@ailib-official/ai-protocol 0.8.4 → 1.0.0

package/schemas/v2/tool-calling.json

@@ -0,0 +1,61 @@
+{
+  "$schema": "https://json-schema.org/draft/2020-12/schema",
+  "$id": "https://raw.githubusercontent.com/ailib-official/ai-protocol/main/schemas/v2/tool-calling.json",
+  "title": "Tool Calling Capability Declaration",
+  "description": "Describes native function calling support and text fallback configuration for a provider.",
+  "type": "object",
+  "properties": {
+    "native": {
+      "type": "object",
+      "properties": {
+        "supported": { "type": "boolean", "default": true },
+        "reliability": {
+          "type": "string",
+          "enum": ["full", "partial", "unreliable"],
+          "default": "full"
+        },
+        "parallel": { "type": "boolean" },
+        "streaming": { "type": "boolean" },
+        "notes": { "type": "string" }
+      },
+      "additionalProperties": false
+    },
+    "text_fallback": {
+      "type": ["object", "null"],
+      "properties": {
+        "format": {
+          "type": "string",
+          "enum": ["xml_json", "markdown_fence", "custom"]
+        },
+        "wrapper": { "type": "string" },
+        "body": {
+          "type": "string",
+          "enum": ["json", "yaml", "plain"]
+        },
+        "name_location": {
+          "type": "string",
+          "enum": ["attribute", "json_field"]
+        },
+        "args_key": { "type": "string" },
+        "known_dialects": {
+          "type": "array",
+          "items": {
+            "type": "object",
+            "required": ["tag", "map_to"],
+            "properties": {
+              "tag": { "type": "string" },
+              "map_to": { "type": "string" }
+            },
+            "additionalProperties": false
+          }
+        },
+        "prompt_level": {
+          "type": "string",
+          "enum": ["L1", "L2", "L3"]
+        }
+      },
+      "additionalProperties": false
+    }
+  },
+  "additionalProperties": false
+}

package/v1/models/deepseek-chat.yaml

@@ -4,8 +4,8 @@ protocol_version: "1.1"
 models:
   deepseek-chat:
     provider: deepseek
-    model_id: "deepseek-chat"
-    display_name: "DeepSeek Chat"
+    model_id: "deepseek-v4-pro"
+    display_name: "DeepSeek Chat (V4 Pro)"
     verification:
       status: verified
       verified_at: "2026-01-28"
@@ -20,8 +20,8 @@ models:
 
   deepseek-reasoner:
     provider: deepseek
-    model_id: "deepseek-reasoner"
-    display_name: "DeepSeek Reasoner"
+    model_id: "deepseek-v4-flash"
+    display_name: "DeepSeek Reasoner (V4 Flash)"
     verification:
       status: verified
       verified_at: "2026-03-06"

package/v1/models/gemini.yaml

@@ -1,5 +1,5 @@
 $schema: "https://raw.githubusercontent.com/ailib-official/ai-protocol/main/schemas/v1.json"
-protocol_version: "1.1"
+protocol_version: "1.5"
 
