# A Request: From Start to Finish Source: https://wp.farlabs.ai/a-request-from-start-to-finish The journey of an inference request through the FAR AI network is designed to be fast, transparent, and verifiable at every step. * A developer submits a chat or completion request to the FAR AI API in standard format. * The orchestrator identifies eligible nodes those with the right hardware and the model loaded in memory. * The node with the highest Reliability Score for that model and hardware tier receives the job. * The node processes the request and streams tokens back through the orchestrator to the developer in real time. * Upon completion, the node reports timing and energy metrics. These are verified and recorded for billing and analytics. # Our Products Source: https://wp.farlabs.ai/about-far-labs/products/our-products FARAI FAR AI is a secure, efficient and distributed AI inference. We connect consumer and enterprise GPUs into a single network, intelligently routing inference requests to optimal nodes for performance and reliability unlocking the potential of over 3B idle GPUs worldwide and giving their owners the opportunity to get paid for providing compute resources. While developers use a simple API to run AI inference across existing systems, integrate AI into products, and build new applications and startups. COMPANION LEAGUE A living arena where AI agents battle, evolve, and discover new strategies 24/7. These autonomous tournaments keep the game meta fresh, power endless competitive content, and create a new kind of esports experience built on ever-evolving AI. AI COMPANION Your personal in-game coach and partner. The  Companion learns your  playstyle, adapts to your emotions, and supports  you with tips, actions, and encouragement in real time. Always by your  side, it grows with you making every match smarter and more personal. AI PLAYER Human-like AI teammates and opponents that make every match feel alive. Trained to play like real people with natural  quirks, pacing, and  advanced tactics they keep queues full and games  balanced, even when  human players aren’t available. FARCHAT FAR Chat is our consumer product that turns prompts into a social and monetizable layer. It helps people find, verify, and reuse the prompt behind AI content, then save or publish it as a prompt pack. Built to drive demand for the network, it brings users, creators, and repeat usage into one loop. # Introduction Source: https://wp.farlabs.ai/abstract *FAR AI is a distributed AI inference network that makes powerful artificial intelligence fast, affordable, and accessible, by harnessing GPU hardware that already exists around the world.* Today, accessing state-of-the-art AI requires routing every request through centralized cloud data centers. That model is expensive and capacity-constrained. FAR AI turns the equation on its head: instead of building more data centers, we connect developers who need AI compute with everyday GPU owners who have compute to spare. Developers access FAR AI through a standard API and get fast, affordable inference on the world's best open-weight language models. GPU owners from individual enthusiasts to gaming cafes to enterprise fleet operators, install the FAR AI Node Client and earn rewards for every request their hardware serves. Underpinning the network is a cryptographic trust layer that continuously verifies every node is running the model it claims, completing jobs honestly, and reporting energy usage accurately. This gives developers the reliability and quality they expect from a cloud provider, while keeping the economics and the compute distributed across a global network of operators. "FAR AI connects the world's idle GPU capacity with the world's growing demand for AI, creating a network that scales with people, not data centers." # AI Inference Is Too Expensive Source: https://wp.farlabs.ai/ai-inference-is-too-expensive Running a large language model is resource-intensive. The specialized hardware required high-end data center GPUs costs tens of thousands of dollars per unit and consumes hundreds of watts under load. Cloud providers who own these GPUs price their inference services accordingly, putting advanced AI out of reach for many developers, startups, and researchers. Meanwhile, the same AI workloads that strain cloud budgets could realistically run on consumer and prosumer GPUs already owned by millions of people worldwide. These cards sit idle for most of the day during work hours, overnight, between gaming sessions, representing enormous untapped compute potential. # Available Models Source: https://wp.farlabs.ai/available-models FAR AI supports the leading open-weight language models, with availability determined by the VRAM distribution of the active node fleet. As the network grows and higher-VRAM nodes join, the range of available models expands automatically. Developers can query the API at any time to see which models are currently available and their associated pricing. # Centralized Infrastructure Doesn't Scale Sustainably Source: https://wp.farlabs.ai/centralized-infrastructure-doesnt-scale-sustainably