Idle

The first universal compute layer. Turn any idle device - GPU, laptop, PC - into a node on the global AI compute network. |

[Idle @IdleProtocol]

$IDLE is now live on Solana.

AjLhrxN2yrCe45Y2KGPMZCkBm6NpN43jWqPkdZq6pump

Quick rundown on what it does and why it exists.

IDLE Protocol takes 15% of every gateway payment. That fee is the entire token economy. 10% auto-swaps to $IDLE on Jupiter every hour and sends it to the burn address. 5% covers infrastructure and development. The other 85% goes straight to the resource owner's wallet, same block.

Staking gives you a bigger cut. Base split is 85/15. Stake 10k $IDLE and you keep 87%. Stake 50k and it's 90%. 250k gets you 95%. The protocol adjusts your split ratio on the next payout.

Burning $IDLE unlocks features: priority routing (your jobs match first), advanced task types, analytics on your gateway traffic, and custom endpoint domains. Priority routing means lower latency and higher uptime for your resources.

The thesis: $IDLE supply can only go down. Every gateway call generates a burn. More resources on the network means more API calls, which means more fees, which means more buying pressure and more tokens removed from circulation permanently. The flywheel is mechanical, not speculative.

https://t.co/0n8skIkzCO

[Idle @IdleProtocol]

Introducing IDLE.

Your PC earns from AI agent tasks. Your wallet compounds yield. Your agent sells its downtime.

Everything idle. Now earning.

[Idle @IdleProtocol]

Your GPU is an appreciating asset now.

New Bloomberg data: Nvidia H100s still rent at almost 80% of their launch price - in their fourth year. AWS hasn't retired six-year-old A100 servers because demand won't allow it. GPU rental prices climbed all year as demand for AI compute outstripped supply of new chips.

For two decades the rule was: compute gets cheaper over time. That rule is dead. Compute is scarce, priced like it, and getting scarcer.

Which means the GPU sitting in your PC doing nothing is leaving money on the table every hour it idles.

IDLE Protocol connects it. Real inference jobs routed to your hardware. USDC on Solana per completed job. 47,000+ operators already earning.

The market repriced compute. Time to reprice what yours is worth.

https://t.co/3T6ELd892l

[Idle @IdleProtocol]

IDLE Protocol now supports AMD Radeon GPUs.

Until today, every distributed compute network required NVIDIA CUDA. AMD owners locked out. That ends now.

IDLE nodes now run natively on AMD Radeon RX 7900 XTX, RX 9070 XT, RX 9070, and Radeon AI PRO R9700. Powered by ROCm 7.2 and vLLM - the same production inference stack running on our NVIDIA nodes. Continuous batching. PagedAttention. OpenAI-compatible endpoint.

RX 7900 XTX runs Llama 3.1 8B at 96 tokens/second - 75% of RTX 4090 throughput at half the price. RX 9070 XT delivers 24GB GDDR7 for $599. AMD holds 5-8% of the GPU market and growing fast - that's tens of millions of consumer GPUs that just came online for distributed compute.

Every AMD node earns USDC on Solana per completed job. No CUDA required. No NVIDIA tax.

The distributed compute network just doubled its addressable hardware.

[Idle @IdleProtocol]

distributed, hardware-agnostic, permissionless, global, autonomous, on-chain, open-weight, agent-native, verifiable, resilient, composable, unstoppable, real

[Solana @solana]

fast, affordable, global, permissionless, resilient, responsive, scalable, lindy, efficient, accessible, interoperable, available, expressive, observable, secure, verifiable, concurrent, composable, robust

[Idle @IdleProtocol]

The AI inference market is undergoing the biggest architectural shift since the cloud.

IDC forecasts 80% of AI inference will run locally by 2027. Enterprises spent $40 billion on cloud AI inference in 2024 - and every major vendor is now scrambling to build edge platforms. Cisco. Nutanix. Red Hat. Amazon just hiked GPU prices 15%.

The reason: cloud inference has fundamental limits. High latency. High cost. Privacy exposure. Centralized failure. Every serious enterprise is now moving inference to the edge.

IDLE Protocol has been building the edge inference layer for months.