 # Gemini Model Instances
 models:
@@ -30,3 +30,33 @@ models:
     capabilities: ["chat", "vision", "audio", "video", "tools", "streaming"]
     status: active
     tags: ["google", "multimodal", "fast", "cost-effective"]
+
+  # VERIFIED_AT: 2026-05-17 (official model card: gemini-2.5-flash-lite)
+  gemini-2.5-flash-lite:
+    provider: gemini
+    model_id: "gemini-2.5-flash-lite"
+    name: "Gemini 2.5 Flash-Lite"
+    verification:
+      status: verified
+      verified_at: "2026-05-17"
+      source: "https://ai.google.dev/gemini-api/docs/models/gemini-2.5-flash-lite"
+      notes: "Free-tier RPM/RPD vary by account and region; typical AI Studio free band is high concurrency vs Flash/Pro."
+    context_window: 1048576
+    capabilities: ["chat", "vision", "audio", "video", "tools", "streaming", "reasoning"]
+    status: active
+    tags: ["google", "multimodal", "flash-lite", "cost-effective", "free-tier"]
+
+  # VERIFIED_AT: 2026-05-17 (official model card: gemini-3.1-flash-lite-preview)
+  gemini-3.1-flash-lite-preview:
+    provider: gemini
+    model_id: "gemini-3.1-flash-lite-preview"
+    name: "Gemini 3.1 Flash-Lite (Preview)"
+    verification:
+      status: verified
+      verified_at: "2026-05-17"
+      source: "https://ai.google.dev/gemini-api/docs/models/gemini-3.1-flash-lite-preview"
+      notes: "Preview model; availability and quotas may change by region."
+    context_window: 1048576
+    capabilities: ["chat", "vision", "audio", "video", "tools", "streaming", "reasoning"]
+    status: active
+    tags: ["google", "multimodal", "flash-lite", "preview", "free-tier"]

package/v1/providers/gemini.yaml

@@ -20,14 +20,18 @@ auth:
   token_env: "GEMINI_API_KEY"
 payload_format: "gemini_style"
 
-# Gemini native API families
-api_families: ["generate_content"]
+# Gemini native API plus OpenAI-compatible chat (same API key; Bearer on /openai/*).
+api_families: ["generate_content", "chat_completions"]
 default_api_family: "generate_content"
 endpoints:
   chat:
     path: "/models/{model}:generateContent"
     method: "POST"
     adapter: "gemini"
+  openai_chat:
+    path: "/openai/chat/completions"
+    method: "POST"
+    adapter: "openai"
 
 # V1.5 Service Endpoints
 services:
@@ -227,3 +231,7 @@ availability:
 experimental_features:
   - "multimodal_video"
   - "multimodal_audio"
+
+notes:
+  - "OpenAI-compatible Chat Completions: POST {base_url}/openai/chat/completions with Authorization: Bearer <GEMINI_API_KEY> and OpenAI-style JSON body. See https://ai.google.dev/gemini-api/docs/openai"
+  - "Native GenerateContent continues to use ?key= query authentication as configured under auth.type query_param."

package/v2/contracts/anthropic-messages.contract.yaml

@@ -0,0 +1,55 @@
+# Anthropic Messages API — ProviderContract (content_block_mapping reference)
+# PT-079-R1 | Drives manifest encoder for Claude document blocks
+$schema: "https://raw.githubusercontent.com/ailib-official/ai-protocol/main/schemas/v2/provider-contract.json"
+
+contract_version: "1.0"
+provider_id: anthropic
+api_style: anthropic_messages
+api_version: "2023-06-01"
+
+request_mapping:
+  message_format: anthropic_messages
+  role_mapping:
+    system: system
+    user: user
+    assistant: assistant
+    tool: user
+  system_message_handling: top_level_field
+  content_block_mapping:
+    text:
+      wrapper: content_blocks
+    image:
+      format: anthropic_source
+      base64_field: source.data
+      url_field: source.url
+    document:
+      format: anthropic_document
+      type_field: document
+      source_wrapper: anthropic_source
+      base64_field: source.data
+      mime_type_field: source.media_type
+      default_mime_type: application/pdf
+      ref_resolution: error_before_encode
+
+response_mapping:
+  content_path: "$.content[?(@.type=='text')].text"
+  finish_reason_path: "$.stop_reason"
+  usage_path: "$.usage"
+  finish_reason_mapping:
+    end_turn: stop
+    max_tokens: length
+    tool_use: tool_calls
+    stop_sequence: stop
+
+capability_contracts:
+  streaming:
+    protocol: sse
+    supports_usage_in_stream: true
+    supports_tool_streaming: true
+    done_signal: message_stop
+  tools:
+    definition_format: anthropic_tools
+    supports_parallel: true
+
+authentication_contract:
+  method: api_key_header

