@ailib-official/ai-protocol 0.8.4

package/v2/providers/google.yaml

@@ -0,0 +1,245 @@
+# Google Gemini V2 正式提供商清单 — 全模态支持（音频/视频/CU），合并 v2-alpha 与深度调研数据
+# AI-Protocol V2 Provider Manifest
+# Provider: Google | Models: Gemini 3 Pro / Flash
+# Last Updated: 2026-02-16
+$schema: "https://raw.githubusercontent.com/ailib-official/ai-protocol/main/schemas/v2/provider.json"
+
+# === Ring 1: Core Skeleton ===
+id: google
+protocol_version: "2.0"
+name: "Google Gemini"
+version: "3.0.0"
+status: stable
+category: ai_provider
+official_url: "https://ai.google.dev/docs"
+support_contact: "https://support.google.com/gemini"
+
+endpoint:
+  base_url: "https://generativelanguage.googleapis.com/v1beta"
+  chat: "/models/{model}:generateContent"
+  embeddings: "/models/{model}:embedContent"
+  auth:
+    type: "query_param"
+    param_name: "key"
+    token_env: "GEMINI_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:
+    "INVALID_ARGUMENT": "invalid_request"
+    "UNAUTHENTICATED": "authentication"
+    "PERMISSION_DENIED": "permission_denied"
+    "NOT_FOUND": "not_found"
+    "RESOURCE_EXHAUSTED": "rate_limited"
+    "INTERNAL": "server_error"
+    "UNAVAILABLE": "overloaded"
+    "DEADLINE_EXCEEDED": "timeout"
+
+# === Ring 2: Capability Mapping ===
+capabilities:
+  required:
+    - text
+    - streaming
+    - tools
+  optional:
+    - vision
+    - audio
+    - video
+    - parallel_tools
+    - reasoning
+    - embeddings
+    - structured_output
+    - computer_use
+    - mcp_client
+  feature_flags:
+    structured_output: true
+    parallel_tool_calls: true
+    extended_thinking: true
+    streaming_usage: false
+    system_messages: true
+
+capability_profile:
+  phase: "ios_v1"
+  inputs:
+    modalities: ["text", "image", "audio", "video"]
+  outcomes:
+    types: ["text_completion", "structured_json", "tool_call_sequence"]
+  systems:
+    requires: ["mcp", "search"]
+
+parameters:
+  temperature: { type: float, range: [0.0, 2.0], default: 1.0 }
+  maxOutputTokens: { type: integer, min: 1, max: 65536, alias: "max_tokens" }
+  topP: { type: float, range: [0.0, 1.0], alias: "top_p" }
+  topK: { type: integer, min: 0, alias: "top_k" }
+
+streaming:
+  decoder:
+    format: "gemini_sse"
+    strategy: "gemini_generate"
+  event_map:
+    - match: "$.candidates[0].content.parts[0].text"
+      emit: "PartialContentDelta"
+      extract:
+        content: "$.candidates[0].content.parts[0].text"
+    - match: "$.candidates[0].content.parts[0].functionCall"
+      emit: "PartialToolCall"
+      extract:
+        tool_call: "$.candidates[0].content.parts[0].functionCall"
+    - match: "$.candidates[0].finishReason"
+      emit: "StreamEnd"
+      extract:
+        finish_reason: "$.candidates[0].finishReason"
+  candidate:
+    candidate_id_path: "$.candidates[*].index"
+    fan_out: true
+
+multimodal:
+  input:
+    vision:
+      supported: true
+      formats: [jpeg, png, gif, bmp, tiff, webp]
+      max_file_size: "20MB"
+      max_resolution: "4096x4096"
+      encoding_methods: [base64_inline, url, file_uri]
+      document_understanding: true
+    audio:
+      supported: true
+      formats: [mp3, wav, flac, ogg, aac, m4a]
+      max_duration: "8h"
+      speech_recognition: true
+    video:
+      supported: true
+      formats: [mp4, mov, avi, flv, wmv]
+      max_duration: "1h"
+      frame_extraction: automatic
+      audio_track: true
+      temporal_reasoning: true
+  output:
+    text: true
+    audio:
+      supported: false
+    image:
+      supported: true
+      formats: [png, jpeg]
+  omni_mode:
+    supported: false
+
+computer_use:
+  supported: true
+  status: ga
+  implementation: tool_based
+  actions:
+    screenshot:
+      supported: true
+      formats: [png, jpeg]
+    mouse:
+      supported: true
+      operations: [click, double_click, drag, scroll]
+    keyboard:
+      supported: true
+      operations: [type, key_press, shortcut]
+    browser:
+      supported: true
+      operations: [navigate, fill_form, click_element, scroll]
+  safety:
+    confirmation_required: true
+    sandbox_mode: recommended
+    action_logging: true
+    domain_allowlist: false
+    sensitive_data_protection: true
+  environment:
+    browser: true
+    desktop: true
+  provider_mapping:
+    tool_type: "computer_use"
+    config_method: sdk_config
+
+mcp:
+  client:
+    supported: true
+    protocol_version: "2025-11-25"
+    transports: [streamable_http, sse]
+    auth_methods: [bearer, oauth2, api_key]
+    capabilities:
+      tools: true
+      resources: false
+      prompts: false
+      sampling: false
+    provider_mapping:
+      config_method: sdk_config
+  server:
+    supported: false
+
+# === Ring 3: Advanced Extensions ===
+api_families: ["generate_content"]
+default_api_family: "generate_content"
+
+endpoints:
+  chat:
+    path: "/models/{model}:generateContent"
+    method: "POST"
+    adapter: "gemini"
+  stream_chat:
+    path: "/models/{model}:streamGenerateContent"
+    method: "POST"
+    adapter: "gemini"
+  embeddings:
+    path: "/models/{model}:embedContent"
+    method: "POST"
+    adapter: "gemini"
+
+services:
+  list_models:
+    path: "/models"
+    method: "GET"
+    response_binding: "models"
+  get_model:
+    path: "/models/{model}"
+    method: "GET"
+
+retry_policy:
+  strategy: "exponential_backoff"
+  max_retries: 3
+  min_delay_ms: 1000
+  max_delay_ms: 30000
+  jitter: "full"
+  retry_on_http_status: [429, 500, 503]
+
+termination:
+  source_field: "finishReason"
+  mapping:
+    STOP: "end_turn"
+    MAX_TOKENS: "max_tokens"
+    SAFETY: "refusal"
+    RECITATION: "refusal"
+    OTHER: "other"
+    BLOCKLIST: "refusal"
+    PROHIBITED_CONTENT: "refusal"
+
+metadata:
+  models:
+    gemini-3-pro:
+      context_window: 1000000
+      max_output_tokens: 64000
+      status: public_preview
+      pricing: { input_per_1m: 2.00, output_per_1m: 12.00 }
+    gemini-3-flash:
+      context_window: 1000000
+      max_output_tokens: 64000
+      status: public_preview
+      pricing: { input_per_1m: 0.15, output_per_1m: 0.60 }
+  context_caching:
+    supported: true
+    pricing: "$0.20-0.40 per 1M tokens/hour"
+  sdk:
+    python: google-genai
+    javascript: "@google/genai"
+    go: "cloud.google.com/go/vertexai"

