GM Wants to Be Bittensor’s Private Door to Every Frontier Model

Subnet 28, branded “gm” (@say_gm_ on X), published its roadmap, and that announcement tells you exactly what kind of subnet it’s trying to be.

Three cards: Good Morning, Good Afternoon, Good Night. Now, Next, Later.

The vibe feels less like a protocol spec and more like a product launch, which is the point. GM is one of the few subnets built for people who don’t care that it’s a subnet.

The subnet is aiming to become the confidential gateway to Claude, GPT, and Gemini, routed through a Trusted Execution Environment (TEE) so prompts and responses stay private and verifiable.

It gives OpenRouter’s unified access plus Venice-style privacy, wired into Bittensor’s incentive economy with on-chain payments, staking, and miner competition.

GM is Now Live on Testnet

Two things are currently in motion.

First, drop-in compatible APIs: GM speaks the OpenAI, Anthropic, and Gemini formats, so existing code switches over with minimal changes. Point your client at GM and go without SDK rewrite or new mental model.

Second, hardware verification from day one: the gateway runs inside a TEE on testnet right now. This is the load-bearing claim. Privacy isn’t a policy promise you have to trust but it’s enforced by a secure hardware enclave, the same class of approach (Intel TDX/SGX, confidential-compute GPUs) that research has shown can run full LLM inference at single-digit throughput overhead. The data path is private because the hardware won’t let it be otherwise.

Next is Closed Beta on Mainnet

This is where GM moves from “private proxy” to “platform.” Four features:

• Attested Logs: a signed audit log of which model answered what. Cryptographically verifiable, so you can prove after the fact that your request hit the model you paid for.
• Processors: opt-in, per-API-key middleware that auto-compresses context, redacts PII, and enforces JSON schemas before requests leave the enclave.
• Teams: one bill, one place to set budgets, per-user audit logs. The unglamorous feature that makes a tool usable inside an organization.
• Bring Your Own Key: encrypt your data with a key only your TEE session can use. GM’s own framing: they can’t read it.

Later: Open to Everyone

The endgame leans fully into Bittensor’s economy:

• Pay with Alpha: lock the subnet’s alpha token to prepay gateway fees at preferential rates, on-chain. This is the mechanism that ties usage back to the token instead of leaving it as a speculative side-asset.
• Custom Processors SDK: write your own processors, sandboxed inside the TEE and attested per request. The closed processor list from the beta phase becomes an open surface.
• Hosted Agents: spin up tools like Hermes or OpenClaw on GM’s hardware, attested per request.

Why It Matters

You need to have a PhD in ML before you can explain what some subnets do. GM doesn’t. The progression (works-with-your-code today, enterprise controls next, token-native and extensible later) is a normal SaaS arc that’s quickly understandable to anyone.

GM’s TEE-enforced privacy is another edge that’s widely talked about in the space. With a disclosed roadmap, the community awaits how far the subnet can go from here.

Testnet is currently live, closed mainnet beta next, waitlist open at saygm.com.