package/v2/contracts/gemini-generate.contract.yaml

@@ -0,0 +1,52 @@
+# Google Gemini generateContent — ProviderContract (content_block_mapping reference)
+# PT-079-R1 | Drives manifest encoder for Gemini inline document payloads
+$schema: "https://raw.githubusercontent.com/ailib-official/ai-protocol/main/schemas/v2/provider-contract.json"
+
+contract_version: "1.0"
+provider_id: google
+api_style: gemini_generate
+api_version: "v1beta"
+
+request_mapping:
+  message_format: gemini_contents
+  role_mapping:
+    system: user
+    user: user
+    assistant: model
+    tool: user
+  system_message_handling: prepend_to_user
+  content_block_mapping:
+    text:
+      field: text
+      wrapper: array_of_parts
+    image:
+      format: gemini_inline_data
+      base64_field: inlineData.data
+      mime_type_field: inlineData.mimeType
+    document:
+      format: gemini_inline_data
+      source_wrapper: gemini_inline_data
+      base64_field: inlineData.data
+      mime_type_field: inlineData.mimeType
+      default_mime_type: application/pdf
+      ref_resolution: error_before_encode
+
+response_mapping:
+  content_path: "$.candidates[0].content.parts[?(@.text)].text"
+  finish_reason_path: "$.candidates[0].finishReason"
+  usage_path: "$.usageMetadata"
+  finish_reason_mapping:
+    STOP: stop
+    MAX_TOKENS: length
+    SAFETY: content_filter
+    RECITATION: content_filter
+
+capability_contracts:
+  streaming:
+    protocol: sse
+    supports_usage_in_stream: true
+  tools:
+    definition_format: gemini_declarations
+
+authentication_contract:
+  method: query_param

package/v2/packs/examples/README.md

@@ -0,0 +1,9 @@
+# Pack examples
+
+JSON pack manifests validated against `schemas/v2/pack.json`.
+
+| File | Description |
+|------|-------------|
+| [deepseek-economy-pack.json](./deepseek-economy-pack.json) | Cost-optimized DeepSeek routes with Groq fallback (PR-PP-001) |
+
+Provider and model ids must exist in `v2/providers/*.yaml` manifests.

package/v2/packs/examples/deepseek-economy-pack.json

@@ -0,0 +1,43 @@
+{
+  "$schema": "https://raw.githubusercontent.com/ailib-official/ai-protocol/main/schemas/v2/pack.json",
+  "id": "deepseek-economy-pack",
+  "name": "DeepSeek Economy Pack",
+  "version": "0.1.0",
+  "protocol_version": "2.0",
+  "description": "Cost-optimized chat routes centered on DeepSeek models. Primary: deepseek-chat; fallback: deepseek-reasoner for reasoning-tagged requests. NOT PRODUCTION SLA — example pack for Prism Pack contract (PR-PP-001).",
+  "status": "draft",
+  "optimize_default": "cost",
+  "provider_routes": [
+    {
+      "provider": "deepseek",
+      "model": "deepseek-chat",
+      "priority": 0,
+      "cost_weight": 1.0,
+      "capability_tags": ["chat", "economy"],
+      "notes": "Primary economy route; legacy alias of deepseek-v4-pro per provider manifest"
+    },
+    {
+      "provider": "deepseek",
+      "model": "deepseek-reasoner",
+      "priority": 1,
+      "cost_weight": 0.85,
+      "capability_tags": ["chat", "reasoning"],
+      "notes": "Reasoning-capable fallback; higher cost than deepseek-chat"
+    },
+    {
+      "provider": "groq",
+      "model": "llama-3.1-8b-instant",
+      "priority": 2,
+      "cost_weight": 0.7,
+      "capability_tags": ["chat", "fallback"],
+      "notes": "Cross-provider fallback when DeepSeek is unhealthy (aligns with prism-core cost_router fixtures)"
+    }
+  ],
+  "metadata": {
+    "publisher": "ailib-prism",
+    "task_ref": "PR-PP-001",
+    "docs": "https://github.com/ailib-official/ai-protocol/blob/main/docs/PACK_SPECIFICATION.md",
+    "billing_hint": "pay_per_use",
+    "sla": "NOT PRODUCTION SLA — contract example only"
+  }
+}

