@ailib-official/ai-protocol 0.8.4 → 1.0.0

package/v2/providers/nvidia.yaml

@@ -0,0 +1,405 @@
+# NVIDIA NIM V2 provider — OpenAI-compatible chat completions (Eos default aggregator)
+# Last Updated: 2026-06-26 (add tool_calling)
+$schema: "https://raw.githubusercontent.com/ailib-official/ai-protocol/main/schemas/v2/provider.json"
+
+id: nvidia
+protocol_version: "2.0"
+name: "NVIDIA NIM"
+version: "1.0.0"
+status: stable
+category: third_party_aggregator
+official_url: "https://build.nvidia.com/explore/discover"
+support_contact: "https://docs.api.nvidia.com/nim/docs"
+
+endpoint:
+  base_url: "https://integrate.api.nvidia.com/v1"
+  chat: "/chat/completions"
+  auth:
+    type: "bearer"
+    header: "Authorization"
+    prefix: "Bearer"
+    token_env: "NVIDIA_API_KEY"
+
+error_classification:
+  by_http_status:
+    "400": "invalid_request"
+    "401": "authentication"
+    "403": "permission_denied"
+    "404": "not_found"
+    "429": "rate_limited"
+    "500": "server_error"
+    "503": "overloaded"
+  by_error_code:
+    "context_length_exceeded": "request_too_large"
+    "model_not_found": "not_found"
+    "rate_limit_exceeded": "rate_limited"
+
+availability:
+  required: false
+  regions:
+    - global
+  check:
+    method: GET
+    path: "/models"
+    expected_status: [200, 401]
+    timeout_ms: 5000
+
+capabilities:
+  required:
+    - text
+    - streaming
+    - tools
+  optional:
+    - vision
+    - parallel_tools
+    - agentic
+    - reasoning
+  feature_flags:
+    parallel_tool_calls: true
+    streaming_usage: true
+    system_messages: true
+
+  tool_calling:
+    native:
+      supported: true
+      reliability: "full"
+      parallel: true
+
+parameters:
+  temperature: { type: float, range: [0.0, 2.0], default: 1.0 }
+  max_tokens: { type: integer, min: 1, max: 8192 }
+  top_p: { type: float, range: [0.0, 1.0] }
+  stream: { type: boolean }
+
+streaming:
+  decoder:
+    format: "sse"
+    strategy: "openai_chat"
+    done_signal: "[DONE]"
+    prefix: "data: "
+  content_path: "$.choices[0].delta.content"
+  tool_call_path: "$.choices[0].delta.tool_calls"
+  usage_path: "$.usage"
+  event_map:
+    - match: "exists($.choices[*].delta.content)"
+      emit: "PartialContentDelta"
+      fields:
+        content: "$.choices[*].delta.content"
+    - match: "exists($.choices[*].delta.tool_calls)"
+      emit: "PartialToolCall"
+      fields:
+        tool_calls: "$.choices[*].delta.tool_calls"
+    - match: "exists($.usage)"
+      emit: "Metadata"
+      fields:
+        usage: "$.usage"
+    - match: "$.choices[*].finish_reason != null"
+      emit: "StreamEnd"
+      fields:
+        finish_reason: "$.choices[*].finish_reason"
+  stop_condition: "$.choices[0].finish_reason != null"
+
+multimodal:
+  input:
+    vision:
+      supported: true
+      formats: [jpeg, png, gif, webp]
+      encoding_methods: [base64_inline, url]
+    audio:
+      supported: false
+    video:
+      supported: false
+  output:
+    text: true
+    audio:
+      supported: false
+    image:
+      supported: false
+
+api_families: ["chat_completions"]
+default_api_family: "chat_completions"
+
+endpoints:
+  chat:
+    path: "/chat/completions"
+    method: "POST"
+    adapter: "openai"
+
+services:
+  list_models:
+    path: "/models"
+    method: "GET"
+    response_binding: "data"
+
+retry_policy:
+  strategy: "exponential_backoff"
+  max_retries: 3
+  min_delay_ms: 1000
+  max_delay_ms: 30000
+  jitter: "full"
+  retry_on_http_status: [429, 500, 502, 503]
+
+termination:
+  source_field: "finish_reason"
+  mapping:
+    stop: "end_turn"
+    length: "max_tokens"
+    tool_calls: "tool_use"
+
+metadata:
+  api_compatibility: openai
+  models:
+    # NVIDIA Nemotron Series
+    nvidia/nemotron-3-ultra-550b-a55b:
+      context_window: 1000000
+      max_output_tokens: 8192
+      verification:
+        status: operational
+        source: provider_catalog
+        notes: "Flagship 550B MoE, 1M context, agentic reasoning"
+    nvidia/nemotron-3-super-120b-a12b:
+      context_window: 1000000
+      max_output_tokens: 8192
+      verification:
+        status: operational
+        source: provider_catalog
+        notes: "120B MoE, 1M context, agentic reasoning"
+    nvidia/nemotron-3-nano-30b-a3b:
+      context_window: 1000000
+      max_output_tokens: 8192
+      verification:
+        status: operational
+        source: provider_catalog
+        notes: "30B MoE, 1M context, efficient reasoning"
+    nvidia/nemotron-3-nano-omni-30b-a3b-reasoning:
+      context_window: 131072
+      max_output_tokens: 8192
+      verification:
+        status: operational
+        source: provider_catalog
+        notes: "Omni-modal (image/video/audio/text) reasoning"
+    nvidia/nemotron-4-340b-instruct:
+      context_window: 128000
+      max_output_tokens: 8192
+      verification:
+        status: operational
+        source: provider_catalog
+    nvidia/llama-3.3-nemotron-super-49b-v1:
+      context_window: 128000
+      max_output_tokens: 8192
+      verification:
+        status: operational
+        source: provider_catalog
+        notes: "Llama 3.3 Nemotron Super 49B chat model"
+    nvidia/llama-3.3-nemotron-super-49b-v1.5:
+      context_window: 128000
+      max_output_tokens: 8192
+      verification:
+        status: operational
+        source: provider_catalog
+        notes: "Updated v1.5 with improved reasoning"
+    nvidia/llama-3.1-nemotron-70b-instruct:
+      context_window: 128000
+      max_output_tokens: 8192
+      verification:
+        status: operational
+        source: provider_catalog
+    nvidia/llama-3.1-nemotron-ultra-253b-v1:
+      context_window: 128000
+      max_output_tokens: 8192
+      verification:
+        status: operational
+        source: provider_catalog
+        notes: "Nemotron fine-tuned Llama 3.1 253B MoE"
+    nvidia/llama3-chatqa-1.5-70b:
+      context_window: 32768
+      max_output_tokens: 4096
+      verification:
+        status: operational
+        source: provider_catalog
+    nvidia/nemotron-nano-12b-v2-vl:
+      context_window: 131072
+      max_output_tokens: 4096
+      verification:
+        status: operational
+        source: provider_catalog
+        notes: "Vision-language model for image/video understanding"
+
+    # Embedding Models
+    nvidia/llama-nemotron-embed-1b-v2:
+      context_window: 512
+      max_output_tokens: 0
+      verification:
+        status: operational
+        source: provider_catalog
+    nvidia/nv-embed-v1:
+      context_window: 512
+      max_output_tokens: 0
+      verification:
+        status: operational
+        source: provider_catalog
+
+    # DeepSeek
+    deepseek-ai/deepseek-v4-flash:
+      context_window: 1000000
+      max_output_tokens: 8192
+      verification:
+        status: operational
+        source: provider_catalog
+        notes: "284B MoE, 13B activated, 1M context"
+    deepseek-ai/deepseek-v4-pro:
+      context_window: 1000000
+      max_output_tokens: 8192
+      verification:
+        status: operational
+        source: provider_catalog
+        notes: "1.6T MoE, 49B activated, 1M context"
+
+    # Mistral
+    mistralai/mistral-large-3-675b-instruct-2512:
+      context_window: 128000
+      max_output_tokens: 8192
+      verification:
+        status: operational
+        source: provider_catalog
+        notes: "Mistral flagship 675B MoE VLM"
+    mistralai/mistral-small-4-119b-2603:
+      context_window: 256000
+      max_output_tokens: 8192
+      verification:
+        status: operational
+        source: provider_catalog