Every percentage point of growth in AI adoption translates directly into more data center construction, more power consumption, and more capital expenditure by a handful of large providers. This creates a structural bottleneck: the pace at which AI can be made available to the world is limited by how quickly a few companies can build and finance new facilities. A distributed model removes this bottleneck entirely. When compute is sourced from existing hardware across the globe, capacity grows organically as more node operators join with no single point of investment, failure, or control. # Product Timeline Source: https://wp.farlabs.ai/conclusion-copied-1 FAR AI is building toward a future where AI inference is open, affordable, verifiable, and private by default. The roadmap progresses through three phases, each building on the foundations of the last. | **Phase** | **Timeline** | **What It Delivers** | | ---------------------------- | -------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | | Phase 1 - Alpha Network | Live | Full security runtime, Reliability Score routing,
TCP and homogeneous distributed inference,
OpenAI-compatible API. Calibration of routing
weights and tiered model catalog in progress. | | Phase 2 - FAR-ZK Lazy | Near-term | Zero-knowledge proofs on the dispute
resolution path. When a developer disputes
a response, the node produces a cryptographic
proof of correct execution. Dispute resolution
becomes automatic and mathematically certain. | | Phase 3 - FAR-ZK Eager | Mid-term | Per-inference zero-knowledge proofs. Every
completed job is accompanied by a proof of
correct execution, verified before payment is
authorized. Correctness and payment become
atomic. | | Long Horizon - Input Privacy | Research phase | Goal: the developer's prompt is never visible
in plaintext to any FAR-controlled component. | Each phase is additive, earlier guarantees remain in place as new capabilities are introduced. The Alpha network's statistical verification does not go away when ZK proofs arrive; it runs alongside them, providing defense in depth. # Core Advantages Source: https://wp.farlabs.ai/conclusion-copied-2 | **Open by design** | **Affordable by structure** | | -------------------------------------------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | No proprietary hardware, no vendor lock-in.
FAR AI runs on GPUs already owned by millions
of people worldwide. | Routing inference through existing hardware
removes data center overhead from the cost
equation entirely. | | **Trustworthy by cryptography** | **Fair by construction** | | Continuous verification means developers
get cloud-grade reliability without centralized
infrastructure. | Tiered routing ensures that reliability not just
raw GPU power determines which operators
earn the most. | | **Scalable by nature** | **Private by roadmap** | | Capacity grows every time a new node joins.
There is no bottleneck at the infrastructure layer. | ZK proofs ensure prompts stay hidden from
the FAR AI orchestrator. A confidentiality layer -
TEE today, FHE as the long-term target extends
that guarantee to the executing node. | FAR AI is not just a cheaper way to access AI. It is a fundamentally different infrastructure model, one where the network grows stronger with every node that joins. # Start Here Source: https://wp.farlabs.ai/conclusion-copied-3 Whether you are a developer looking for a more affordable path to powerful AI, or a GPU owner looking to put idle hardware to work, FAR AI is built for you. **Developers:** Visit [farlabs.ai](http://farlabs.ai) to create an account and access the API documentation. **Node Operators:** Download the FAR AI Node Client from [farlabs.ai](http://farlabs.ai) and start earning with your GPU today. # Drop-In Compatibility Source: https://wp.farlabs.ai/drop-in-compatibility FAR AI exposes a fully OpenAI-compatible API. Any application already integrated with OpenAI's API can switch to FAR AI by updating a single configuration value, the endpoint URL. No SDK changes, no new request formats, no migration work. Streaming responses, tool calling, multi-turn conversation history, system prompts, and vision inputs are all supported out of the box. The experience is identical to a major cloud provider, at a different price point. # How Node Operators Earn Source: https://wp.farlabs.ai/economic-model-1 Node operators are paid for the work their hardware performs. Earnings are calculated transparently based on two factors: the number of tokens their node generates, and the energy their GPU consumes in doing so. This ties payouts directly to the value delivered, operators who run faster hardware, serve more requests, and maintain high uptime earn proportionally more. Payouts are tracked in real time on the operator dashboard, with full visibility into tokens served, energy consumed, and earnings accumulated across any time period. # System Requirements Source: https://wp.farlabs.ai/ecosystem FAR AI is designed to run on widely available consumer and prosumer GPU hardware. The following table summarizes the minimum and recommended specifications for a node. | **Component** | **Minimum** | **Recommended** | | ---------------- | -------------------------------------------------- | ----------------------------------------------------------------- | | GPU | NVIDIA, 10 GB VRAM,
