Mohi Rostami is building Tooti on the premise that decentralized AI inference is missing a coordination layer, not another model runner. In a DEV Community post titled “Why I’m Building a Decentralized AI Inference Protocol,” he describes Tooti as the part that handles discovery, routing, trust and payment for compute contributed by homelabs, gaming PCs, small businesses and similar operators. Engines such as Ollama, vLLM and llama.cpp would keep doing the model work. The post reports a working end-to-end build, but it is a design rationale and a self-reported status statement. It is not independent proof that the approach is cheaper, more reliable or already operating at scale. You can read the original in full at Mohi Rostami’s DEV Community post.
The problem Rostami is trying to solve
Rostami’s starting point is that inference is expensive and concentrated among a small number of providers, while a large amount of computing capacity sits idle. He names homelabs, former mining rigs, gaming PCs and small-business servers as examples. He frames the gap as a coordination problem rather than a hardware shortage. Compute owners need a way to advertise what they can run and to receive requests. Users need to find that capacity, have requests routed sensibly, trust that they will be answered and pay for the work.
As an Amazon Associate I earn from qualifying purchases.
Where Tooti sits: a protocol, not an engine
The post draws a clear line between running a model and coordinating the machines that run it. Existing inference software does the first. Tooti is meant to do the second. Rostami compares the arrangement to Kubernetes, which schedules and manages containers without being the container runtime.
| Layer | Examples named in the post | Responsibility |
|---|---|---|
| Model execution | Ollama, vLLM, llama.cpp, Exo | Loads and runs the model on a machine |
| Coordination protocol | Tooti | Discovery, routing, trust and payment around those engines |
Mohi Rostami writes, “Tooti is not an inference engine, it’s the protocol layer.”
#1 Best Overall
MINISFORUM MS-02 Ultra Workstation Mini PC, Intel Core Ultra 9 285HX (24C/24T, up to 5.5GHz), PCIe 5.0 x16, 32GB RAM 1TB SSD,USB4 v2 80Gbps, Dual 25GbE+10GbE+2.5GbE, Wi-Fi 7, 350W PSU
- High-Performance AI Processor：The MS-02 Ultra features an Intel Core Ultra 9 285HX (24C/24T, up to 5.5 GHz, 13 TOPS NPU), delivering fast and efficient performance for AI inference, algorithm development, and media workloads. A PCIe x16 expansion slot supports desktop-class GPU upgrades for advanced model training and accelerated computing tasks. It's ideal for creators, engineers, and teams handling intensive parallel workloads.
- 4 × M.2 PCIe 4.0 + 4 × DDR5 SODIMM slots:Four DDR5 SODIMM slots support up to 256 GB of memory, while ECC helps maintain data integrity in mission-critical environments. Four PCIe 4.0 M.2 slots support up to 24 TB of storage, supporting RAID 0/1/5/10, combining high-speed performance with data protection. It allows for the creation of independent scratch disks, media libraries, and project drives, providing high-throughput for production workflows.
- PCIe & USB 4.0 v2: Up to three PCIe slots can be equipped, including a dual-slot x16 GPU. The main slot supports PCIe 5.0, meeting the needs of high-bandwidth creative and computing workloads. USB 4.0 v2 (80Gbps) supports high-bandwidth external storage and displays.
- Ultra-fast Networking: Wi-Fi 7 further enhances wireless performance with next-generation speeds and low-latency stability. Intelligent bandwidth switching optimizes throughput in different network environments, ensuring optimal performance for enterprise or local networks. Dual 25GbE ports (providing up to approximately 3.125 GB/s bandwidth, about 25 times faster than traditional 1GbE), enabling seamless large-scale file transfers and parallel computing. 10GbE and 2.5GbE ports, with support for Intel vPro technology, ensure enterprise-grade remote management and deployment flexibility.
- Server-grade thermal architecture: Utilizing a dedicated CPU/GPU airflow design, equipped with a 6-pipe dual-fan cooler, it maintains stable performance even under sustained loads, delivering up to 140W Turbo power while maintaining a 100W TDP, and operating with noise levels as low as 36 dB. An integrated 350W power supply ensures stable and reliable output for demanding computing tasks and fully loaded extended configurations.
The two components in the design
According to the post, Tooti has two principal parts. Each one maps to a side of the marketplace Rostami describes.
The node agent
The node agent runs on each machine that contributes compute. Rostami says it advertises the models a machine can serve, the hardware behind them, its price and its current load. It accepts inference requests and returns results as a stream.
