This changes everything.
Introducing MapleâŻProxy â a powerful bridge that lets you call Mapleâs endâtoâendâencrypted LLMs with any OpenAIâcompatible client, without changing a line of code. https://blossom.primal.net/19887ff7b4b7c95e1b195c5b7790f9457035a75cb5e6690dab0129fc4cb2dc2c.mp4
Read the full announcement or check out the thread
Why it matters: Most AIâpowered apps (calorie trackers, CRMs, study companions, relationship apps eyes đ) send every prompt to OpenAI, exposing all user data. With MapleâŻProxy, that data never leaves a hardwareâisolated enclave.
https://blossom.primal.net/21b4ee5b782d240d3eb06c3db41ffa73ec79734f104ee94113ff2dc6d7e771c3.mp4
How it works:
đ Inference runs inside a Trusted Execution Environment (TEE).
đ Endâtoâend encryption keeps prompts and responses private.
â Cryptographic attestation proves youâre talking to genuine secure hardware.
đ« Zero data retention â no logs, no training data.
Readyâtoâuse models (payâasâyouâgo, perâŻmillion tokens):
- llama3â3â70b â general reasoning
- gptâossâ120b â creative chat
- deepseekâr1â0528 â advanced math & coding
- mistralâsmallâ3â1â24b â conversational agents
- qwen2â5â72b â multilingual work & coding
- qwen3âcoderâ480b â specialized coding assistant
- gemmaâ3â27bâitâfp8âdynamic â fast image analysis
Realâworld use cases:
đïž A calorieâcounting app replaces public OpenAI calls with MapleâŻProxy, delivering personalized meal plans while keeping dietary data private.
đ A startupâs internal knowledgeâbase search runs through the proxy, so confidential architecture details never leave the enclave.
đ©âđ» A codingâassistant plugâin for any IDE points to http://localhost:8080/v1 and suggests code, refactors, and explains errors without exposing proprietary code
Getting started is simple:
Desktop app (fastest for local dev)
- Download from trymaple.ai/downloads
- Sign up for a Pro/Team/Max plan (starts at $20/mo)
- Purchase $10+ of credits
- Click âStart Proxyâ â API key & localhost endpoint are ready.
Docker image (productionâready)
- `docker pull ghcr.io/opensecretcloud/maple-proxy:latest`
- Run with your MAPLE_API_KEY and MAPLE_BACKEND_URL
- You now have a secure OpenAIâcompatible endpoint at http://localhost:8080/v1
Compatibility: Any library that lets you set a base URL worksâLangChain, LlamaIndex, Amp, Open Interpreter, Goose, Jan, and virtually every OpenAIâcompatible SDK.
Need more detail? Check the technical writeâup, full API reference on GitHub, or join the Discord community for realâtime help.
https://blog.trymaple.ai/maple-proxy-documentation/
Start building with private AI today: download the app or pull the Docker image, upgrade to a plan, add a few dollars of credits, point your client to http://localhost:8080/v1, and secure all your apps.
Interesting use of TEEs!
What about Claudo Code?
Some info below, not sure if itâs 100% private
https://github.com/anthropics/claude-code/issues/7178?utm_source=chatgpt.com
Love it
So now MapleAI via routstr paid with ecash instead of a login?
not the same type of proxy but would be great! nostr:nprofile1qqs8msutuusu385l6wpdzf2473d2zlh750yfayfseqwryr6mfazqvmgpz3mhxue69uhhyetvv9ukzcnvv5hx7un89uqnqamnwvaz7tmfw33ks7fdvahkcer9deex7epdve6hycm0de6xsmmsdpskwatn9eekxctjv93zu6td9ufpdr9u should run a routstr proxy
Routstr doesnât speak OpenAI API?
It does in that it can use openai upstream and issue api keys from the routstr client to use in other openai clients.
I think Maple may sit more as an upstream provider from a rouststr perspective. Like that my routstr provider would use maple as an upstream provider and users would query my provider, pay in cashu and receive the response from maple via my proxy.
Maple AI could run a routstr proxy and offer no account kyc free access to maple.. but all I see is that maple could add nostr login and save encrypted chats to nostr relays like routstr is doing now.
Just spit balling what I think is the configuration.
it does but we need end to end encryption so the routstr node should not touch the unencrypted part, and it would be sick if marple hosts a routstr node to serve its own models
I'm insanely bullish on confidential compute and nostr:nprofile1qyjhwumn8ghj7en9v4j8xtnwdaehgu3wvfskuep0dakku62ltamx2mn5w4ex2ucpxpmhxue69uhkjarrdpuj6em0d3jx2mnjdajz6en4wf3k7mn5dphhq6rpva6hxtnnvdshyctz9e5k6tcqyp7u8zl8y8yfa87nstgj2405t2shal4rez0fzvxgrseq7k60gsrx6zeuh5t / nostr:nprofile1qyshwumn8ghj7en9v4j8xtnwdaehgu3wvfskuep0w3uhqetnvdexjur5qyfhwumn8ghj7ur4wfcxcetsv9njuetn9uqzphchxfm3ste32hfhkvczzxapme9gz5qvqtget6tylyd7wa8vjecgaq9enl
AI will provide a tremendous amount of value for the world. But it will also allow for mind manipulation on steroids that makes what happened with web 2.0 cos like Facebook look tame. But there's now an alternative to the closed source mega models and hosting - OpenSecret / MapleAI effectively offer developers a publicly accessible (and open source) version of Apple's encrypted inference cloud: https://security.apple.com/blog/private-cloud-compute/
Thank god for cryptography, the work of people like Anthony / nostr:nprofile1qydhwumn8ghj7mnjv4kxz7fwvvkhxar9d3kxzu3wdejhgtcprpmhxue69uhhqun9d45h2mfwwpexjmtpdshxuet5qqsgafy9ye4j9p2x8vfmlq6equtpcg4m8ks7v545g0d3f7wwueeq5scz9g5a6, and organizations like Apple
Wen NWC payment option? I just want to let it rip with a budget.
Hey nostr:nprofile1qqsdy27dk8f9qk7qvrm94pkdtus9xtk970jpcp4w48k6cw0khfm06mspz4mhxue69uhkummnw3ezummcw3ezuer9wchsz9mhwden5te0wfjkccte9ehx7um5wghxyctwvshsh9tjpk reaching out on here as well. Can we make something happen to get Maple encrypted LLMs available on PPQ?
Thatâs because OpenAIâs open weight is available on AWS now, right?
We have the OpenAI open-source gpt-oss-120b model running in a secure GPU. That gives users end-to-end encryption. It is comparable to the GPT-4 family of models.
Hey guys, this is great, thank you for building.
Can you please enable monthly bitcoin payments.
Incredible. Big unlock for privacy focused AI
LFG đ„đ
â
I need to try Maple soon đ
Well worth the annual subscription!! Love the product
Do I understand that correctly: I need to be a pro subscriber and than purchase API credits on top?!
I love the zero knowledge approach you guys have, but for the occasional document analysis I can't justify 20$/month. Would gladly pay twice what other models charge on a pay-per-token basis like on ppq.ai
