{
    "$schema": "./provider-catalog.schema.json",
    "id": "ambient",
    "displayName": "Ambient",
    "aliases": [],
    "tier": 2,
    "wire": {
        "baseURL": "https://api.ambient.xyz/v1"
    },
    "models": {
        "default": "ambient/large",
        "fallbacks": ["z-ai/glm-5.2", "qwen/qwen3.6-27b", "qwen/qwen3.8-27b"],
        "defaultContextWindow": 202752,
        "defaultMaxOutputTokens": 202752,
        "catalog": {
            "ambient/large": {
                "contextWindow": 202752,
                "maxOutputTokens": 202752,
                "pricingPerMTok": {
                    "input": 0.6,
                    "output": 2.0,
                    "cachedInput": 0.15
                },
                "vision": false,
                "status": "production",
                "description": "Ambient's own recommended default — the model id used in every example on the vendor's docs (the plain chat-completions curl example and the streaming curl example both send model: \"ambient/large\"). Documented as \"an Ambient-managed alias rather than a distinct model. Its specs mirror whatever concrete model it currently points to\" (https://ambient.xyz/docs/build/models, retrieved 2026-09-28); today it mirrors z-ai/glm-5.2 exactly — identical context_length, max_output_length and pricing in the unauthenticated GET https://api.ambient.xyz/v1/models response taken 2026-09-28. Text-only (input_modalities: [\"text\"]); no vision. The live roster's supported_features for this id are tools, json_mode, structured_outputs, reasoning, logprobs — this is the vendor's own server-reported metadata, not a live POST probe (none was possible credential-free), so capabilities.tools and capabilities.structuredOutput are marked true on that basis alone, while the combined capabilities (tools+response_format in one request, tool-call deltas under streaming) are unverified and both structuredOutputWithTools and toolsWithStreaming are left false pending a live-keyed probe. maxOutputTokens equals the context window in the vendor's own catalog entry (202,752 tokens each, i.e. \"198K\" as rounded on the docs models page)."
            },
            "z-ai/glm-5.2": {
                "contextWindow": 202752,
                "maxOutputTokens": 202752,
                "pricingPerMTok": {
                    "input": 0.6,
                    "output": 2.0,
                    "cachedInput": 0.15
                },
                "vision": false,
                "status": "production",
                "description": "The concrete underlying model that ambient/large currently aliases. The vendor's docs models page states \"GLM 5.2 ... is the supported production model\" (https://ambient.xyz/docs/build/models, retrieved 2026-09-28); other catalog ids may be previews with no guarantees. Text-only, no vision. Same live-roster numbers (unauthenticated GET https://api.ambient.xyz/v1/models, 2026-09-28) and the same unverified-combined-capability caveats as ambient/large — kept as a fallback pinned to a concrete id in case the ambient/large alias target changes underneath it."
            },
            "qwen/qwen3.6-27b": {
                "contextWindow": 32768,
                "maxOutputTokens": 8192,
                "pricingPerMTok": {
                    "input": 0.32,
                    "output": 3.2,
                    "cachedInput": 0.16
                },
                "vision": true,
                "status": "preview",
                "description": "Qwen3.6 27B. The live roster's input_modalities are [\"text\",\"image\"] and the vendor's own description reads \"natively multimodal (text + image) dense 27B, hybrid attention. Served FP8 at 32k context, thinking off.\" (unauthenticated GET https://api.ambient.xyz/v1/models, 2026-09-28). The vendor's docs models page marks non-GLM-5.2 entries as possible previews served with no guarantees, so status is preview here. vision: true is taken from input_modalities; set as models.visionModel since the default (ambient/large) is text-only. Same unverified-combined-capability caveats as the other entries (no live POST probe was possible)."
            },
            "qwen/qwen3.8-27b": {
                "contextWindow": 32768,
                "maxOutputTokens": 8192,
                "pricingPerMTok": {
                    "input": 0.32,
                    "output": 3.2,
                    "cachedInput": 0.16
                },
                "vision": true,
                "status": "preview",
                "description": "Qwen3.8 27B. The live roster's own description calls it \"natively multimodal (text + image + video) dense 27B, Apache 2.0, hybrid attention\" but input_modalities lists only [\"text\",\"image\"] — no \"video\" entry — so vision: true here is based on input_modalities, not the prose \"video\" claim, and no video-input support is declared. Same preview status and unverified-combined-capability caveats as qwen3.6-27b (unauthenticated GET https://api.ambient.xyz/v1/models, 2026-09-28)."
            }
        },
        "topModels": [
            "ambient/large",
            "z-ai/glm-5.2",
            "qwen/qwen3.6-27b",
            "qwen/qwen3.8-27b"
        ],
        "visionModel": "qwen/qwen3.6-27b"
    },
    "capabilities": {
        "text": true,
        "streaming": true,
        "tools": true,
        "toolsWithStreaming": false,
        "structuredOutput": true,
        "structuredOutputWithTools": false,
        "embeddings": false,
        "thinking": true
    },
    "errorRules": [
        {
            "status": 401,
            "class": "authentication",
            "message": "Invalid Ambient API key. Check {apiKeyEnvVar}. Get one at {setupUrl}"
        },
        {
            "status": 400,
            "pattern": "Unknown model",
            "class": "invalid-model",
            "message": "Ambient model '{model}' not found. Pick a current model from the unauthenticated GET https://api.ambient.xyz/v1/models roster or {setupUrl}"
        }
    ],
    "setup": {
        "url": "https://app.ambient.xyz",
        "apiKeyFormat": null,
        "billingPolicy": "no-free-tier",
        "instructions": [
            "1. Visit: https://app.ambient.xyz and sign in",
            "2. Click \"Get API key\" to create a key",
            "3. Ambient has no free tier — the cheapest option is the $1/mo Starter plan (https://ambient.xyz/pricing); confirm current billing terms before use",
            "4. Set {apiKeyEnvVar} in your .env file"
        ],
        "description": "OpenAI- and Anthropic-compatible endpoint at https://api.ambient.xyz/v1 (vendor's own docs: \"Point your OpenAI or Anthropic SDK at Ambient and get verified inference in minutes, no rewrite required\"). Pay-per-token API billing; no free tier. Inference is cryptographically verified via the vendor's \"Proof of Logits\" protocol."
    },
    "evidence": {
        "rosterVerified": {
            "date": "2026-09-28",
            "method": "unauthenticated GET https://api.ambient.xyz/v1/models — 4 models; credential-free, no key used",
            "status": 200
        },
        "liveMatrix": null,
        "addedInPR": "https://github.com/juspay/neurolink/pull/1848"
    }
}