# Introducing SayGM: An AI Gateway Built Like a Vault Every prompt you send to an AI model passes through someone else's servers first. SayGM is the AI gateway built so that "someone else" structurally can't read it — hardware-verified, not just policy-promised. Here's what that means, and why it's also cheaper than the alternatives. _Source: https://saygm.com/blog/introducing-saygm-an-ai-gateway-built-like-a-vault_ Every morning, millions of prompts leave developers' laptops and land on someone else's servers. Most of us don't think twice about it. That is, until we're the ones building something that can't afford to leak a customer record, a trade secret, or a line of proprietary code into a black box. And most of us are paying a quiet markup for that risk, baked into every provider's sticker price. That's the problem SayGM was built to solve. This is our first blog post, so consider it the front door: what SayGM is, what an AI gateway actually does, why it's cheaper than calling providers directly, and why we built its privacy on hardware you can verify instead of a policy you have to trust. Table of Contents ----------------- - [What Is SayGM?](#what-is-saygm) - [What Is an AI Gateway, and Why Use One?](#what-is-an-ai-gateway-and-why-use-one) - [Two Levels of Privacy: Anonymous Routing vs. End-to-End TEE](#two-levels-of-privacy-anonymous-routing-vs-end-to-end-tee) - [How SayGM Keeps Costs Down](#how-saygm-keeps-costs-down) - [One Unified API for OpenAI, Anthropic, and Gemini](#one-unified-api-for-openai-anthropic-and-gemini) - [Built on Decentralized, Verifiable Infrastructure](#built-on-decentralized-verifiable-infrastructure) - [What's Coming Next](#whats-coming-next) - [Say GM to AI](#say-gm-to-ai) What Is SayGM? -------------- An AI aggregator gives you one API key and one dashboard to reach a wide range of frontier and open models, instead of juggling separate accounts and SDKs for every provider. SayGM is all of that, plus a vault. SayGM gives you that same convenience: - One API key for 35+ frontier and open models. - Plug-and-play into Cursor, Cline, Claude Code, or any OpenAI-compatible tool. Just point it at SayGM and go, no code rewrites. - At or below the price the model maker charges, with no subscription lock-in. Here's where the comparison to a typical aggregator breaks down: those platforms sit in the middle of every request and can read everything you send. SayGM can't. That's the difference hardware makes, and it's the rest of this post. What Is an AI Gateway, and Why Use One? --------------------------------------- An AI gateway sits between your application and the model providers you call (OpenAI, Anthropic, Google, Moonshot etc) and gives you one consistent interface instead of separate SDKs, keys, and rate limits for each. Most gateways exist to solve routing and cost problems: failover when a provider goes down, load balancing across models, and centralized billing. If you're evaluating openrouter alternatives, this is the same category SayGM competes in - with one structural difference layered on top. ![An image showing the flow from your app via SayGm's TEE privacy enclave through to LLM models and back.](https://assets.saygm.com/blog/2026/08/eead487c-ea6f-4960-bd81-52cc37cb13cf-app-to-gm-to-llm-flow-1902x1066.png) That difference is who can see the data passing through the gateway itself. A typical gateway can technically log or inspect every prompt and completion that flows through it. SayGM is built so that it structurally can't. You can read the full technical picture in [our documentation](https://docs.saygm.com/), including a [quickstart guide](https://docs.saygm.com/getting-started/quickstart/) that gets you from signup to first request in a few minutes. Two Levels of Privacy: Anonymous Routing vs End-to-End TEE ---------------------------------------------------------- "Private" isn't one setting. SayGM actually offers two distinct levels, and it's worth knowing the difference so you can pick the right one for what you're building. Anonymous routing works a bit like a VPN for your prompts. The model provider on the other end (Anthropic, OpenAI, whoever) still sees the content of the request, but it never sees you. Your identity is stripped out before the request reaches them, so you get access to frontier models without linking your usage to your account, your organization, or your IP. TEE end-to-end goes a layer further: both the routing and the model inference happen inside the trusted execution environment, so the prompt itself is provably blind to everyone in the chain - SayGM included, not just the model provider. ![SayGm has two levels of privacy: one where only the maker sees your prompt and one where no one does](https://assets.saygm.com/blog/2026/08/407020e4-53e3-4fc4-a3c1-38313d2d2b1e-saygm-two-levels-of-privacy-1035x576.png) In short: anonymous routing hides who's asking while TEE end-to-end hides what's being asked as well. Which one you reach for depends on whether your concern is identity exposure or content exposure, sometimes it's both. The TEE end-to-end tier is possible because SayGM runs inside Intel TDX confidential virtual machines - a hardware-isolated region of the processor that encrypts and protects code and data while it's actually being used, not just at rest or in transit.