package/v2/providers/anthropic.yaml

@@ -1,14 +1,13 @@
-# Anthropic V2 正式提供商清单 — MCP 创建者，合并 v2-alpha 结构与审计修正后数据
-# AI-Protocol V2 Provider Manifest
-# Provider: Anthropic | Models: Claude Opus 4.6 / Sonnet 4.5
-# Last Updated: 2026-02-16
+# Anthropic V2 Provider Manifest — MCP 创建者
+# Provider: Anthropic | Models: Claude Opus 4.8 / Sonnet 4.6 / Haiku 4.5
+# Last Updated: 2026-06-26 (4.6→4.8, add tool_calling)
 $schema: "https://raw.githubusercontent.com/ailib-official/ai-protocol/main/schemas/v2/provider.json"
 
 # === Ring 1: Core Skeleton ===
 id: anthropic
 protocol_version: "2.0"
 name: "Anthropic"
-version: "4.6.0"
+version: "4.8.0"
 status: stable
 category: ai_provider
 official_url: "https://docs.anthropic.com"
@@ -45,6 +44,16 @@ error_classification:
     "api_error": "server_error"
     "overloaded_error": "overloaded"
 
+availability:
+  required: false
+  regions:
+    - global
+  check:
+    method: GET
+    path: "/models"
+    expected_status: [200, 401]
+    timeout_ms: 3000
+
 # === Ring 2: Capability Mapping ===
 capabilities:
   required:
@@ -67,6 +76,13 @@ capabilities:
     streaming_usage: true
     system_messages: true
 
+  tool_calling:
+    native:
+      supported: true
+      reliability: "full"
+      parallel: true
+      streaming: true
+
 capability_profile:
   phase: "ios_v1"
   inputs:
@@ -218,16 +234,21 @@ termination:
 
 metadata:
   models:
-    claude-opus-4-6:
+    claude-opus-4-8:
       context_window: 1000000
-      max_output_tokens: 8192
-      release_date: "2026-02-05"
-      pricing: { input_per_1m: 5.00, output_per_1m: 25.00 }
-    claude-sonnet-4-5:
-      context_window: 200000
-      max_output_tokens: 8192
-      release_date: "2025-11-24"
+      max_output_tokens: 32768
+      release_date: "2026-06-15"
+      pricing: { input_per_1m: 15.00, output_per_1m: 75.00 }
+    claude-sonnet-4-6:
+      context_window: 1000000
+      max_output_tokens: 32768
+      release_date: "2026-06-11"
       pricing: { input_per_1m: 3.00, output_per_1m: 15.00 }
+    claude-haiku-4-5:
+      context_window: 200000
+      max_output_tokens: 64000
+      release_date: "2025-10-01"
+      pricing: { input_per_1m: 0.80, output_per_1m: 4.00 }
   context_management:
     long_context_beta: "context-1m-2025-08-07"
     compaction_api: true

package/v2/providers/cohere.yaml

@@ -1,7 +1,5 @@
 # Cohere V2 Provider Manifest — Chat + Rerank
-# AI-Protocol V2 Provider Manifest
-# Provider: Cohere | Rerank v3.5/v4, Command R+
-# Last Updated: 2026-02-19
+# Last Updated: 2026-06-26 (add tool_calling)
 $schema: "https://raw.githubusercontent.com/ailib-official/ai-protocol/main/schemas/v2/provider.json"
 
 # === Ring 1: Core Skeleton ===
@@ -35,6 +33,16 @@ error_classification:
     "500": "server_error"
     "503": "overloaded"
 
+availability:
+  required: false
+  regions:
+    - global
+  check:
+    method: GET
+    path: "/models"
+    expected_status: [200, 401]
+    timeout_ms: 3000
+
 # === Ring 2: Capability Mapping ===
 capabilities:
   required:
@@ -48,6 +56,11 @@ capabilities:
     structured_output: true
     system_messages: true
 
