@ailib-official/ai-protocol 0.8.4

package/v1/providers/minimax.yaml

@@ -0,0 +1,172 @@
+$schema: "https://raw.githubusercontent.com/ailib-official/ai-protocol/main/schemas/v1.json"
+
+id: minimax
+protocol_version: "1.5"
+
+name: MiniMax
+version: "v1"
+status: stable
+category: ai_provider
+official_url: "https://platform.minimaxi.com/document/ChatCompletion%20v2"
+support_contact: "https://www.minimaxi.com/contact"
+
+endpoint:
+  base_url: "https://api.minimax.chat/v1"
+  protocol: https
+  timeout_ms: 60000
+
+auth:
+  type: bearer
+  token_env: "MINIMAX_API_KEY"
+
+payload_format: "openai_style"
+
+# MiniMax uses OpenAI-compatible API
+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"
+
+termination:
+  source_field: "finish_reason"
+  mapping:
+    stop: "end_turn"
+    length: "max_tokens"
+    tool_calls: "tool_use"
+    content_filter: "refusal"
+  notes:
+    - "OpenAI-compatible finish_reason."
+    - "MiniMax is a leading Chinese AI company known for abab model series."
+
+tooling:
+  source_model: "openai_tool_calls"
+  tool_use:
+    id_path: "id"
+    name_path: "function.name"
+    input_path: "function.arguments"
+    input_format: "json_string"
+  notes:
+    - "Supports OpenAI-compatible function calling."
+
+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]
+
+error_classification:
+  by_http_status:
+    "400": "invalid_request"
+    "401": "authentication"
+    "403": "permission_denied"
+    "404": "not_found"
+    "429": "rate_limited"
+    "500": "server_error"
+    "503": "overloaded"
+
+parameter_mappings:
+  temperature: "temperature"
+  max_tokens: "max_tokens"
+  stream: "stream"
+  top_p: "top_p"
+  stop_sequences: "stop"
+  tools: "tools"
+  tool_choice: "tool_choice"
+  n: "n"
+
+response_format: "openai_style"
+
+response_paths:
+  content: "choices[0].message.content"
+  tool_calls: "choices[0].message.tool_calls"
+  usage: "usage"
+  finish_reason: "choices[0].finish_reason"
+
+streaming:
+  event_format: "data_lines"
+  decoder:
+    format: "sse"
+    delimiter: "\n\n"
+    prefix: "data: "
+    done_signal: "[DONE]"
+  content_path: "choices[0].delta.content"
+  tool_call_path: "choices[0].delta.tool_calls"
+  usage_path: "usage"
+  frame_selector: "exists($.choices) || exists($.error)"
+  event_map:
+    - match: "exists($.choices[*].delta.content)"
+      emit: "PartialContentDelta"
+      fields:
+        content: "$.choices[*].delta.content"
+    - match: "exists($.choices[*].delta.tool_calls[*].function.name)"
+      emit: "ToolCallStarted"
+      fields:
+        tool_call_id: "$.choices[*].delta.tool_calls[*].id"
+        tool_name: "$.choices[*].delta.tool_calls[*].function.name"
+        index: "$.choices[*].delta.tool_calls[*].index"
+    - match: "exists($.choices[*].delta.tool_calls[*].function.arguments)"
+      emit: "PartialToolCall"
+      fields:
+        arguments: "$.choices[*].delta.tool_calls[*].function.arguments"
+        index: "$.choices[*].delta.tool_calls[*].index"
+    - match: "exists($.usage)"
+      emit: "Metadata"
+      fields:
+        usage: "$.usage"
+    - match: "exists($.choices[*].finish_reason)"
+      emit: "FinalCandidate"
+      fields:
+        finish_reason: "$.choices[*].finish_reason"
+  stop_condition: "$.choices[0].finish_reason != null"
+
+features:
+  multi_candidate:
+    support_type: "native"
+    param_name: "n"
+  response_mapping:
+    tool_calls:
+      path: "choices[0].message.tool_calls"
+      fields:
+        id: "id"
+        name: "function.name"
+        args: "function.arguments"
+    error:
+      message_path: "error.message"
+      code_path: "error.code"
+      type_path: "error.type"
+
+capabilities:
+  streaming: true
+  tools: true
+  vision: true
+  agentic: true
+  parallel_tools: true
+  reasoning: false
+
+availability:
+  required: false
+  regions:
+    - cn
+    - global
+  check:
+    method: GET
+    path: "/models"
+    expected_status: [200, 401]
+    timeout_ms: 5000
+
+experimental_features:
+  - "voice_synthesis"
+  - "music_generation"
+  - "video_generation"

package/v1/providers/mistral.yaml

@@ -0,0 +1,186 @@
+$schema: "https://raw.githubusercontent.com/ailib-official/ai-protocol/main/schemas/v1.json"
+
+id: mistral
+protocol_version: "1.5"
+
+name: Mistral AI
+version: "v1"
+status: stable
+category: ai_provider
+official_url: "https://docs.mistral.ai"
+support_contact: "https://mistral.ai/contact"
+
+endpoint:
+  base_url: "https://api.mistral.ai/v1"
+  protocol: https
+  timeout_ms: 10000
+auth:
+  type: bearer
+  token_env: "MISTRAL_API_KEY"
+payload_format: "openai_style"
+
+# OpenAI-compatible API families
+api_families: ["chat_completions"]
+default_api_family: "chat_completions"
+endpoints:
+  # Runtime uses UnifiedRequest.operation (currently "chat") to resolve endpoints.
+  # Keep keys aligned with runtime expectations.
+  chat:
+    path: "/chat/completions"
+    method: "POST"
+    adapter: "openai"
+
+# Optional service endpoints (OpenAI-compatible).
+services:
+  list_models:
+    path: "/models"
+    method: "GET"
+    response_binding: "data"
+
+# Termination reason normalization (see v1/spec.yaml: standard_schema.streaming_events.termination_reasons)
+termination:
+  source_field: "finish_reason"
+  mapping:
+    stop: "end_turn"
+    length: "max_tokens"
+    tool_calls: "tool_use"
+    content_filter: "refusal"
+  notes:
+    - "OpenAI-compatible: finish_reason is reported per candidate."
+
+# Tool invocation normalization (see v1/spec.yaml: standard_schema.content_blocks)
+tooling:
+  source_model: "openai_tool_calls"
+  tool_use:
+    id_path: "id"
+    name_path: "function.name"
+    input_path: "function.arguments"
+    input_format: "json_string"
+  notes:
+    - "OpenAI-compatible tool_calls.function.arguments is typically a JSON string; runtimes SHOULD parse it into an object when normalizing."
+
+# Rate limits (Mistral uses standard HTTP headers)
+# Note: Mistral may use different header names, but we follow OpenAI-compatible pattern
+rate_limit_headers:
+  requests_limit: "x-ratelimit-limit-requests"
+  requests_remaining: "x-ratelimit-remaining-requests"
+  requests_reset: "x-ratelimit-reset-requests"
+  tokens_limit: "x-ratelimit-limit-tokens"
+  tokens_remaining: "x-ratelimit-remaining-tokens"
+  tokens_reset: "x-ratelimit-reset-tokens"
+  retry_after: "retry-after"
+
+# Retry policy (OpenAI-compatible guidance: exponential backoff for 429; retry 5xx after brief wait)
+retry_policy:
+  max_retries: 3
+  max_delay_ms: 30000
+  strategy: "exponential_backoff"
+  min_delay_ms: 1000
+  jitter: "full"
+  retry_on_http_status: [429, 500, 502, 503]
+  notes:
+    - "OpenAI-compatible: exponential backoff is recommended for rate limit errors (429)."
+    - "Retry 500 after a brief wait."