47,000+ nodes deployed globally on consumer hardware. Sub-100ms latency to users because compute happens on devices near them, not in a data center on the other side of the country. No cloud GPU markup. No data leaving the region.

The market is finally catching up to where IDLE already is.

https://t.co/9bN0AUM5eJ

[Idle @IdleProtocol]

A look at the ecosystem around IDLE.

Discoverable on Anthropic's MCP Registry, Coinbase x402 Bazaar, Amazon Bedrock AgentCore, Hugging Face Inference Providers, LangChain, RapidAPI, https://t.co/EoWuiH6bi8, and DePINscan.

Models routed through IDLE: NVIDIA NIM (incl. Nemotron 3 Ultra), Mistral, Google Gemini, Kimi K2.6, Nous Research Hermes, Microsoft MAI, DeepSeek V4, Zai GLM 5.2.

Demand routed in from SAID Protocol, Xona Agent, Hatcher Labs, SolRouter, and Prova, and more.

Powered by Alchemy, IBM Partner Plus, SKALE, PayAI, and Privacy Cash.

[Idle @IdleProtocol]

IDLE has surpassed 549,000 on-chain transactions.

Here's where the network stands:

47,698 active network operators
549,973 transactions since launch
$19,596 distributed to operators

Every number verifiable on-chain. Live data tracked on @Dune, updated in real time.

[Idle @IdleProtocol]

IDLE Protocol now supports https://t.co/S37SUBM9nE's GLM family.

https://t.co/S37SUBM9nE just released GLM 5.2 - the first open-weight model to top the leaderboard on real coding tasks. Vercel's CEO called it "genuinely impressed, almost shocked."

The full family is now available through IDLE:

GLM 5.2 - routed through NIM for frontier tasks
GLM 4.7-Flash - served across consumer GPU nodes for high-throughput inference
GLM-4.5-Air - mid-range coding on IDLE's 16GB+ VRAM tier

Paid per request in USDC on Solana.

The frontier just went open. IDLE serves it end to end.

[Idle @IdleProtocol]

Every DePIN compute network has the same unsolved problem: nodes can lie about their hardware. Claim an H100. Deliver an RTX 3090. Run jobs in a VM. Fake the specs.

The only solution today is enterprise hardware attestation - NVIDIA CC mode, Intel TDX. Neither works on consumer GPUs.

IDLE just solved it.

Every node on IDLE now runs continuous behavioral fingerprinting - clock jitter analysis, thermal signature verification, memory bandwidth attestation, and randomized benchmark challenges. Each measurement signed and anchored to Solana. Nodes that don't match their claimed hardware get flagged and removed automatically.

Hardware attestation without enterprise hardware. Mathematical proof on consumer GPUs.

This is what makes distributed compute actually work at scale.

https://t.co/zYlHMXZixN

[Idle @IdleProtocol]

IDLE Protocol now supports @NVIDIA Nemotron 3 Nano Omni - the first multimodal model on the IDLE network.

Nemotron 3 Nano Omni unifies vision, audio, and language in a single model - 9x more efficient than comparable systems for agentic AI workloads. Built for autonomous agents that need to see, hear, and reason in one pass.

Available as an NVIDIA NIM microservice. IDLE already routes inference through NVIDIA NIM. The integration is one endpoint away.

Multimodal AI, served by IDLE's distributed compute network. Paid per request in USDC on Solana. No subscriptions.

[PayAI Network | x402 Facilitator @PayAINetwork]

4/ Another integration to distribute compute around Solana ft. IDLE Protocol
https://t.co/3hpVeCM3O3

[Idle @IdleProtocol]

(3/3) How to connect your GPU and start earning.

If you have an RTX with 16GB+ VRAM, you can join the network and start serving gpt-oss-20b inference jobs immediately. Hardware tier is detected automatically on registration. Jobs route to you based on capacity, reliability score, and latency.

Every completed job gets paid in USDC on Solana automatically. 85% of every request fee goes to the node operator. Settlement runs every 10 minutes.

The compute the AI economy needs is already in people's hands. IDLE connects it.

[Idle @IdleProtocol]

1/ IDLE Protocol now supports OpenAI gpt-oss-20b.

