Private inference router for onchain agents. Models run in secure enclaves, every response is signed. 0x47b537f7c2c2183967fe5ba883291b8ada490db6
Joined October 2026
- Tweets126
- Following11
- Followers369
- Likes187
Your onchain agent thinks, pays, connects and trades. Today, every step is watched.
OffRouter is the private router for onchain agents:
01 Private Inference: 18 models in GPU enclaves
02 Private Payments: no account, ZK-funded keys
03 Private Gateway: any endpoint, behind an enclave
04 Private Swap: sealed orders, live in beta
Built on @RobinhoodCrypto.
Every swap you make is a signal.
Bots see your token, your size and your slippage before you execute, then buy ahead of you and sell back into you. You pay for being seen.
Not anymore. Private Swap is live in beta on @RobinhoodChain: your order goes to an enclave and reaches the chain only as a hash until it’s filled.
Nothing to read. Nothing to front-run. 🧵
How it works:
1. Get a quote and confirm once in your wallet
2. Your payment goes into escrow under a hash of your order. What you’re buying, your minimum and where it’s paid go only to the enclave, never on chain
3. The enclave fills the open orders for each pair together in one swap, then pays each recipient after a short random wait
Your funds can only go two ways: into your swap, or back to you. Not filled at your minimum? It’s sent back.
Once an order expires, anyone can trigger the refund on chain, and it can only return to the wallet that paid.
Don’t trust, verify. The executor’s hardware report is public, so you can check the key filling your order is the enclave’s. The swap page also rebuilds every message itself and refuses to sign if anything differs from what you asked for.
Beta for now: ETH ↔ $OFFROUTER
More pairs and higher limits are next.
For the most privacy, send to a fresh address.
Swap privately 👉 offrouter.org/swap
Four new private models on OffRouter:
→ glm-5.2: reasoning and tools, $1.26 / $3.00
→ qwen3.6-27b: reasoning and vision, $0.32 / $3.25
→ qwen3.6-35b-a3b: reasoning and vision, $0.20 / $1.27
→ qwen3-vl-30b: vision, $0.20 / $0.70
18 models now, every one sealed in a GPU enclave with a signed receipt. Pay in stablecoins, no account.
offrouter.org/models
AI infrastructure shouldn't become a single point of exposure.
Your API keys, usage data, and inference requests deserve stronger protection.
OffRouter is building for that future.
Private inference. Sealed GPU enclaves. Cryptographic receipts.
Your intelligence stays yours.
🇺🇸 🚨 USA: OPENROUTER.AI USER DATABASE AND API KEYS ALLEGEDLY FOR SALE
A threat actor on an underground forum claims to be selling a database associated with openrouter[.]ai, including user records and API keys.
The actor claims:
* 251,996 user rows (about 250,000 unique users)
* 207,847 API keys (described as sk-or-v1 format; active / revoked / expired mix)
Alleged user fields include emails, usernames, password material (hashed and some plaintext claimed), names, signup dates, credits, plan and country. Alleged API-key fields include user linkage, key values, creation/last-used timestamps, spend and rate limits.
The claim has not been independently verified.
⚠️ Analyst Note:
If authentic, exposed API keys would enable unauthorized LLM spend and account takeover, and user emails/credentials would fuel phishing and credential stuffing. Key values and sample rows are redacted from our images and are not reproduced here. Unverified threat-actor claim, NOT confirmation of an OpenRouter compromise.
Source: this is for subscribers only. Subscribe: xitter.cf/DailyDarkWeb/creator-s…
#DDW #DarkWeb #USA #OpenRouter #APIKeys #DataBreach #ThreatIntelligence
Roadmap update: more of OffRouter is live.
→ 14 private models, each in a GPU enclave
→ Unlinkable payments: fund a key with a ZK proof, no wallet attached
→ Private Gateway (beta): put any OpenAI-compatible endpoint behind an enclave
→ OffChat: private chat, nothing stored server side
Next up: Private Swap and private agents.
offrouter.org/roadmap
OffRouter retweeted
Data centers can't access debt financing?
Come to Decentralized AI where you can access
> Financing from emissions and/or directly from investors
> Idle heterogenous GPU, inference, talents, and everything you need
> Devout community supporting open-source, decentralized technology
If this is not a signal for Onchain and Decentralized AI, I don't know what is
> Kimi K3.1 Pro matching frontier capabilities
> Local conversational model that sounds more real than humans
> Renewed wave of interests for Confidential Inference
I know everything is going down but I can't imagine a world where these models don't get immediately integrated by onchain providers and sold for 30-80%+ off
Open Frontier Intelligence at affordable prices
November is gonna be an exciting month for Onchain Inference
onchain inference.
what does everything below have in common?
wallets, CEXes, DEXes, onchain credit, onchain experiments, Stock Tokens, fixed-rate markets, institutional custody, block explorers, analytics, data APIs, on-ramps, cross-chain swaps, index products, liquidity infrastructure
it’s all live on Robinhood Chain
First repo: an MCP server for our private models.
It runs on your machine, verifies the enclave, then encrypts your prompt before it leaves. Our gateway only ever sees ciphertext.
Free to try:
claude mcp add offrouter -- npx -y github:offrouter/mcp
github.com/offrouter/mcp
OffRouter is on GitHub now.
We keep saying private inference should be something you can check, not something you take our word for. So the code is going public, one repo at a time.
github.com/offrouter
Your AI provider is a server too.
If it gets owned, what’s in its request log?
Behind OffRouter Private Gateway:
→ no IP, no user id, just our enclave
→ [EVM_ADDRESS_1] instead of your wallet
→ [API_KEY_1] instead of your key
→ [SEED_PHRASE_1] instead of your seed
Two lines of code, any OpenAI-compatible API. Your app still gets the real values back, with a signed receipt.
Free in beta → offrouter.org/gateway
Lesson from PHP zero-day: assume the server gets owned.
Then ask: what could the attacker read?
For most AI APIs: your prompts, in plaintext.
For confidential inference: ciphertext. The model runs inside a GPU enclave that even the host can’t look into.
Privacy should survive a breach, not just a policy.