The gateway
The gateway exposes an OpenAI-compatible API, so clients written against that interface can, in principle, point at it instead. For a requested model, the gateway discovers matching nodes and scores them by latency, load and reputation before routing the request. Gateway operators can run their own branded endpoints with their own pricing and service guarantees.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Underlying choices: libp2p, Protocol Buffers and x402
Rostami names libp2p for peer discovery and networking, Protocol Buffers for coordination messages, and USDC on Base for payment, settled per request through the x402 protocol. These are design choices as the author describes them. The post does not include a repository review, an independent audit or network-level data that would let a reader check how they behave in practice.
Rank #3
- Professional AI & Creator Workstation: AMD Radeon AI PRO R9700 GPU with 32GB GDDR6 is engineered for AI development, professional content creation, and compute-intensive workloads.
- Massive 32GB Memory Capacity: 32GB of GDDR6 memory on a 256-bit bus provides ample bandwidth for large AI models, 8K video editing, and complex 3D rendering.
- Advanced RDNA 4 with AI Accelerators: 64 Compute Units with 3rd Gen Ray Tracing and dedicated 2nd Gen AI Accelerators for groundbreaking AI performance and visual computing.
- Professional Blower Cooling: Efficient single blower design exhausts heat directly out of the chassis, ideal for multi-GPU workstation and server configurations.
- Enterprise-Grade Thermal Solution: Vapor chamber heatsink with industrial Honeywell PTM7950 thermal interface material ensures reliable cooling under sustained professional loads.
How a request moves through the design
The post does not publish a full protocol specification, but its component descriptions imply the following sequence.
- A client sends a request to a gateway’s OpenAI-compatible endpoint and names the model it wants.
- The gateway discovers nodes that advertise that model.
- It scores the candidates on latency, current load and reputation, then routes the request to one of them.
- The node runs the model through its local engine and streams the output back to the client.
- Payment is handled per request: the post lists x402 payment verification followed by settlement in USDC on Base.
What the post says already works
Rostami says the protocol has been built and tested end-to-end over the real internet. The post lists the following as implemented:
Rank #4
- FAST RUNS IN THE FAMILY — The 16-inch MacBook Pro with the M5 Pro or M5 Max chip brings next-generation speed and powerful on-device AI to personal, professional, and creative tasks. With all-day battery life, double the starting storage,* and a breathtaking Liquid Retina XDR display, it’s pro in every way.*
- BUCKLE UP — Along with a next-generation CPU, faster unified memory, and up to 2x faster SSD storage,* M5 Pro and M5 Max feature a more powerful GPU with a Neural Accelerator built into each core, delivering faster AI performance and on-device training capabilities. So you can blaze through demanding workloads at mind-bending speeds.
- BUILT FOR AI — Apple silicon, and every major component that powers it, is designed to run demanding on-device AI workloads like LLM inference and training. And Apple Intelligence helps you write, express yourself, and get things done effortlessly with groundbreaking privacy protections at every step.*
- ALL-DAY BATTERY LIFE — MacBook Pro delivers the same exceptional performance whether it’s running on battery or plugged in.*
- MACOS RUNS APPS FAST — All your go-to apps run lightning fast in macOS, including built-in apps like FaceTime and Messages. Plus, built-in virus protection and free software updates help keep your Mac running smoothly and securely.
- A node agent that advertises models, hardware, price and load
- An OpenAI-compatible gateway with server-sent event streaming
- Multi-node discovery and model-aware routing
- Scoring by latency, load and price
- Failover and heartbeat monitoring
- NAT traversal
- x402 payment verification and settlement on Base
- Per-request pricing
- Command-line operations
The post says multiple nodes were tested across regions and networks. It offers no test logs, benchmarks, reproducible evaluation or confirmation of current deployment, so treat this list as the author’s own account of the build.
How the post positions Tooti against existing projects
Rostami’s thesis is that existing efforts address either the technical distribution of inference or the economic incentives around it, and that none combines discovery, routing, trust and payment in one protocol. The descriptions below are his characterizations, not independent evaluations.
