KosCompute Provider Guide

Verification status: this entry is docs- and roster-verified only — not yet live-verified. It was onboarded credential-free: no KosCompute API key was ever created or used. Every field below comes from KosCompute's public docs (https://api.koscompute.com/docs/) and an unauthenticated GET /v1/models call (no auth header). No POST request was ever sent, so no generate, stream, tool-calling or structured-output capability has been exercised end to end. Treat capabilities.tools, .structuredOutput and .thinking as documentation-grade, not wire-proven, until evidence.liveMatrix is filled in by a live run.

KosCompute is a Tier-2 catalog provider: OpenAI-wire-compatible with no behavioural quirks, so its entire integration is one JSON file (src/lib/providers/catalog/koscompute.json) rather than hand-written code. That file is the source of truth for everything on this page.

Key Facts

Provider id: koscompute
Protocol: OpenAI-compatible (/v1/chat/completions); the vendor also exposes a Responses API and an Anthropic-Messages-compatible surface for the same model, neither of which NeuroLink uses here.
Base URL: https://api.koscompute.com/v1
Default model: qwen/qwen3.8-27b
Models in catalog: 1 — the only chat-capable id KosCompute serves. The other 6 ids on the vendor's /v1/models roster (openai/whisper-large-v3, openai/whisper-large-v3-turbo, kosmik/tts-piper-fast, kosmik/tts-kokoro-quality, kosmik/tts-moss-nano, kosmik/tts-supertonic-3) are Whisper transcription and TTS models, not chat models, so they are excluded from this catalog.
Streaming: documented (stream: true, text/event-stream, terminal data: [DONE]) — not live-probed.
Tool calling: documented (tools array, tool_choice auto/none/required/named) and the served model's own supported_features (from the live /v1/models response) list tool_calling — not live-probed.
Structured output: documented (response_format with json_object and json_schema) and the model's supported_features list json_mode and structured_outputs — not live-probed. Combining tools with a JSON schema in one request is not documented either way, so structuredOutputWithTools is declared false until a combined probe is run.
Embeddings: not documented anywhere in the vendor's docs or roster.
Vision: the default model accepts image input (input_modalities: ["text","image"], vision.max_pixels: 1048576) per the live roster and the vendor's vision guide (HTTPS image URLs and data URLs) — not live-probed.
Reasoning / thinking: the model reports reasoning: true, extended_thinking: true, thinking_default: true, max_thinking_tokens: 8192, and supported_reasoning_efforts: ["none", "low", "medium", "xhigh"] on the live roster.
Billing: unconfirmed. No pricing, free-tier or top-up page is published; the docs only mention an account "balance" in passing. This entry sets billingPolicy: "no-free-tier" as the conservative default in the absence of any documented free allocation — not as a confirmed vendor policy.
Signup: no public signup, dashboard, or key-management page is published. The site's only call-to-action ("Start building") points at the docs quickstart, which itself assumes you already hold a key.
Key format: none declared.

Quick Start

1. Get an API key

KosCompute does not publish a public signup or key-issuance flow. The quickstart docs assume the key already exists:

Obtain a key out of band (no public signup page exists as of 2026-09-28).
Confirm current billing terms directly with KosCompute — none are public.
Set KOSCOMPUTE_API_KEY in your .env file.
Optional sanity check before spending a real request: curl https://api.koscompute.com/v1/models needs no key and returns the live roster.

2. Configure

export KOSCOMPUTE_API_KEY=your-api-key
export KOSCOMPUTE_MODEL=qwen/qwen3.8-27b      # optional — overrides the default model
export KOSCOMPUTE_BASE_URL=https://api.koscompute.com/v1   # optional — self-hosted or proxy

3. Use it

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

const neurolink = new NeuroLink();

const result = await neurolink.generate({
  input: { text: "Explain speculative decoding in one paragraph." },
  provider: "koscompute",
  model: "qwen/qwen3.8-27b",
});

console.log(result.content);

# CLI
npx @juspay/neurolink generate "Hello" --provider koscompute

Per-request credentials work as they do for every provider:

await neurolink.generate({
  input: { text: "Hello" },
  provider: "koscompute",
  credentials: { koscompute: { apiKey: process.env.KOSCOMPUTE_API_KEY } },
});

Models

Model	Context	Vision	$/M in · out	Notes
`qwen/qwen3.8-27b` ⭐	262,144 (text) / 131,072 (image)	yes	$0.35 / $2.20 (cached $0.09)	Only chat-capable model in the roster. Reasoning-capable (`extended_thinking`, up to 8,192 thinking tokens, effort levels none/low/medium/xhigh). Max output 32,768 tokens.

Context window and max output shown are the catalog defaults (262,144 / 32,768 tokens), taken directly from the model's own entry in the live, unauthenticated GET /v1/models response (context_length, max_output_length) on 2026-09-28. Pricing is that same response's precomputed per-MTok fields (input, output, cache_read).

Fallback order: none — qwen/qwen3.8-27b is the only chat model KosCompute serves, so it is both the default and its own (schema-exempt) single-entry fallback list.

Verification status

Tier-2 onboarding requires evidence before a provider is accepted, and pnpm run verify:provider-onboarding gates it in CI. This is what the catalog records for KosCompute — and, just as importantly, what it does not yet record:

Probe	Result
Roster	unauthenticated GET `/v1/models`, HTTP 200, 7 models, 2026-09-28. No API key was used or required for this call.
Auth rejection	Not probed. Would require a `POST` with an invalid key, which this credential-free onboarding pass does not send.
Live capability sweep	Not run. `evidence.liveMatrix` is `null`. `capabilities.tools`, `.structuredOutput`, `.toolsWithStreaming` and `.thinking` are set from the vendor's docs (tool-calling, structured-output and streaming guides) plus the served model's own `supported_features` array on the live roster — not from an executed chat/stream/tool-call/schema request. `structuredOutputWithTools` is `false` because no combined `tools` + `response_format` request was — or could be — attempted without a key.

Until a live run fills in evidence.liveMatrix, treat this provider as documentation-grade: the wire shape (base URL, auth header, request/response field names) is well-documented and consistent with the standard OpenAI-compatible contract, but no NeuroLink call has actually exercised it.

Troubleshooting

Symptom	Cause	Fix
`Invalid KosCompute API key`	`KOSCOMPUTE_API_KEY` unset, wrong, expired, or lacks access	KosCompute's docs describe 401 as "missing, invalid, expired, inactive, or conflicting credentials" but publish no example error body or code string
Model not found	Wrong model id, or the roster changed since 2026-09-28	Pick a current id from an unauthenticated `GET https://api.koscompute.com/v1/models`
429 with `concurrent_limit_exceeded`	Per-key concurrency limit exceeded	Wait for an in-flight request to finish, or reduce parallel requests, before retrying
No signup page found	Expected — KosCompute publishes no public signup or key-management page	Obtain a key out of band; there is no self-serve console to link to

Key Facts​

Quick Start​

1. Get an API key​

2. Configure​

3. Use it​

Models​

Verification status​

Troubleshooting​

See also​