OpenAI's first open-weight model in six years. 20 billion parameters, matching o3-mini reasoning performance, fully Apache 2.0 licensed.

Now running on IDLE's distributed compute network - served by consumer GPUs globally, paid per request in USDC on Solana.

The closed lab that defined modern AI just released their first open model. IDLE is the first decentralized networks serving it.

[Idle @IdleProtocol]

(2/3) How it actually works.

gpt-oss-20b uses a Mixture-of-Experts architecture - 21 billion total parameters with 32 experts, but only ~3.6 billion activate per token via top-4 expert routing. Combined with native MXFP4 quantization, the model fits in just 16GB of VRAM.

That means any node on the IDLE network with an RTX 4080, RTX 5080, RTX 3090, RTX 4090, or RTX 5090 can run it locally - no data center required.

Nodes use vLLM with continuous batching and PagedAttention. Multiple concurrent requests get processed in parallel. GPU utilization stays high. Throughput stays consistent.

[Idle @IdleProtocol]

A look at the ecosystem around IDLE.

Where developers and agents discover the network:
- @AnthropicAI MCP Registry - installable in one command for Claude, Cursor, and Windsurf
- @Coinbase x402 Bazaar - discoverable to every x402-compatible agent
- Amazon Bedrock AgentCore - listed as a compute provider for AWS agents
- @huggingface - IDLE is now an Inference Provider
- @LangChainAI + LangGraph - IDLE is natively callable from any LangChain agent
- RapidAPI - accessible to 4M developers
- https://t.co/EoWuiH6bi8 - listed compute infrastructure
- DePINscan

Partners powering the stack:
Alchemy - RPC infrastructure + backed by $20M Solana Fund
IBM - Partner Plus Service Partner
SKALE Network - gasless x402 dual-network
PayAINetwork - x402 facilitator
@ThePrivacyCash - ZK-shielded operator payouts
@SolRouter - private inference venue
@SaidInfra - IDLE as default execution layer
@AskProva - default inference venue, strategic investment, on-chain proof layer
@XonaAgent - agent infrastructure integration

[Idle @IdleProtocol]

1/ IDLE Protocol just crossed 540,000 on-chain transactions.

Where the network stands right now:

17,345 active network operators
542,934 transactions since launch
$18,903 distributed to operators in USDC

Every number verifiable on-chain. Live data tracked on @Dune, updated in real time.

[Idle @IdleProtocol]

IDLE Protocol is now natively compatible with @LangChain.

Any LangGraph agent can connect to IDLE's distributed compute network via our MCP server - web scraping, inference, DNS, health checks, Solana data queries, agent tasks. All 16 endpoints. Available as tools your agent can discover and call automatically.

No new code. No new SDK. LangGraph already supports MCP natively via langchain-mcp-adapters.

LangGraph powers agents at Uber, JP Morgan, and Klarna. Those agents can now route compute through IDLE's node network, paying per request in USDC on Solana automatically.

[Idle @IdleProtocol]

https://t.co/Qpnc9CA1PZ

[Idle @IdleProtocol]

We've been building this for months. Today we wrote it all down.

How IDLE Protocol runs distributed inference across thousands of consumer GPUs - the architecture, the hardware tiers, vLLM under the hood, consensus validation, and why data parallelism at the network level beats tensor parallelism in a data center for this problem.

Full technical breakdown.

[Idle @IdleProtocol]

2/ We believe on-chain verification for AI will become standard infrastructure on @solana - the way oracles became standard for DeFi.

@AskProva is building that layer. 1,000+ proofs anchored. Program verified by OtterSec. Open to any developer, any app, any agent.

We wanted to be early. This is just the beginning.

[Idle @IdleProtocol]

We're making a strategic investment into @AskProva - the on-chain proof layer for AI on @solana.

IDLE processes thousands of AI inference jobs daily across a global operator network. The question we keep hearing: how do you prove the model actually ran? Not trust. Prove it. On-chain.

Prova built the answer. Their Solana program verifies every proof at the chain level - ECDSA signature recovery, TEE key registry linked to real NVIDIA/Intel attestation. Fake a proof and Solana rejects it. We integrated it. We tested it. Then we invested.

https://t.co/ypLMGwo2Od