Skip to main content

SubQ Provider Guide

A provider of decide — the same typed boolean / choice / score answers as TypeSafe's Jev, Laya, XOR, Perplexity and Cloudflare Clef, from Subquadratic's hosted SubQ Snap. It emits no text at all.

Overview​

subq-snap-preview is Subquadratic's decision model, in public preview. You send one state plus named, typed questions. The model answers every question in a single pass and returns a typed answer for each, with a probability instead of a sentence. SubQ speaks the same System One wire as TypeSafe, so NeuroLink describes it with one catalog file, src/lib/providers/catalog/subq.json, and no provider code of its own. NeuroLink calls SubQ's public endpoint, so a key alone configures it: there is no base URL to set.

Key Facts​

Provider idsubq; the aliases subquadratic and subq-snap name it too
Inference typedecide only
Modelsubq-snap-preview, the only model on SubQ's Decisions endpoint. SUBQ_MODEL sets it
KeySUBQ_API_KEY, or credentials.subq.apiKey. Keys come from console.subq.ai; the console shows one only once
Endpointhttps://api.subq.ai/v1/systemone; SUBQ_BASE_URL or credentials.subq.baseURL can name another origin
Questions per requestno cap declared
State window376,000 estimated tokens, advisory: reported by decisionLimits(), never enforced. SubQ's own 400 response states a maximum model length of 393,216 tokens; see Limits
Medianone. Images attached to a decide() call are refused before any request; see Images
Default timeout5 seconds; timeoutMs or --timeout replaces it
Precedenceused automatically unless TypeSafe, Laya or XOR is configured; it comes ahead of Perplexity, Cloudflare Clef, OpenAI Decisions and ScaleDown

Quick Start​

1. Get a key​

Sign up at console.subq.ai and create an API key. The console shows it once, so copy it then. SubQ answers HTTP 402 to a request made with no credits left.

2. Configure​

Set the key in the environment:

export SUBQ_API_KEY=...                           # the key from the console
export SUBQ_MODEL=subq-snap-preview               # optional
export SUBQ_BASE_URL=https://api.subq.ai          # optional; an origin, without /v1

or in the config passed to the SDK, exactly as for any other provider. Values passed per call override the constructor's, which override the environment:

const neurolink = new NeuroLink({
  credentials: {
    subq: { apiKey: process.env.MY_SUBQ_KEY },
  },
});

SUBQ_BASE_URL and credentials.subq.baseURL name an origin. NeuroLink appends /v1/systemone and drops any trailing slashes, so a value that already ends in /v1 would send the request to /v1/v1/systemone.

3. Use it​

import { NeuroLink, readDecisionChoice } from "@juspay/neurolink";

const neurolink = new NeuroLink();
const result = await neurolink.decide({
  provider: "subq",
  state: "We were billed twice for March. Please refund it today.",
  questions: {
    team: {
      type: "choice",
      instructions: "Which team should handle this?",
      criteria: {
        billing: "Payments and refunds",
        technical: "Bugs",
        sales: "Pricing",
      },
    },
    urgent: { type: "boolean", instructions: "Is this urgent?" },
  },
});
const team = readDecisionChoice(result.answers, "team");

From the CLI​

neurolink decide "We were billed twice for March." --provider subq \
  --questions '{"urgent":{"type":"boolean","instructions":"Is this urgent?"}}'

Every built-in consumer of decide — model routing, relevance-driven compaction, tool routing and RAG planning — asks for the default decision provider. That is the first one that is configured, in the environment or in the credentials passed to the SDK, in the order TypeSafe, Laya, XOR, SubQ, Perplexity, Cloudflare Clef, OpenAI Decisions, then ScaleDown. SubQ counts with its key alone. It sits ahead of Perplexity, Clef and OpenAI Decisions because its key belongs to no other provider, so setting it can only be meant for SubQ, whereas theirs are shared with text providers and must not displace it. ScaleDown, whose key is also its own, was placed last on purpose. So:

  • Only a SubQ key: built-in features use SubQ.
  • A SubQ key, plus a Perplexity key, the Cloudflare pair, OPENAI_API_KEY or SCALEDOWN_API_KEY: built-in features use SubQ; Perplexity, Cloudflare Clef, OpenAI Decisions and ScaleDown run only where a caller asks for provider: "perplexity-decider", provider: "cloudflare-clef", provider: "openai-decisions" or provider: "scaledown".
  • A SubQ key, plus TypeSafe's key, or Laya's or XOR's key and base URL: built-in features use TypeSafe, Laya or XOR, in that order, whichever is configured; SubQ runs only where a caller asks for provider: "subq".
  • A blank SUBQ_API_KEY: no key. SubQ is not configured.