+  tool_calling:
+    native:
+      supported: true
+      reliability: "full"
+
 capability_profile:
   phase: "ios_v1"
   inputs:

package/v2/providers/deepseek.yaml

@@ -1,21 +1,23 @@
-# DeepSeek V2 正式提供商清单 — OpenAI 兼容 API，超高性价比
+# DeepSeek V4 Provider Manifest — OpenAI/Anthropic 双兼容 API
 # AI-Protocol V2 Provider Manifest
-# Provider: DeepSeek AI | Models: V3.2 / R1
-# Last Updated: 2026-02-16
+# Provider: DeepSeek AI | Models: V4-Pro / V4-Flash
+# Last Updated: 2026-06-26 (updated from V3.2 → V4)
+# Source: https://api-docs.deepseek.com/updates (2026-04-24)
+# Important: deepseek-chat / deepseek-reasoner DEPRECATED — retired 2026-07-24
 $schema: "https://raw.githubusercontent.com/ailib-official/ai-protocol/main/schemas/v2/provider.json"
 
 # === Ring 1: Core Skeleton ===
 id: deepseek
 protocol_version: "2.0"
 name: "DeepSeek"
-version: "3.2.0"
+version: "4.0.0"
 status: stable
 category: ai_provider
 official_url: "https://api-docs.deepseek.com/"
 support_contact: "https://api-docs.deepseek.com/"
 
 endpoint:
-  base_url: "https://api.deepseek.com/v1"
+  base_url: "https://api.deepseek.com"
   chat: "/chat/completions"
   auth:
     type: "bearer"
@@ -27,6 +29,7 @@ error_classification:
   by_http_status:
     "400": "invalid_request"
     "401": "authentication"
+    "402": "insufficient_quota"
     "403": "permission_denied"
     "404": "not_found"
     "413": "request_too_large"
@@ -39,6 +42,17 @@ error_classification:
     "rate_limit_exceeded": "rate_limited"
     "insufficient_quota": "quota_exhausted"
 
+availability:
+  required: false
+  regions:
+    - cn
+    - global
+  check:
+    method: GET
+    path: "/models"
+    expected_status: [200, 401]
+    timeout_ms: 3000
+
 # === Ring 2: Capability Mapping ===
 capabilities:
   required:
@@ -46,7 +60,6 @@ capabilities:
     - streaming
     - tools
   optional:
-    - vision
     - parallel_tools
     - agentic
     - reasoning
@@ -59,6 +72,25 @@ capabilities:
     streaming_usage: true
     system_messages: true
 
+  tool_calling:
+    native:
+      supported: true
+      reliability: "partial"
+      parallel: true
+      streaming: true
+      notes: "DeepSeek native tool calling is unreliable — LLM often outputs <shell> instead; text fallback required"
+    text_fallback:
+      format: "xml_json"
+      wrapper: "tool_call"
+      body: "json"
+      name_location: "json_field"
+      args_key: "arguments"
+      known_dialects:
+        - { tag: "shell", map_to: "shell" }
+        - { tag: "bash", map_to: "shell" }
+        - { tag: "function", map_to: "" }
+      prompt_level: "L2"
+
 capability_profile:
   phase: "ios_v1"
   inputs:
@@ -66,7 +98,7 @@ capability_profile:
   outcomes:
     types: ["text_completion", "structured_json", "tool_call_sequence"]
   systems:
-    requires: ["mcp", "search"]
+    requires: ["mcp"]
 
 parameters:
   temperature: { type: float, range: [0.0, 2.0], default: 1.0 }
@@ -75,14 +107,7 @@ parameters:
   frequency_penalty: { type: float, range: [-2.0, 2.0] }
   presence_penalty: { type: float, range: [-2.0, 2.0] }
   stream: { type: boolean }
-
-parameter_mappings:
-  temperature: "temperature"
-  max_tokens: "max_tokens"
-  stream: "stream"
-  top_p: "top_p"
-  tools: "tools"
-  tool_choice: "tool_choice"
+  reasoning_effort: { type: string, enum: ["low", "medium", "high"], default: "medium" }
 
