary to manage multiple sensor data streams simultaneously, while its onboard storage options ensure low-latency data logging.
2. Relevant Technical Stack & Experience I maintain a rigorous, professional approach to hardware design and software optimization:
Firmware & Software: Highly proficient in C++ (Arduino/ESP-IDF framework) and Python for writing bare-metal routines, sensor communication drivers, and low-level memory-optimized scripts.
Edge AI & Processing: Experienced in taking machine learning models, quantizing them for resource-constrained hardware, and deploying them locally for real-time sensor analytics.
System Integration: Adept at managing hardware-level communications (I2C, SPI, UART, and high-speed serial protocols) to bridge microcontrollers with high-performance single-board computers.
3. Concrete Deliverables for the Banana Pi Community If selected for this giveaway, I am committed to delivering high-value technical assets back to the open-source hardware ecosystem:
Industrial Implementation Tutorial: I will author a comprehensive, step-by-step guide on the Banana Pi BBS/Forum detailing how to interface the BPI-M3 with low-power microcontroller nodes for industrial data aggregation, complete with clean wiring diagrams and core configuration steps.
Open-Source Repositories: All driver configurations, optimized multi-threaded data logging scripts, and local processing pipelines developed for the BPI-M3 during Project Omni Guard will be open-sourced on GitHub. Hardware Performance Analysis: I will publish a thorough technical review analyzing the board’s processing throughput, multi-core efficiency, and thermal stability under continuous, real-time industrial monitoring loads.
Thank you for your time, consideration, and dedication to supporting independent hardware engineering and open-source innovation. I look forward to showcasing the capabilities of the BPI-M3 in a robust, real-world application.
Best regards, Dipesh Kachhi Embedded Systems Engineer & Hardware Innovator
I’d like to participate in this as well. I am backend Golang engineer specialising in writing Golang, Kubernetes, Docker and Linux code. In my spare time I like to tinker with routers and embedded stuff hence why I am here.
May you could do a raffle and select at random everyone that has shown interest in this thread.
If I get a unit, I will blog about at https://blog.bana.io and also possibly make a very detailed video in English.
Regards,
- Bana
Hi BPI team,
I’ve been working with BPI-R4 and BPI-R4 Pro 8X hardware for some time now, and I’d like to offer something rather than ask for anything.
I’m a Linux & embedded systems developer with a focus on OpenWrt, kernel networking, and WiFi 7. Here’s what we’ve built and published on your forum recently:
- BPI-R4 OpenWrt Deploy System — 4GB/8GB, SD / eMMC / NVMe
- BPI-R4 OpenWrt — UniFi Protect (NVR) + Network Application on NVMe
- Luci-app-wifimgr v2.0.0 — WiFi 7 / MLO manager for BPI-R4 (MT7988A)
- BPI-R4 Pro 8X — Deploy System with UniFi Protect & Network (first release)
On the kernel side, I’m an active netdev contributor with patches in review and one already accepted into net-next:
What I’m offering: if you have hardware that needs open-source SW support — whether it’s OpenWrt integration, kernel drivers, or tooling — I’m happy to collaborate with your dev team. No strings attached, just good open-source work that benefits the community.
If that sounds interesting to you, I’m open to a conversation.
Best regards,
Petr Wozniak
Hello BPI team,
I’m Lexi McVay, founder of Argus Orbital, Inc., a small hardware startup in San Francisco. I’d like to apply for a BPI-R4 (and, if possible, the BPI-R4-NIC-BE14 WiFi 7 module).
The work I plan to do:
- Build a stationary WiFi 7 + 5G router on the BPI-R4 with mainline OpenWrt, a Quectel RM520N-GL 5G modem, and the BE14 WiFi 7 NIC.
- Document the full build on this forum: bring-up, the 5G modem setup (ModemManager / QMI packages, APN config), 6GHz WPA3/SAE setup, WireGuard and OpenVPN running natively on the router, and throughput and thermal results.
- Publish a step-by-step guide and a video of the build, so other users can replicate it.
This would replace a locked carrier gateway as my main internet connection, so it will get real, daily, long-term use, and I will post follow-up results.
Thank you for considering it! Lexi
It looks great. You could start a separate forum; I’m sure it would get a good response.
Hello BPI team,
I’d like to apply for one BPI-M7 (RK3588).
What I’ll do with it: run a complete offline AI assistant on one board and push the RK3588 to its limits.
B.A.I.S. is our private AI system (UK). It already runs a house and a business every day on a CPU-only mini PC: voice, memory, local language models, vision and Home Assistant control, with no cloud. I want to move the whole stack onto the M7 and publish exactly where it holds up and where it breaks:
- LLM on the NPU: 1.5B to 8B models with RKLLM vs llama.cpp on the A76/A55 cores (tokens/s, time to first token, RAM)
- Speech: faster-whisper / whisper.cpp speech recognition and Piper text-to-speech (real-time factors)
- Vision on the NPU at the same time: person/face/object detection on camera streams with RKNN
- Whole-house voice: several room satellites (Android TV boxes, ESP32 mics) talking to the one M7: how many simultaneous conversations before latency hurts
- Full load test: LLM + speech + vision + Home Assistant together; thermals with and without cooling, throttling curve, power draw at the wall
- Hardware I/O: NVMe vs eMMC model loading, 2.5GbE for multi-room audio, HDMI as a local dashboard
Everything will be published openly: a step-by-step image/guide, scripts and configs on GitHub, benchmark tables and a video, on our site (www.baisukltd.co.uk) and here on the forum, crediting Banana Pi.