RTX 30 series or newer | RTX 3080 / 3090 / 4090 or
equivalent, 16–24 GB VRAM | | System RAM | 16 GB | 32 GB or more | | Storage | 100 GB free for model
cache, SATA SSD | 500 GB or more, NVMe
SSD strongly preferred | | Internet | 100 Mbps for initial setup;
2–3 Mbps ongoing | 200 Mbps or more; under
50 ms latency to
orchestrator | | Power Supply | 150 W headroom above
GPU rated power | Stable supply with UPS
protection recommended | | Operating System | Windows 10 (build 19041+)
or Windows 11 | Windows 11, latest updates
applied | VRAM is the most important factor for earning potential. More VRAM allows a node to serve larger, more valuable models. NVIDIA drivers must be installed before running the FAR AI installer, the installer handles all other software dependencies automatically. Storage speed has a meaningful impact on job start latency. A large model loads significantly faster from a high-speed NVMe drive than from a standard SATA SSD, and faster load times mean higher effective availability and more earning opportunities. The installer automatically selects the best available drive for model caching. # Fair Earnings Across Hardware Tiers Source: https://wp.farlabs.ai/fair-earnings-across-hardware-tiers A potential pitfall in any distributed inference network is that faster GPUs earn far more per hour than slower ones, even when both are equally reliable. This would make the Reliability Score meaningless as an earnings signal for operators with mid-range hardware. FAR AI addresses this differently depending on the inference mode in use. **Single-machine and same-LAN inference** For models that fit within a single node or a local cluster of nodes, FAR AI uses a tiered model catalog. Each model is tagged with a minimum hardware tier, and the orchestrator routes jobs only to nodes in the appropriate tier. Within each tier, nodes have comparable hardware, so a higher Reliability Score translates directly into proportionally more work and higher earnings. When demand temporarily exceeds a tier's capacity, a hardware-aware rate adjustment ensures nodes serving above their native tier are compensated fairly for the additional cost. **Cross-network distributed inference** For models too large for any single device, the orchestrator splits the job across multiple nodes on the broader FAR AI network. In this mode there is no tier concept in the single-machine sense. The orchestrator identifies any combination of reachable nodes whose combined available GPU memory is sufficient to hold the model, assigns each node the weight shard that fits its hardware, and compensates each node proportionally to the memory it contributed and the energy it consumed. An operator with a smaller GPU is not below any tier, they contribute the slice they can carry and are paid for exactly that slice. Reliability Score still governs which nodes are selected when multiple combinations could satisfy the memory requirement. # For Developers Source: https://wp.farlabs.ai/for-developers **For Developers: A Familiar API, Unfamiliar Economics** Developers interact with FAR AI through a standard, OpenAI-compatible API. Switching from an existing provider requires changing a single line of code, the endpoint URL. Everything else works as before: streaming responses, tool calling, vision inputs, multi-turn conversations, and system prompts are all fully supported. What changes is the cost. By routing inference through a distributed network of existing hardware rather than purpose-built data centers, FAR AI can offer substantially lower per-token pricing without sacrificing reliability or response quality. # For Developers: Predictable, Competitive Pricing Source: https://wp.farlabs.ai/for-developers-predictable-competitive-pricing Developers pay per token, with pricing that reflects the cost of the hardware tier serving their request. Because FAR AI routes work to existing hardware rather than maintaining its own data center fleet, the cost structure is fundamentally different from centralized providers and the savings are passed on to developers. Developers can see model availability, current pricing, and historical performance metrics through the FAR AI developer portal, enabling informed decisions about model selection and budget planning. # For Node Operators Source: https://wp.farlabs.ai/for-node-operators **For Node Operators: Idle Hardware Becomes Income** Anyone with a compatible NVIDIA GPU can become a FAR AI node operator. The Node Client software installs in minutes on Windows and runs quietly in the background. Operators earn rewards for every request their hardware serves, with payouts calculated transparently based on tokens generated and energy consumed. Node operators don't need to manage model selection, routing, or developer relationships. The FAR AI orchestrator handles all of that automatically, directing work to the right hardware at the right time and handling all billing and settlement. | **Earn passively** | **No configuration needed** | | :-------------------------------------------------------------------------------------------- | :------------------------------------------------------------------- | | Your GPU works while you sleep,