package/v2/providers/jina.yaml

@@ -0,0 +1,55 @@
+# Jina AI V2 Provider Manifest — Rerank 专用
+# AI-Protocol V2 Provider Manifest
+# Provider: Jina AI | Reranker v2/v3 多语言
+# Last Updated: 2026-02-19
+$schema: "https://raw.githubusercontent.com/ailib-official/ai-protocol/main/schemas/v2/provider.json"
+
+# === Ring 1: Core Skeleton ===
+id: jina
+protocol_version: "2.0"
+name: "Jina AI"
+version: "v3"
+status: stable
+category: ai_provider
+official_url: "https://jina.ai"
+support_contact: "https://jina.ai/contact"
+
+endpoint:
+  base_url: "https://api.jina.ai"
+  rerank: "/v1/rerank"
+  auth:
+    type: "bearer"
+    header: "Authorization"
+    prefix: "Bearer"
+    token_env: "JINA_API_KEY"
+
+error_classification:
+  by_http_status:
+    "400": "invalid_request"
+    "401": "authentication"
+    "403": "permission_denied"
+    "404": "not_found"
+    "429": "rate_limited"
+    "500": "server_error"
+
+# === Ring 2: Capability Mapping ===
+capabilities:
+  required:
+    - rerank
+  optional: []
+
+capability_profile:
+  phase: "ios_v1"
+  inputs:
+    modalities: ["text"]
+  outcomes:
+    types: ["other"]
+  systems:
+    requires: ["search"]
+
+parameters:
+  top_n: { type: integer, min: 1 }
+  model: { type: string, default: "jina-reranker-v2-base-multilingual" }
+
+metadata:
+  rerank_models: [jina-reranker-v2-base-multilingual, jina-reranker-v3]

