Recommended Free Tools
iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Samuel James Hiotis says he built a 34-agent AI swarm on an Android phone by defining many lightweight agents but running only a few at once. The reported design combines Termux, Python, local and hosted inference, Redis Streams for agent messaging, and an orchestrator that splits tasks and gathers results. It is an account of one build, not a verified recipe: the article does not identify the phone or provide enough setup detail to reproduce its performance reliably.
What “34 agents” means in this build
The number refers to agents Hiotis says he defined, not 34 models doing inference simultaneously. In his description, each agent is a lightweight wrapper around an inference call. An orchestrator chooses which agents to involve for a task, limits how many are active, and passes their results to a synthesizer.
That distinction is central to the design. The system’s value comes from assigning different parts of a task to specialized roles, while keeping concurrent work bounded to suit a phone’s memory and heat limits. It does not mean the phone can run 34 large language models at once.
How the reported workflow fits together
- Decompose the request. An orchestrator turns the incoming task into smaller subtasks that can be assigned to agents.
- Dispatch bounded work. It sends subtasks to selected agents through a shared message bus, with a semaphore limiting concurrent activity. Hiotis says this guard is important to avoid exhausting the phone’s memory.
- Run agent inference. Each agent wraps an inference call. The account describes small quantized models running locally, with requests able to be escalated to hosted models for final synthesis.
- Collect and synthesize. The orchestrator gathers agent outputs and sends them to a synthesizer, which combines them into a response.
- Review and format when needed. In the article’s RISC-V research-brief demo, three researcher agents handled subtasks; a synthesizer, reviewer, and formatter then produced the brief. The reviewer is one stage in that reported workflow, not independent proof that the result was accurate.
The author describes using Redis Streams as the message bus. In practical terms, that gives components a shared way to pass task messages and results without requiring every agent to communicate directly with every other agent. The article also mentions Uvicorn and a simple Flask interface, alongside Android, Termux, Python 3.11, and Ollama.
#1 Best Overall
- Universal unlocked. Compatible with all major U.S. carriers, including Verizon, AT&T, T-Mobile and other prepaid carriers.
- Super-bright, super-smooth 6.7" display. See your screen clearly even outdoors in sunlight, and enjoy seamless views with a fast-refreshing 120Hz display.*
- AI-powered camera system. Take stunning photos in any light with the 50MP camera**, look your best with a 32MP selfie cam*****, and capture extreme close-ups.
- Superfast 5G performance. Unleash your entertainment at 5G speed*** with the MediaTek Dimensity 6300 chipset and up to 12GB of RAM with RAM Boost****.
- Long-lasting battery + TurboPower charging. Power through day after day with a 5200mAh battery, then get hours of power in just minutes.****
What you would need to adapt the idea
The source describes the architecture and names its software stack, but it does not provide a repository or sufficiently detailed setup instructions to establish a reproducible phone build. Treat the following as the design decisions to work through, not as a tested installation procedure.
Choose a small initial set of roles
Hiotis recommends starting with three agents and expanding only after task routing works. A small trial makes it easier to see whether decomposition is useful and whether the orchestrator is assigning the right work. Add roles only when they have distinct jobs; more defined roles do not automatically improve an answer.
Rank #2
- YOUR CONTENT, SUPER SMOOTH: The ultra-clear 6.7" FHD+ Super AMOLED display of Galaxy A17 5G helps bring your content to life, whether you're scrolling through recipes or video chatting with loved ones.¹
- LIVE FAST. CHARGE FASTER: Focus more on the moment and less on your battery percentage with Galaxy A17 5G. Super Fast Charging powers up your battery so you can get back to life sooner.²
- MEMORIES MADE PICTURE PERFECT: Capture every angle in stunning clarity, from wide family photos to close-ups of friends, with the triple-lens camera on Galaxy A17 5G.
- NEED MORE STORAGE? WE HAVE YOU COVERED: With an improved 2TB of expandable storage, Galaxy A17 5G makes it easy to keep cherished photos, videos and important files readily accessible whenever you need them.³
- BUILT TO LAST: With an improved IP54 rating, Galaxy A17 5G is even more durable than before.⁴ It’s built to resist splashes and dust and comes with a stronger yet slimmer Gorilla Glass Victus front and Glass Fiber Reinforced Polymer back.
Decide what runs locally and what can be hosted
The reported setup uses quantized local models for some inference and allows escalation to hosted models for final synthesis. That is a hybrid design, not an entirely offline swarm. Hosted inference introduces a network dependency and latency; the source does not specify the model names, configurations, or escalation rules needed to reproduce its behavior.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Set a concurrency limit before scaling
Keep the semaphore in the orchestration path so the number of simultaneous inference calls stays bounded. Hiotis’s account links this limit to phone memory pressure. The right limit for another device cannot be inferred from his figures because the phone model, operating-system build, and model configurations are not identified.