Boards I work with today: Intel mini PCs, RK3528 and Amlogic Android TV boxes as voice satellites, and an Orange Pi 5 Plus (RK3588) arriving for review, so I can also publish a fair RK3588 board-to-board comparison.
Thank you, Richie Ellis, B.A.I.S.
Ok ,judy will contact you.
Hello Banana Pi team,
I would like to apply for one BPI-M7 with 32 GB RAM for an open-source, offline USB protocol observatory. I own an original EINSTAR handheld scanner and have started EINSTAR USB Observatory, which currently includes a Python tool for summarizing USB trace metadata and tests using synthetic records. I have not captured real scanner traffic or validated this workflow on Banana Pi hardware yet.
My plan is to bring up the BPI-M7 as a dedicated Linux capture and analysis host, check USB 2 enumeration and trace collection using a device I own, and compare sanitized trace summaries with captures from my existing PC. I would publish exact setup steps, code changes, reproducible synthetic examples, and measured CPU, memory, power, and sustained-capture behavior. If practical, I would also test offline small-model summarization of sanitized metadata and report model size, latency, memory use, and failures rather than claiming NPU support in advance. No vendor binaries, raw private captures, or identifying device data would be published.
I will share an honest board bring-up and project results on this forum and in the public repository. Please confirm whether the 32 GB variant is available for this program, whether power/cooling accessories are included, and whether the board can be shipped to Virginia, USA with no cost to me for the board, shipping, taxes, or duties, and that I may keep the board without a required return. I would share my mailing address only in a private message after those terms are clear.
Thank you, Bradley Grubbs
Hi - I’m lead developer of Frugal IoT which is a fully open-source project to make IoT solutions affordable in highly resource constrained environments. Its a full stack from hardware designs; firmware; server; client. One of the elements is a server that can run locally when the backbone to the internet is down, and typically runs on a Raspberry Pi. Some of the people in the Community Networks / Remote Internet access space are experimenting with bring services to newly connected remote communities, including IoT and also are experimenting with the Banana Pi, and have asked me if the Frugal-IoT server could run on the Banana Pi. I really don’t know - its Linux based - but would have to work within the constrained OpenWRT environment, which is probably do-able, but hard without anything to test on. Hence this request.
I see you have a bunch of router products, its the Open WRT One that I’m getting requests for support for, and would like to get a sample to test.
A lot of the applications I imagine being used for would be as a GSM->WiFi router, and notice it doesn’t have a SIM card, is that an add-on, or something planned for a later version? I notice you also have BPI-R3 that has a SIM slot, but if I read the web page correctly it doesn’t support OpenWRT? I’m not familiar with your Zero range, if I’m reading the website correctly, the BPI-P2 Zero, BPI-M2 Zero, BPI-M4 Zero all support Linux. I’ve tested the frugal-iot-server against some of the Raspberry PI range, with success on the RPI4. The Pi-Zero-W, sometimes works, but the installer fails more than 50% of the time for no detectable reason so we don’t recommend it. And we don’t have a Pi-Zero-2W to see if that is reliable. It would be interesting to see which of your lower end boards were viable servers, though I think the combination with the router in OpenWRT-One makes more sense if that works.
Generally the node code of Frugal IoT works on any ESP32 variant, so I have no reason to think it wouldn’t work on your ESP32-S2 and ESP32-S3 boards, but if you wanted to send me any at the same time, I’ll check them and add them to the supported list on the Wiki.
Thanks - Mitra
Hi Banana Pi Team!
I’m Aditya Upadhye, a software engineer and recent MS Computer Science graduate from Stony Brook University. My experience includes building AI agents, real-time speech-processing systems, and LLM-powered applications.
A few projects that reflect my work:
- Master’s Thesis: Agentic UI Orchestration Agentic UI Orchestration: A Dynamic LLM-Driven Framework for Accessible Desktop Task Automation - ProQuest - Built an autonomous LLM-driven desktop agent achieving 89% task completion across 214 tasks.
- Hack for Humanity: Miami 2026 Open Source Track winner for AidNow https://youtube.com/shorts/vf8QP_vxo7Q, an offline multilingual emergency copilot.
- eMerge AI Hackathon 2026: 3rd Place, Archie Track, for Find My School https://youtu.be/9q-oK6NEY5A, an AI-powered school navigation assistant.
GitHub: aditya-upadhye (Aditya Upadhye) · GitHub
Board requested: Banana Pi BPI-M7 (preferably 32GB RAM, but flexible).
I’m interested in exploring a practical question: How well can modern AI models actually run on affordable edge hardware?
I’d like to develop an open-source project, EdgeProof, to evaluate small language models, speech recognition, and computer vision workloads on the BPI-M7.
My focus would be on:
- Benchmarking inference speed, latency, memory usage, and thermal performance.
- Exploring CPU vs. NPU acceleration using RKNN/RKLLM.
- Demonstrating a practical, privacy-preserving offline AI application.
I’ll publish the source code, reproducible benchmarks, and setup guides on GitHub. I’ll also write a detailed technical post on the Banana Pi community forum, sharing my experiments, performance results, challenges, and practical recommendations for developers exploring on-device AI.
The BPI-M7’s RK3588 platform is particularly interesting to me because I’d like to investigate the gap between advertised NPU capabilities and real-world AI performance.
I’d be excited to contribute useful, independently tested resources to the Banana Pi ecosystem!
Thank you! Aditya.
Good day!
Thank you for your support!
This is Klaus , from Banana Pi. You can contact me : [email protected]
I will follow up on this matter.
@sinovoip @aditya
Thank you, Klaus! I appreciate your response and willingness to follow up. I’ve reached out to you via email regarding my BPI-M7 developer board request. Looking forward to hearing from you!