package/v2/providers/moonshot.yaml

@@ -0,0 +1,186 @@
+# Moonshot Kimi V2 正式提供商清单 — Agent Swarm 先驱，OpenAI 兼容 API
+# AI-Protocol V2 Provider Manifest
+# Provider: Moonshot AI | Models: Kimi K2.5 / K2-Thinking
+# Last Updated: 2026-02-16
+$schema: "https://raw.githubusercontent.com/ailib-official/ai-protocol/main/schemas/v2/provider.json"
+
+# === Ring 1: Core Skeleton ===
+id: moonshot
+protocol_version: "2.0"
+name: "Moonshot AI"
+version: "2.5.0"
+status: stable
+category: ai_provider
+official_url: "https://platform.moonshot.ai/docs/"
+support_contact: "https://platform.moonshot.ai/"
+
+endpoint:
+  base_url: "https://api.moonshot.cn/v1"
+  chat: "/chat/completions"
+  auth:
+    type: "bearer"
+    header: "Authorization"
+    prefix: "Bearer"
+    token_env: "MOONSHOT_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"
+    "insufficient_quota": "quota_exhausted"
+
+# === Ring 2: Capability Mapping ===
+capabilities:
+  required:
+    - text
+    - streaming
+    - tools
+  optional:
+    - vision
+    - video
+    - parallel_tools
+    - agentic
+    - reasoning
+    - structured_output
+  feature_flags:
+    structured_output: true
+    parallel_tool_calls: true
+    streaming_usage: true
+    system_messages: true
+
+capability_profile:
+  phase: "ios_v1"
+  inputs:
+    modalities: ["text", "image", "video"]
+  outcomes:
+    types: ["text_completion", "structured_json", "tool_call_sequence"]
+  systems:
+    requires: ["mcp", "search"]
+
+parameters:
+  temperature: { type: float, range: [0.0, 2.0], default: 1.0 }
+  max_tokens: { type: integer, min: 1, max: 33000 }
+  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"
+    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, webp, gif]
+      encoding_methods: [base64_inline, url]
+    audio:
+      supported: false
+    video:
+      supported: true
+      formats: [mp4, avi, mov]
+  output:
+    text: true
+    audio:
+      supported: false
+    image:
+      supported: false
+
+# === Ring 3: Advanced Extensions ===
+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"
+
+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
+  open_source:
+    license: "Modified MIT"
+    hugging_face: "https://huggingface.co/moonshotai/Kimi-K2.5"
+  agent_swarm:
+    supported: true
+    status: beta
+    max_agents: 100
+    max_tool_calls: 1500
+    coordination: "PARL (Parallel Agent RL)"
+  models:
+    kimi-k2-5:
+      context_window: 256000
+      max_output_tokens: 33000
+      architecture: { type: moe, total: "1T", active: "32B", experts: 384 }
+      pricing: { input_per_1m: 0.60, output_per_1m: 3.00 }
+    kimi-k2-thinking:
+      context_window: 256000
+      max_output_tokens: 33000
+      pricing: { input_per_1m: 0.60, output_per_1m: 2.50 }
+  regional:
+    primary: china
+    international_partners: [fireworks_ai, together_ai, openrouter, nvidia_nim]
+  sdk:
+    note: "OpenAI-compatible — use openai SDK with base_url override"
+    python: openai
+    typescript: openai