between gaming sessions, or
during business hours. | The orchestrator assigns models and
routes jobs automatically. | | **Transparent payouts** | **Scale at your pace** | | :------------------------------------------------------------------ | :--------------------------------------------------------------------------- | | Earnings are tracked in real time
on your operator dashboard. | Run one GPU or dozens, the network
accommodates all fleet
sizes. | # How Verification Works Source: https://wp.farlabs.ai/how-verification-works The FAR AI security runtime operates on four pillars: * **Model integrity -** when a node commits to serving a specific model, it also commits to a cryptographic fingerprint of that model's weights. The orchestrator continuously issues random challenges that the node can only answer correctly if it is genuinely holding and running the committed model. A node serving a different model cannot pass these challenges. * **Hardware presence -** specialized proof-of-work challenges verify that the node's declared GPU is genuinely present, active, and holding the committed model in memory.These challenges have strict timing requirements that cannot be met by a node fetching data from a remote source on demand. * **Network topology -** the orchestrator verifies that a node's declared network configuration is consistent with its observed latency and connectivity patterns, making it impossible to misrepresent hardware location or connectivity. * **Settlement integrity -** every completed job generates a tamper-evident, cryptographically chained record linking the job to the node, the model, and the energy consumed. The orchestrator reconciles these records against its own independent history. Any gap, replay, or discrepancy is detected automatically. Every node is verified continuously, not just at onboarding, but throughout every job it runs. # About FAR Labs Source: https://wp.farlabs.ai/index FAR Labs is an AI native lab building infrastructure, AI as Service (AIaS) and physical AI systems. \ \ We are building FAR AI -  secure, efficient and distributed AI inference. We connect consumer and enterprise GPUs into a single network, intelligently routing inference requests to optimal nodes for performance and reliability unlocking the potential of over 3B idle GPUs worldwide and giving their owners the opportunity to get paid for providing compute resources. While developers use a simple API to run AI inference across existing systems, integrate AI into products, and build new applications and startups. 3 \ **Join our socials** # Trust and Verification Source: https://wp.farlabs.ai/model *FAR AI's security runtime provides continuous, cryptographic proof that every node is doing what it claims without requiring trust in any individual operator.* FAR Labs # Model Management Source: https://wp.farlabs.ai/model-management Node operators don't choose which models to run. The orchestrator assigns models based on network demand and the operator's hardware capabilities. Models are downloaded and cached locally on the node, so load times drop to near zero for frequently requested models and the orchestrator can remotely load or unload models as demand patterns shift. This means operators always serve the models where they can add the most value, and developers always have access to the models they need, without any manual coordination between the two. # Multi-Machine Setups Source: https://wp.farlabs.ai/multi-machine-setups Node operators who want to serve larger models can connect multiple machines on the same local network to pool their GPU memory. All machines in the group must be connected by wired gigabit Ethernet, Wi-Fi is not suitable for the low-latency communication required between nodes in a distributed inference session. # No Capital Gates Source: https://wp.farlabs.ai/no-capital-gates Any node meeting the minimum hardware requirements can join the FAR AI network and begin receiving work. There is no capital staking requirement and no entry fee beyond the cost of qualifying hardware. Newer nodes start with a limited score history and accrue routing priority as they demonstrate consistent behaviour over time. This design keeps the network open and competitive. Established operators earn their advantage through demonstrated reliability, not through having joined first or paid the most. # Overview Source: https://wp.farlabs.ai/security *The Reliability Score is how FAR AI ensures that the best developers get routed to the best nodes automatically, continuously, and without operator intervention.* FAR Labs # Solution Source: https://wp.farlabs.ai/semantic *FAR AI is a three-sided platform connecting developers, node operators, and the orchestration layer that makes them work seamlessly together.*