+
+parameter_mappings:
+  temperature: "temperature"
+  max_tokens: "max_tokens"
+  stream: "stream"
+  stop_sequences: "stop"
+  tools: "tools"
+  tool_choice: "tool_choice"
+  frequency_penalty: "frequency_penalty"
+  presence_penalty: "presence_penalty"
+  top_p: "top_p"
+  random_seed: "random_seed"
+  n: "n"
+  parallel_tool_calls: "parallel_tool_calls"
+  safe_prompt: "safe_prompt"
+  response_format: "response_format"
+  prompt_mode: "prompt_mode"
+  prediction: "prediction"
+  metadata: "metadata"
+
+response_format: "openai_style"
+response_paths:
+  content: "choices[0].message.content"
+  tool_calls: "choices[0].message.tool_calls"
+  usage: "usage"
+
+streaming:
+  event_format: "data_lines"
+  decoder:
+    format: "sse"
+    delimiter: "\n\n"
+    prefix: "data: "
+    done_signal: "[DONE]"
+  content_path: "choices[0].delta.content"
+  tool_call_path: "choices[0].delta.tool_calls"
+  usage_path: "usage"
+  frame_selector: "exists($.choices) || exists($.error)"
+  event_map:
+    # Text content delta
+    - match: "exists($.choices[*].delta.content)"
+      emit: "PartialContentDelta"
+      fields:
+        content: "$.choices[*].delta.content"
+    # Tool call start
+    - match: "exists($.choices[*].delta.tool_calls[*].function.name)"
+      emit: "ToolCallStarted"
+      fields:
+        tool_call_id: "$.choices[*].delta.tool_calls[*].id"
+        tool_name: "$.choices[*].delta.tool_calls[*].function.name"
+        index: "$.choices[*].delta.tool_calls[*].index"
+    # Tool call arguments streaming
+    - match: "exists($.choices[*].delta.tool_calls[*].function.arguments)"
+      emit: "PartialToolCall"
+      fields:
+        arguments: "$.choices[*].delta.tool_calls[*].function.arguments"
+        index: "$.choices[*].delta.tool_calls[*].index"
+    # Usage metadata
+    - match: "exists($.usage)"
+      emit: "Metadata"
+      fields:
+        usage: "$.usage"
+    # Finish with reason
+    - match: "exists($.choices[*].finish_reason)"
+      emit: "FinalCandidate"
+      fields:
+        finish_reason: "$.choices[*].finish_reason"
+        candidate_index: "$.choices[*].index"
+  stop_condition: "$.choices[0].finish_reason != null"
+
+features:
+  multi_candidate:
+    support_type: "native"
+    param_name: "n"
+  response_mapping:
+    tool_calls:
+      path: "choices[0].message.tool_calls"
+      fields:
+        id: "id"
+        name: "function.name"
+        args: "function.arguments"
+    error:
+      message_path: "error.message"
+      code_path: "error.code"
+      type_path: "error.type"
+      param_path: "error.param"
+
+capabilities:
+  streaming: true
+  tools: true
+  vision: false
+  agentic: true
+  parallel_tools: true
+  reasoning: false
+
+availability:
+  required: false
+  regions:
+    - global
+  check:
+    method: GET
+    path: "/models"
+    expected_status: [200, 401]
+    timeout_ms: 3000

package/v1/providers/moonshot.yaml

@@ -0,0 +1,176 @@
+$schema: "https://raw.githubusercontent.com/ailib-official/ai-protocol/main/schemas/v1.json"
+
+id: moonshot
+protocol_version: "1.5"
+
+name: 月之暗面 Kimi (Moonshot)
+version: "v1"
+status: stable
+category: ai_provider
+official_url: "https://platform.moonshot.cn/docs"
+support_contact: "https://www.moonshot.cn/contact"
+
+endpoint:
+  base_url: "https://api.moonshot.cn/v1"
+  protocol: https
+  timeout_ms: 60000
+
+auth:
+  type: bearer
+  token_env: "MOONSHOT_API_KEY"
+
+payload_format: "openai_style"
+
+# OpenAI-compatible API
+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"
+  # Kimi supports file upload for context
+  upload_file:
+    path: "/files"
+    method: "POST"
+  list_files:
+    path: "/files"
+    method: "GET"
+
+termination:
+  source_field: "finish_reason"
+  mapping:
+    stop: "end_turn"
+    length: "max_tokens"
+    tool_calls: "tool_use"
+  notes:
+    - "OpenAI-compatible finish_reason."
+    - "Kimi is known for ultra-long context (up to 200K tokens) and file understanding."
+
+tooling:
+  source_model: "openai_tool_calls"
+  tool_use:
+    id_path: "id"
+    name_path: "function.name"
+    input_path: "function.arguments"
+    input_format: "json_string"
+  notes:
+    - "Supports OpenAI-compatible function calling."
