{
    "$schema": "./provider-catalog.schema.json",
    "id": "laya",
    "displayName": "Laya",
    "aliases": [],
    "inferenceKinds": ["decide"],
    "wire": {},
    "models": {
        "default": "typed-decisions",
        "topModels": ["typed-decisions", "multilingual", "english", "auto"],
        "catalog": {
            "typed-decisions": {
                "contextWindow": 1024,
                "maxOutputTokens": 256,
                "vision": false,
                "status": "production",
                "description": "Recommended - Laya fine-tuned for typed decisions"
            },
            "english": {
                "contextWindow": 512,
                "maxOutputTokens": 256,
                "vision": false,
                "status": "production",
                "description": "Laya, English base checkpoint (512-token window)"
            },
            "multilingual": {
                "contextWindow": 1024,
                "maxOutputTokens": 256,
                "vision": false,
                "status": "production",
                "description": "Laya, 100+ languages"
            },
            "auto": {
                "contextWindow": 512,
                "maxOutputTokens": 256,
                "vision": false,
                "status": "production",
                "description": "Let Laya pick a checkpoint by language"
            }
        }
    },
    "timeouts": { "decideMs": 5000 },
    "decisionLimits": {
        "maxStateTokens": 320,
        "maxQuestions": 64,
        "nonAsciiTokensPerChar": 1.5,
        "models": {
            "typed-decisions": { "maxStateTokens": 768 },
            "multilingual": { "maxStateTokens": 768, "nonAsciiTokensPerChar": 0.6 },
            "english": { "maxStateTokens": 320 },
            "auto": { "maxStateTokens": 320 }
        }
    },
    "decide": {
        "dialect": "system-one",
        "order": 2,
        "baseURLRequired": true,
        "baseURLNormalize": ["strip-trailing-slash"],
        "endpoint": "{baseURL}/predict",
        "model": { "placement": "body-unless-alias", "alias": "auto" },
        "requestIdHeaders": ["x-litellm-call-id"],
        "response": { "modelPaths": ["routing.model", "model"] },
        "errors": {
            "rules": [
                { "status": [401, 403], "kind": "authentication" },
                { "status": 413, "kind": "max_tokens_exceeded" },
                { "status": 429, "kind": "rate_limit" },
                { "status": 503, "kind": "overloaded" },
                { "statusMin": 500, "kind": "server" }
            ],
            "default": "invalid_request",
            "bodyShapes": [
                { "path": "error.message" },
                {
                    "path": "detail",
                    "isArray": true,
                    "formatter": "fastapi-validation",
                    "kind": "invalid_request",
                    "retryable": false
                },
                { "path": "detail" }
            ]
        }
    },
    "setup": {
        "url": "https://github.com/NandhaKishorM/laya",
        "apiKeyFormat": null,
        "billingPolicy": "free-tier",
        "description": "API key for your Laya endpoint",
        "instructions": [
            "1. Run a Laya server, or use a LiteLLM proxy with a pass-through route to one",
            "2. Set LAYA_BASE_URL to its base URL (NeuroLink calls <LAYA_BASE_URL>/predict)",
            "3. Set {apiKeyEnvVar} to the key it accepts (on LiteLLM, a virtual key with that route in its Allowed Routes)"
        ]
    },
    "evidence": {
        "rosterVerified": {
            "date": "2026-09-24",
            "method": "Checkpoint ids typed-decisions, multilingual and english as listed in docs/getting-started/providers/laya.md and answered by a live Laya 0.3.5 server (routing.model); auto sends no model and lets Laya's router choose"
        },
        "liveMatrix": {
            "date": "2026-09-24",
            "result": "A 60,000-character state was answered 200 on a 1,024-token window, so Laya truncates silently and decisionLimits refuses over-window input (768 state tokens on typed-decisions and multilingual, 320 on english and auto). Chinese measured about 1.4 tokens per character on the English tokenizer and 0.56 on multilingual."
        },
        "addedInPR": "https://github.com/juspay/neurolink/pull/1978"
    }
}