Rank #3
- Charger NOT Included, 6.7" Super AMOLED FHD+, 90Hz Refresh Rate, 385 ppi, 800 nits (HBM), 1080x2340px, 5000mAh Battery
- 128GB, 4GB RAM, microSDXC, Exynos 1330 (5nm), Octa-Core, Mali-G68 MP2 or Mali-G57 MC2 GPU
- Rear Camera: 50MP, f/1.8 (wide) + 5MP, f/2.2 (ultrawide) + 2MP, f/2.4 (macro), LED flash, panorama, HDR; Front Camera: 13MP, f/2.0, Android 14, up to 6 major Android upgrades, One UI 6.1
- 3G: HSDPA 850/900/1700(AWS)/1900/2100; 4G LTE: 1/2/3/4/5/7/12/13/14/20/25/26/28/29/30/38/39/40/41/48/66/71, 5G: 2/5/25/41/66/71/77/78 SA/NSA/Sub6/mmWave - Nano-SIM + eSIM
- US Model – Global Connectivity – Compatible with Most GSM Carriers like T-Mobile, AT&T, MetroPCS, etc. Will Also work with CDMA Carriers Such as Verizon, Straight Talk.
Plan for idle work and failures
The author reports using idle timeouts, batching, health checks, and automatic restarts. Those choices address different operational problems: idle timeouts reduce unnecessary activity, while health checks and restarts help recover from process failures. They do not guarantee that a workload will stay within a device’s thermal or battery limits.
Reported figures—and what they do and do not establish
These numbers are Hiotis’s own figures in his September 30, 2025 DEV Community account. They have not been independently validated in the available source.
Rank #4
- YOUR CONTENT, SUPER SMOOTH: The ultra-clear 6.7" FHD+ Super AMOLED display of Galaxy A17 5G helps bring your content to life, whether you're scrolling through recipes or video chatting with loved ones.¹
- LIVE FAST. CHARGE FASTER: Focus more on the moment and less on your battery percentage with Galaxy A17 5G. Super Fast Charging powers up your battery so you can get back to life sooner.²
- MEMORIES MADE PICTURE PERFECT: Capture every angle in stunning clarity, from wide family photos to close-ups of friends, with the triple-lens camera on Galaxy A17 5G.
- NEED MORE STORAGE? WE HAVE YOU COVERED: With an improved 2TB of expandable storage, Galaxy A17 5G makes it easy to keep cherished photos, videos and important files readily accessible whenever you need them.³
- BUILT TO LAST: With an improved IP54 rating, Galaxy A17 5G is even more durable than before.⁴ It’s built to resist splashes and dust and comes with a stronger yet slimmer Gorilla Glass Victus front and Glass Fiber Reinforced Polymer back.
| Figure | What the author reports | Qualification |
|---|---|---|
| 34 agents | That many agents defined in the system. | The account also describes only a smaller subset active at once. |
| Four to six active agents | The author’s sparse-activation range. | Not a universal safe concurrency limit for Android phones. |
| About 200 MB per agent; 6.8 GB for 34 | The author’s context-memory estimate and multiplication. | The article does not provide measurements or explain enough about model and runtime memory to treat this as a general memory formula. |
| A 6 GB phone | A comparison device capacity mentioned by the author. | No make or model is identified. |
| About 1.2 GB | The reported APK-less footprint, attributed mostly to model weights. | The exact models and measurement method are not specified. |
| About 90 seconds | The reported time for a RISC-V research brief. | A demo result without an independent benchmark or stated test conditions. |
| 30 seconds | The reported inactivity period before agents sleep. | A setting in this account, not a recommended default for every workload. |
| Two to three crashes daily; throttling after about 10 minutes of heavy load | The author’s operational reports. | Not measured across devices or independently reproduced. |
| “90% quality at 30% size” | The author’s claim about Q4_K_M quantization. | No evaluation method or benchmark is supplied, so it should not be generalized as a model-quality result. |
Practical limits to expect on a phone
Hiotis identifies memory use, network latency, battery draw, process crashes, and thermal throttling as problems encountered in his build. He reports using sparse activation, batching, idle timeouts, automatic restarts, health checks, and local quantized models to manage them. He also says active cooling helped with heat. These are implementation choices from his account, not tested recommendations for every device; a clip-on phone cooler is an optional category to investigate, not a requirement or a tested product recommendation.
The account does not specify the exact handset, OS build, energy measurements, heat readings, or model configurations. Its reported speed, memory, and reliability figures therefore cannot establish what another phone will achieve. Start with a small workload, monitor the device, and reduce concurrent activity if it becomes unstable or overheats.
Best Value
- Carrier: This phone is locked to Tracfone, which means this device can only be used on the Tracfone wireless network. Tracfone plan required, activating is easy, just 3 steps.
- DISPLAY: Immersive viewing on a 6.7-inch super-bright 120Hz display with powerful stereo speakers and Bass Boost for cinematic entertainment.
- CAMERA SYSTEM: Advanced 50MP Quad Pixel camera captures sharp, detailed photos and videos in any lighting condition
- PERFORMANCE: Lightning-fast 5G connectivity paired with a powerful processor and RAM Boost for smooth multitasking.
- BATTERY LIFE: Long-lasting 5000mAh battery with TurboPower charging technology delivers hours of power in minutes.
What the account demonstrates—and what it leaves open
Hiotis’s September 30, 2025 article is a useful example of a mobile agent architecture: separate task decomposition, bounded worker execution, message passing, and synthesis rather than treating “34 agents” as 34 simultaneous models. It reports a working demonstration and practical operating problems, but it does not provide the materials or controlled measurements needed to verify the build or predict results on a different phone.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