[ Intel's own overview of TDX](https://www.intel.com/content/www/us/en/developer/tools/trust-domain-extensions/overview.html) describes it as isolating trust domains from the hypervisor and host operating system entirely, and[ Google Cloud's confidential computing documentation](https://docs.cloud.google.com/confidential-computing/docs/confidential-computing-overview) frames this as the missing third leg of encryption - protecting data in use, alongside encryption at rest and in transit. In practice, that means your prompt is sealed for the entire time it's being processed: SayGM can't read it, not merely that we've promised we won't. You can read more in [our privacy guide](https://docs.saygm.com/security/privacy/). How SayGM Keeps Costs Down -------------------------- Privacy is one half of the pitch. The other is that routing through SayGM shouldn't cost more than calling a provider directly - it should cost less, or the same at worst. On frontier models, SayGM runs about 5% cheaper than OpenRouter, since OpenRouter adds a top-up fee that SayGM doesn't charge. Structurally, this is possible because SayGM's gateway runs on decentralized miner infrastructure rather than a single company's cloud bill - which removes a layer of margin that centralized aggregators typically stack on top of provider list prices. We know cost is a deciding factor for a lot of teams, so there's something coming very soon to push prices down even further and make SayGM the natural choice. One Unified API for OpenAI, Anthropic, and Gemini ------------------------------------------------- None of this requires you to rebuild anything. SayGM supports OpenAI-compatible Chat Completions and Responses, Anthropic's Messages API, and Gemini's generateContent. So if you're already calling one of these APIs, switching is a base URL and an API key change, not a rewrite. That's the whole idea behind a genuinely unified API: one integration surface, multiple model providers behind it, and the freedom to route between them without touching your application code. Our[ integrations page](https://docs.saygm.com/integrations/) covers connecting Claude Code, Codex CLI, Cursor, and other tools you're probably already using. For teams juggling several provider SDKs just to keep options open, this collapses that overhead into a single, drop-in integration -  with the privacy layer built underneath it rather than bolted on top. Built on Decentralized, Verifiable Infrastructure ------------------------------------------------- SayGM's gateway doesn't run on a single company's servers. It runs on [Bittensor](https://www.bittensor.com/), an open network where independent subnets of miners and validators produce digital commodities and get paid in the network's native token, TAO.[ ](https://docs.bittensor.com/) SayGM operates as a subnet (#28) on this network, which means the infrastructure running your requests is permissionless at the network layer (anyone can operate hardware and compete to serve inference) while remaining permissioned at the model layer, since only attested, TEE-verified nodes running approved models are eligible to serve traffic.  Decentralization here isn't a buzzword; it's what makes the attestation model auditable instead of something you just have to take our word for. What's Coming Next ------------------ SayGM was launched about 1 month ago, and 400+ people are already signed up and using the platform. Now that we're live, the features we have coming are: - Processors: Auto-compressing context, PII redaction, and JSON schema enforcement are rolling out from here. - Attested logs: a signed, verifiable audit trail of exactly which model answered which request. - Bring Your Own Key: encryption tied to your own TEE session. - Teams:  one bill, one place to set budgets, an audit log per user - is on the roadmap for when we're serving larger organizations. Further out, we're adding paying gateway fees in Alpha at preferential on-chain rates, a Custom Processors SDK so you can write and sandbox your own request-processing logic inside our TEE, and hosted, attested agents you can spin up directly on our hardware. We'll cover each of these in more depth as they ship. Say GM to AI ------------ If you're building something where the prompt itself is the sensitive part - internal tools, regulated workflows, or just a product you don't want to explain to a security team twice - SayGM gives you a gateway you don't have to take on faith. [Read the docs](https://docs.saygm.com/), grab an API key, and say gm to AI. About SayGM ----------- SayGm is a drop-in inference gateway for teams who don't want to just take a company's word that their prompts are private. Every request runs inside a hardware-verified confidential environment - not even SayGm can see what's inside it. That's not a policy, it's provable. Swap in your existing OpenAI, Anthropic, or Gemini code and you're covered in minutes, at transparent, published rates with no hidden markup.\ Say gm to AI at[ saygm.com](https://saygm.com). [Website](https://saygm.com) | [Twitter](https://x.com/say_gm_) | [Discord](https://discord.com/channels/799672011265015819/1343950080465698836) | [Blog](https://saygm.com/blog) | [Medium](https://medium.com/@say_gm_) | [Docs](https://docs.saygm.com/)