+    mistralai/mistral-medium-3.5-128b:
+      context_window: 128000
+      max_output_tokens: 8192
+      verification:
+        status: operational
+        source: provider_catalog
+    mistralai/ministral-14b-instruct-2512:
+      context_window: 256000
+      max_output_tokens: 8192
+      verification:
+        status: operational
+        source: provider_catalog
+    mistralai/codestral-22b-instruct-v0.1:
+      context_window: 256000
+      max_output_tokens: 8192
+      verification:
+        status: operational
+        source: provider_catalog
+        notes: "Coding-specialized 22B"
+
+    # Qwen
+    qwen/qwen3.5-397b-a17b:
+      context_window: 262144
+      max_output_tokens: 81920
+      verification:
+        status: operational
+        source: provider_catalog
+        notes: "Qwen3.5 flagship 397B MoE, 17B activated, vision-language"
+    qwen/qwen3.5-122b-a10b:
+      context_window: 262144
+      max_output_tokens: 81920
+      verification:
+        status: operational
+        source: provider_catalog
+    qwen/qwen3-next-80b-a3b-instruct:
+      context_window: 262144
+      max_output_tokens: 81920
+      verification:
+        status: operational
+        source: provider_catalog
+
+    # Moonshot / Kimi
+    moonshotai/kimi-k2.6:
+      context_window: 262144
+      max_output_tokens: 98304
+      verification:
+        status: operational
+        source: provider_catalog
+        notes: "1T MoE, 32B activated, agentic coding, vision"
+
+    # Z.ai / GLM
+    z-ai/glm-5.1:
+      context_window: 131072
+      max_output_tokens: 8192
+      verification:
+        status: operational
+        source: provider_catalog
+
+    # MiniMax
+    minimaxai/minimax-m2.7:
+      context_window: 128000
+      max_output_tokens: 8192
+      verification:
+        status: operational
+        source: provider_catalog
+
+    # Google
+    google/gemma-4-31b-it:
+      context_window: 256000
+      max_output_tokens: 8192
+      verification:
+        status: operational
+        source: provider_catalog
+    google/diffusiongemma-26b-a4b-it:
+      context_window: 131072
+      max_output_tokens: 4096
+      verification:
+        status: operational
+        source: provider_catalog
+
+    # Meta / Llama
+    meta/llama-4-maverick-17b-128e-instruct:
+      context_window: 1000000
+      max_output_tokens: 4096
+      verification:
+        status: operational
+        source: provider_catalog
+    meta/llama-3.3-70b-instruct:
+      context_window: 128000
+      max_output_tokens: 8192
+      verification:
+        status: operational
+        source: provider_catalog
+
+    # StepFun
+    stepfun-ai/step-3.7-flash:
+      context_window: 262144
+      max_output_tokens: 32768
+      verification:
+        status: operational
+        source: provider_catalog
+        notes: "198B MoE, 11B activated, vision-language, agentic"
+    stepfun-ai/step-3.5-flash:
+      context_window: 131072
+      max_output_tokens: 16384
+      verification:
+        status: operational
+        source: provider_catalog
+
+    # OpenAI OSS
+    openai/gpt-oss-120b:
+      context_window: 128000
+      max_output_tokens: 8192
+      verification:
+        status: operational
+        source: provider_catalog
+    openai/gpt-oss-20b:
+      context_window: 128000
+      max_output_tokens: 8192
+      verification:
+        status: operational
+        source: provider_catalog
+
+    # Legacy chat model (provider catalog)
+    meta/llama-3.1-8b-instruct:
+      context_window: 128000
+      max_output_tokens: 8192
+      verification:
+        status: operational
+        source: provider_catalog
+  sdk:
+    note: "OpenAI-compatible — use openai SDK with base_url override"
+    python: openai
+    typescript: openai