Limits​

The state window is advisory. decisionLimits() reports 376,000 estimated state tokens, and the catalog records a model contextWindow of 393,216. Both come from one live observation made on 2026-10-13 for NeuroLink (not by this change): SubQ refused a request of 399,795 tokens with HTTP 400, and the message names a maximum model length of 393,216 tokens; the largest request it answered had 376,180 input tokens. So 393,216 is the maximum model length SubQ's own 400 response states, and 376,000 is a margin under the largest request it answered. Both are advisory: a longer state is sent, the server decides, and neither figure is a guarantee. The 376,000 counts NeuroLink's estimated tokens, which can differ from the tokens SubQ counts, so a state near it can still be refused.

SubQ answers an over-length request with HTTP 400 and a validation_error body (The decoder prompt (length 399795) is longer than the maximum model length of 393216 ...). NeuroLink reports it as invalid_request, not retried; it is not reported as max_tokens_exceeded.

No question cap is declared. SubQ's documentation gives no count; a choice question takes 2 to 255 options, a score question 2 to 10 levels, and a question name up to 100 characters. SubQ refuses a request outside those with HTTP 422, which NeuroLink reports as invalid_request, with SubQ's reason when the body is the documented object form.

Timeout. The default is 5 seconds per attempt. A state near the window takes longer to read than a short one, so size timeoutMs for the states you send.

Price. SubQ's documentation gives $0.05 per million input tokens and no charge for output tokens (read on 2026-10-13; check the console for the current rate). Each response carries usage.input_tokens, which decide() returns as usage.inputTokens.

Images​

SubQ's documentation describes image input: up to 4 images per request, each an object with a mediaType and a one-line base64 string, with no data: prefix. NeuroLink's decide() prepares images as data: URLs, and a catalog entry cannot yet describe the object form, so SubQ's entry declares no media. An image attached to a decide() call for SubQ is refused before any request, as invalid_request, and the catalog marks the model vision: false. To decide over images today, use XOR, Perplexity or Cloudflare Clef.

Errors​

SubQ's documentation gives an error body of {"detail": {"error_type": "…", "message": "…"}}. NeuroLink shows the message, or a detail that is a plain string. A body in any other form, such as a detail that is an array, gives only SubQ request failed with HTTP <status>; the kind and the retry behaviour do not change.

ReplyKindRetried
401authentication — the provider instance stops retryingno
402invalid_request — the account has no creditsno
403invalid_request — the key cannot use the modelno
400invalid_request, with SubQ's reason (for example a prompt past the maximum model length)no
422invalid_request, with SubQ's reason (documented body form)no
429rate_limityes
503overloadedyes
5xxserveryes
otherinvalid_requestno
imageinvalid_request, with no network call (see Images)no

Only a 401 stops the instance. A 402 is not treated as a rejected key on purpose: after you add credit, the same instance works again, with no restart. An error message never carries the API key; if SubQ echoes it, NeuroLink replaces it before the message reaches a log or an error.

Data retention​

Check SubQ's data retention page before sending anything sensitive. SubQ's documentation, as read on 2026-10-13 (a reading made before this page was written, and not repeated for it), says that prompts are kept indefinitely unless the organization sets otherwise, and that Decisions are unaffected by zero data retention. Check the page for the current terms.

What has and has not been checked​

  • Read on 2026-10-13, before this page was written and not repeated for it: SubQ's documentation (the endpoint, the error body shape, the limits on questions and images, the price and the retention terms), and an unauthenticated GET https://api.subq.ai/v1/models that listed subq-snap-preview.
  • One decision was captured: a single request from SubQ's documentation, answered HTTP 200. Its body is the contract fixture's success body, test/fixtures/decide/subq/contract.json.
  • Observed live on 2026-10-13, in a run made for NeuroLink and not repeated for this page: the over-length 400 above (its body is in the fixture as a seen-live row; the saved text of the message stops after "number of text token") and the largest request answered, 376,180 input tokens.
  • Not captured: any other error body. Every other error in that fixture is a stand-in written from the shape the documentation describes, and is labelled so. Whether a 422 ever carries an array detail is unknown.
  • Not measured: a question cap, and what an image costs. The catalog entry records no live matrix.

Troubleshooting​

  • SubQ requires an API key — set SUBQ_API_KEY, or pass credentials.subq.apiKey. A blank value counts as unset.
  • Authentication failed — SubQ answered 401. Check the key in the console, then build a new NeuroLink instance: an instance does not retry after a rejection.
  • A 402 — the account has no credits. The message says so; add credit and call again.
  • A 403 — the key cannot use the model. The message says why.
  • A request to /v1/v1/systemone — SUBQ_BASE_URL ends in /v1. It should be the origin, https://api.subq.ai.
  • Built-in routing never uses SubQ — another decision provider is also configured and takes precedence.

See also​