 streaming:
   decoder:
@@ -98,6 +123,10 @@ streaming:
       emit: "PartialContentDelta"
       fields:
         content: "$.choices[*].delta.content"
+    - match: "exists($.choices[*].delta.reasoning_content)"
+      emit: "ThinkingDelta"
+      fields:
+        thinking: "$.choices[*].delta.reasoning_content"
     - match: "exists($.choices[*].delta.tool_calls)"
       emit: "PartialToolCall"
       fields:
@@ -115,10 +144,7 @@ streaming:
 multimodal:
   input:
     vision:
-      supported: true
-      formats: [jpeg, png, gif, webp]
-      encoding_methods: [base64_inline, url]
-      document_understanding: true
+      supported: false
     audio:
       supported: false
     video:
@@ -131,19 +157,23 @@ multimodal:
       supported: false
 
 # === Ring 3: Advanced Extensions ===
-api_families: ["chat_completions"]
+api_families: ["chat_completions", "anthropic_messages"]
 default_api_family: "chat_completions"
 
 endpoints:
-  chat:
+  chat_openai:
     path: "/chat/completions"
     method: "POST"
     adapter: "openai"
+  chat_anthropic:
+    path: "/anthropic/v1/messages"
+    method: "POST"
+    adapter: "anthropic"
 
 retry_policy:
   strategy: "exponential_backoff"
   max_retries: 3
-  min_delay_ms: 1000
+  min_delay_ms: 500
   max_delay_ms: 30000
   jitter: "full"
   retry_on_http_status: [429, 500, 502, 503]
@@ -156,21 +186,35 @@ termination:
     tool_calls: "tool_use"
 
 metadata:
-  api_compatibility: openai
+  api_compatibility: [openai, anthropic]
   open_source:
-    license: MIT
-    hugging_face: "https://huggingface.co/deepseek-ai/DeepSeek-V3.2"
+    license: "MIT (weights)"
+    hugging_face: "https://huggingface.co/collections/deepseek-ai/deepseek-v4"
   models:
+    deepseek-v4-pro:
+      context_window: 1048576
+      max_output_tokens: 393216
+      release_date: "2026-04-24"
+      architecture: { type: moe, total: "1.6T", active: "49B" }
+      pricing: { input_per_1m: 0.435, output_per_1m: 0.87, cache_hit_input_per_1m: 0.003625 }
+      thinking: "dual_mode"
+    deepseek-v4-flash:
+      context_window: 1048576
+      max_output_tokens: 393216
+      release_date: "2026-04-24"
+      architecture: { type: moe, total: "284B", active: "13B" }
+      pricing: { input_per_1m: 0.14, output_per_1m: 0.28, cache_hit_input_per_1m: 0.0028 }
+      thinking: "dual_mode"
+  deprecated:
     deepseek-chat:
-      context_window: 128000
-      max_output_tokens: 8192
-      architecture: { type: moe, total: "671B", active: "37B", experts: 256 }
-      pricing: { input_per_1m: 0.28, output_per_1m: 0.42 }
+      retired_on: "2026-07-24"
+      maps_to: "deepseek-v4-flash (non-thinking)"
+      note: "Use deepseek-v4-flash instead"
     deepseek-reasoner:
-      context_window: 128000
-      max_output_tokens: 8192
-      pricing: { input_per_1m: 0.55, output_per_1m: 2.19 }
+      retired_on: "2026-07-24"
+      maps_to: "deepseek-v4-flash (thinking mode)"
+      note: "Use deepseek-v4-flash with thinking mode instead"
   sdk:
-    note: "OpenAI-compatible — use openai SDK with base_url override"
+    note: "OpenAI-compatible — use openai SDK with base_url override. Also supports Anthropic Messages API format."
     python: openai
     typescript: openai