+
+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]
+
+error_classification:
+  by_http_status:
+    "400": "invalid_request"
+    "401": "authentication"
+    "403": "permission_denied"
+    "404": "not_found"
+    "429": "rate_limited"
+    "500": "server_error"
+    "503": "overloaded"
+
+parameter_mappings:
+  temperature: "temperature"
+  max_tokens: "max_tokens"
+  stream: "stream"
+  top_p: "top_p"
+  stop_sequences: "stop"
+  tools: "tools"
+  tool_choice: "tool_choice"
+  n: "n"
+
+response_format: "openai_style"
+
+response_paths:
+  content: "choices[0].message.content"
+  tool_calls: "choices[0].message.tool_calls"
+  usage: "usage"
+  finish_reason: "choices[0].finish_reason"
+
+streaming:
+  event_format: "data_lines"
+  decoder:
+    format: "sse"
+    delimiter: "\n\n"
+    prefix: "data: "
+    done_signal: "[DONE]"
+  content_path: "choices[0].delta.content"
+  tool_call_path: "choices[0].delta.tool_calls"
+  usage_path: "usage"
+  frame_selector: "exists($.choices) || exists($.error)"
+  event_map:
+    - match: "exists($.choices[*].delta.content)"
+      emit: "PartialContentDelta"
+      fields:
+        content: "$.choices[*].delta.content"
+    - match: "exists($.choices[*].delta.tool_calls[*].function.name)"
+      emit: "ToolCallStarted"
+      fields:
+        tool_call_id: "$.choices[*].delta.tool_calls[*].id"
+        tool_name: "$.choices[*].delta.tool_calls[*].function.name"
+        index: "$.choices[*].delta.tool_calls[*].index"
+    - match: "exists($.choices[*].delta.tool_calls[*].function.arguments)"
+      emit: "PartialToolCall"
+      fields:
+        arguments: "$.choices[*].delta.tool_calls[*].function.arguments"
+        index: "$.choices[*].delta.tool_calls[*].index"
+    - match: "exists($.usage)"
+      emit: "Metadata"
+      fields:
+        usage: "$.usage"
+    - match: "exists($.choices[*].finish_reason)"
+      emit: "FinalCandidate"
+      fields:
+        finish_reason: "$.choices[*].finish_reason"
+  stop_condition: "$.choices[0].finish_reason != null"
+
+features:
+  multi_candidate:
+    support_type: "native"
+    param_name: "n"
+  response_mapping:
+    tool_calls:
+      path: "choices[0].message.tool_calls"
+      fields:
+        id: "id"
+        name: "function.name"
+        args: "function.arguments"
+    error:
+      message_path: "error.message"
+      code_path: "error.code"
+
+capabilities:
+  streaming: true
+  tools: true
+  vision: true
+  agentic: true
+  parallel_tools: true
+  reasoning: false
+
+availability:
+  required: false
+  regions:
+    - cn
+  check:
+    method: GET
+    path: "/models"
+    expected_status: [200, 401]
+    timeout_ms: 5000
+
+experimental_features:
+  - "file_upload"
+  - "web_search"
+  - "long_context"

package/v1/providers/nvidia.yaml

@@ -0,0 +1,172 @@
+$schema: "https://raw.githubusercontent.com/ailib-official/ai-protocol/main/schemas/v1.json"
+
+id: nvidia
+protocol_version: "1.5"
+
+name: NVIDIA API Catalog
+version: "v1"
+status: stable
+category: third_party_aggregator
+official_url: "https://build.nvidia.com/explore/discover"
+support_contact: "https://docs.api.nvidia.com/nim/docs"
+
+# NVIDIA API Catalog (NVIDIA Build) - OpenAI-compatible chat completions
+# Base URL: https://integrate.api.nvidia.com (docs.api.nvidia.com/nim/reference/llm-apis)
+endpoint:
+  base_url: "https://integrate.api.nvidia.com/v1"
+  protocol: https
+  timeout_ms: 60000
+
+auth:
+  type: bearer
+  token_env: "NVIDIA_API_KEY"
+
+payload_format: "openai_style"
+
+api_families: ["chat_completions", "completions"]
+default_api_family: "chat_completions"
+endpoints:
+  chat:
+    path: "/chat/completions"
+    method: "POST"
+    adapter: "openai"
+  completions:
+    path: "/completions"
+    method: "POST"
+    adapter: "openai"
+
+services:
+  list_models:
+    path: "/models"
+    method: "GET"
+    response_binding: "data"
+
+termination:
+  source_field: "finish_reason"
+  mapping:
+    stop: "end_turn"
+    length: "max_tokens"
+    tool_calls: "tool_use"
+    content_filter: "refusal"
+  notes:
+    - "OpenAI-compatible: finish_reason is reported per candidate."
