Subconscious is live as a provider on Vercel AI Gateway. If you already route through the gateway, you can switch to GLM 5.3 and DeepSeek V4.1 Flash on our inference system with one line of code.
We designed the Subconscious inference system to be excellent at long-running agentic tasks like coding and cybersecurity. The system uses runtime KV cache compression and custom speculative decoders to make models faster and much more accurate on sessions that consume more than 200k tokens. If you're building a long-running agent, you should try out Subconscious.
We bill only on the tokens our system processes after compression, so a request that sends 1M tokens will only charge for 200k cached tokens at most. Keep that in mind when you compare per-token prices on the gateway. We're often the lowest on a per task basis.
Try it
Call GLM 5.3 or DeepSeek V4.1 Flash through the gateway and set Subconscious as the provider. With the AI SDK it takes one option:
import { generateText } from 'ai';
const { text } = await generateText({
model: 'zai/glm-5.3',
prompt: 'Plan the refactor, then start on step one.',
providerOptions: {
gateway: {
order: ['subconscious'], // One line to use Subconscious!
},
},
});The keyword order chooses Subconscious first with other providers behind us as fallback. Vercel's docs cover provider ordering in more depth. Our provider page lists current models and prices.
Models on the gateway have a listed 1M-token context window, but can process up to 5M tokens thanks to our context compression.
I've been a user and vocal proponent of Vercel since 2020, built my last company on Vercel, and even run the Subconscious platform on Vercel today. I'm incredibly proud to partner with the team, and I hope you take advantage of better inference for agents.