package/v2/providers/openai.yaml

@@ -1,14 +1,13 @@
-# OpenAI V2 正式提供商清单 — 合并 v2-alpha 结构与最新模型数据
-# AI-Protocol V2 Provider Manifest
-# Provider: OpenAI | Models: GPT-5.x series
-# Last Updated: 2026-02-19
+# OpenAI V2 Provider Manifest
+# Provider: OpenAI | Models: GPT-5.5 / GPT-5.4 / GPT-5.4-mini
+# Last Updated: 2026-06-26 (GPT-5→5.5, add tool_calling)
 $schema: "https://raw.githubusercontent.com/ailib-official/ai-protocol/main/schemas/v2/provider.json"
 
 # === Ring 1: Core Skeleton ===
 id: openai
 protocol_version: "2.0"
 name: "OpenAI"
-version: "5.3.0"
+version: "5.5.0"
 status: stable
 category: ai_provider
 official_url: "https://platform.openai.com/docs"
@@ -44,6 +43,16 @@ error_classification:
     "insufficient_quota": "quota_exhausted"
     "server_error": "server_error"
 
+availability:
+  required: false
+  regions:
+    - global
+  check:
+    method: GET
+    path: "/models"
+    expected_status: [200, 401]
+    timeout_ms: 3000
+
 # === Ring 2: Capability Mapping ===
 capabilities:
   required:
@@ -72,6 +81,13 @@ capabilities:
     system_messages: true
     image_generation: true
 
+  tool_calling:
+    native:
+      supported: true
+      reliability: "full"
+      parallel: true
+      streaming: true
+
 capability_profile:
   phase: "ios_v1"
   inputs:
@@ -245,21 +261,27 @@ termination:
 
 metadata:
   models:
-    gpt-5.2:
+    gpt-5.5:
+      context_window: 1048576
+      max_output_tokens: 128000
+      release_date: "2026-06-19"
+      notes: "Latest frontier model for complex reasoning and coding"
+    gpt-5.4:
+      context_window: 400000
+      max_output_tokens: 16384
+      release_date: "2026-03-01"
+    gpt-5.4-mini:
       context_window: 400000
       max_output_tokens: 16384
-      release_date: "2025-12-10"
-      pricing: { input_per_1m: 1.75, output_per_1m: 14.00 }
+      release_date: "2026-03-01"
     gpt-5.3-codex:
       context_window: 400000
       max_output_tokens: 16384
       release_date: "2026-02-05"
       pricing: { input_per_1m: 1.25, output_per_1m: 10.00 }
-    gpt-5.3-codex-spark:
+    gpt-4o-mini:
       context_window: 128000
       max_output_tokens: 16384
-      release_date: "2026-02-12"
-      status: research_preview
   sdk:
     python: openai
     typescript: openai

package/v2/providers/qwen.yaml

@@ -1,10 +1,11 @@
 # Qwen V2 provider manifest — P0 multimodal generative coverage
+# Last Updated: 2026-06-26 (add tool_calling, Qwen3→3.7)
 $schema: "https://raw.githubusercontent.com/ailib-official/ai-protocol/main/schemas/v2/provider.json"
 
 id: qwen
 protocol_version: "2.0"
 name: "Qwen (DashScope)"
-version: "3.0.0"
+version: "3.7.0"
 status: stable
 category: ai_provider
 official_url: "https://help.aliyun.com/zh/model-studio"
@@ -29,6 +30,17 @@ error_classification:
     "500": "server_error"
     "503": "overloaded"
 
+availability:
+  required: false
+  regions:
+    - cn
+    - global
+  check:
+    method: GET
+    path: "/models"
+    expected_status: [200, 401]
+    timeout_ms: 3000
+
 capabilities:
   required:
     - text
@@ -51,6 +63,13 @@ capabilities:
     system_messages: true
     image_generation: true
 
+  tool_calling:
+    native:
+      supported: true
+      reliability: "full"
+      parallel: true
+      streaming: true
+
 capability_profile:
   phase: "ios_v1"
   inputs:
@@ -66,14 +85,6 @@ parameters:
   top_p: { type: float, range: [0.0, 1.0] }
   stream: { type: boolean }
 
-parameter_mappings:
-  temperature: "temperature"
-  max_tokens: "max_tokens"
-  stream: "stream"
-  top_p: "top_p"
-  tools: "tools"
-  tool_choice: "tool_choice"
-
 streaming:
   decoder:
     format: "sse"