+
+tooling:
+  source_model: "openai_tool_calls"
+  tool_use:
+    id_path: "id"
+    name_path: "function.name"
+    input_path: "function.arguments"
+    input_format: "json_string"
+  notes:
+    - "OpenAI-compatible tool_calls.function.arguments is typically a JSON string."
+
+retry_policy:
+  max_retries: 3
+  max_delay_ms: 30000
+  strategy: "exponential_backoff"
+  min_delay_ms: 1000
+  jitter: "full"
+  retry_on_http_status: [429, 500, 502, 503]
+  notes:
+    - "Exponential backoff recommended for 429; retry 5xx after brief wait."
+
+error_classification:
+  by_http_status:
+    "400": "invalid_request"
+    "401": "authentication"
+    "403": "permission_denied"
+    "404": "not_found"
+    "429": "rate_limited"
+    "500": "server_error"
+
+parameter_mappings:
+  temperature: "temperature"
+  max_tokens: "max_tokens"
+  stream: "stream"
+  stop_sequences: "stop"
+  tools: "tools"
+  tool_choice: "tool_choice"
+  frequency_penalty: "frequency_penalty"
+  presence_penalty: "presence_penalty"
+  top_p: "top_p"
+
+response_format: "openai_style"
+response_paths:
+  content: "choices[0].message.content"
+  tool_calls: "choices[0].message.tool_calls"
+  usage: "usage"
+  finish_reason: "choices[0].finish_reason"
+
+streaming:
+  event_format: "data_lines"
+  decoder:
+    format: "sse"
+    delimiter: "\n\n"
+    prefix: "data: "
+    done_signal: "[DONE]"
+  content_path: "choices[0].delta.content"
+  tool_call_path: "choices[0].delta.tool_calls"
+  usage_path: "usage"
+  frame_selector: "exists($.choices) || exists($.error)"
+  event_map:
+    - match: "exists($.choices[*].delta.content)"
+      emit: "PartialContentDelta"
+      fields:
+        content: "$.choices[*].delta.content"
+    - match: "exists($.choices[*].delta.tool_calls[*].function.name)"
+      emit: "ToolCallStarted"
+      fields:
+        tool_call_id: "$.choices[*].delta.tool_calls[*].id"
+        tool_name: "$.choices[*].delta.tool_calls[*].function.name"
+        index: "$.choices[*].delta.tool_calls[*].index"
+    - match: "exists($.choices[*].delta.tool_calls[*].function.arguments)"
+      emit: "PartialToolCall"
+      fields:
+        arguments: "$.choices[*].delta.tool_calls[*].function.arguments"
+        index: "$.choices[*].delta.tool_calls[*].index"
+    - match: "exists($.usage)"
+      emit: "Metadata"
+      fields:
+        usage: "$.usage"
+    - match: "exists($.choices[*].finish_reason)"
+      emit: "FinalCandidate"
+      fields:
+        finish_reason: "$.choices[*].finish_reason"
+        candidate_index: "$.choices[*].index"
+  stop_condition: "$.choices[0].finish_reason != null"
+
+features:
+  multi_candidate:
+    support_type: "native"
+    param_name: "n"
+  response_mapping:
+    tool_calls:
+      path: "choices[0].message.tool_calls"
+      fields:
+        id: "id"
+        name: "function.name"
+        args: "function.arguments"
+    error:
+      message_path: "error.message"
+      code_path: "error.code"
+      type_path: "error.type"
+      param_path: "error.param"
+
+capabilities:
+  streaming: true
+  tools: true
+  vision: true
+  agentic: true
+  parallel_tools: true
+  reasoning: true
+
+availability:
+  required: false
+  regions:
+    - global
+  check:
+    method: GET
+    path: "/models"
+    expected_status: [200, 401]
+    timeout_ms: 5000