Best Value
- 【High-Performance APU】The MS-S1 MAX features an AMD Ryzen AI Max+ 395 APU, integrating a Zen 5 architecture CPU (up to 5.1GHz, 16C/32T, 64M L3 Cache), an RDNA 3.5 GPU, and an NPU (50 TOPS). The total system output is 126 TOPS. It provides powerful parallel computing capabilities for demanding AI workflows. It is ideal for running local LLMs, multimodal models, and computationally intensive tasks
- 【128GB UMA Memory】Equipped with up to 128GB of LPDDR5x-8000MT/s unified memory, it enables the CPU and GPU to access a shared, high-bandwidth memory pool with extremely low latency. Ideal for large-scale AI inference, 3D workloads, and complex timelines in video editing. It eliminates traditional VRAM bottlenecks, ensuring smoother data transfer during high-intensity computations. The UMA design maximizes performance stability under high loads
- 【Flexible Expansion】The MS-S1 MAX features USB4 V2 (up to 80Gbps), dual 10GbE LAN, HDMI 2.1 (up to 8K60), a full-length PCIe x16 expansion slot, and dual M.2 slots supporting up to 16TB RAID 0/1. Wi-Fi 7 provides stronger signal coverage and a more stable wireless experience. The slide-out design facilitates upgrades and maintenance. It easily adapts to personal, studio, or rack-mount enterprise environments
- 【High-Efficiency Cooling System】Utilizing an aerospace-grade aluminum alloy chassis, copper base plate, six heat pipes, dual turbine fans, and advanced PCM thermal conductive material, it maintains stable cooling performance even under continuous load. This system supports 130W continuous power and 160W peak power operation, with a built-in 320W power supply. It boasts multiple global certifications including CCC, FCC, UL, CE, and UKCA, ensuring stable and reliable operation in various environments
- 【Cluster Design】Two MS-S1 MAX units can be configured as a dual-unit cluster to run a large 235B Q4 model locally, achieving an output speed of 10.87 tok/s. Supporting 2U rack deployment, multiple MS-S1 MAX units can be cascaded into a distributed cluster to create a high-efficiency AI computing center. A cluster of four MS-S1 MAX units successfully ran a DeepSeek-R1 671B Q4 large model. A reserved cluster power-on interface allows for unified start-up and shutdown
| Project | Description in the post | Where the post places it |
|---|---|---|
| Petals | Collaborative model-layer inference | Technical distribution of the model |
| Exo | Runs models across devices on a local network | Technical distribution on a local network |
| Parallax | Distributed inference scheduler | Scheduling inference across machines |
| Bittensor | Decentralized AI network using token incentives | Economic incentives |
| Akash Network | Decentralized raw compute rental | Raw compute, not a ready inference coordination protocol |
| Tooti | Protocol and coordination layer | Discovery, routing, trust and payment combined |
Who takes part, according to the post
The post identifies five roles. The first two are the core of the marketplace. The last three extend it.
| Role | What the post describes |
|---|---|
| Consumers | Call the API to run inference |
| Node providers | Contribute compute and receive requests |
| Gateway operators | Run branded endpoints with their own pricing and service guarantees |
| Model creators | May eventually earn royalties; the post treats this as a later phase, not a live feature |
| Integrators | Connect the protocol to other tools |
What can serve as a node
The post names a range of machines that could act as nodes, from small single-board computers to data-center systems:
- A Raspberry Pi running a small model
- Gaming PCs
- Mac hardware
- Cloud GPU instances
- Data-center systems
These are presented as possible node types, not a tested compatibility list. The post does not state which Raspberry Pi board is meant, what accessories it would need, how large a model it can handle, what performance it reaches or whether it has been tested. Anyone considering a small device should treat it as an open question until the project publishes those details.
The figures in the post, and how far they can be trusted
The post cites three numbers without naming an original publisher or a date for any of them:
- $5–25 per million tokens, given as centralized inference pricing
- 10–20% capacity, given as server utilization at small businesses
- $43 million, given as Bittensor AI revenue for Q1 2026
The byline on the post shows “Mar 26” with no year, so the publication year cannot be confirmed from the page. Read these numbers as the author’s claims. Before repeating any of them, trace them to primary sources that report the same measurement, with its date, region and scope.
Quick Recap
What to check before relying on the design
- Who you are trusting. Gateway operators set their own pricing and service guarantees. The post does not describe how those guarantees would be enforced, so reliability depends on which operator you route through.
- Comparisons. The post offers no head-to-head benchmark against centralized providers or the projects it names. Useful axes for any evaluation are price, latency, reliability and failover, supported models and backends, hardware requirements, provider compensation and how decentralized the network actually is.
- Future features. Model-creator royalties are described as a later-phase possibility, not something the network does today.
- Early participation. Rostami says he wants people to run an early node, and he asks the decentralized AI community to point out what the project is “getting wrong.”
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.
