@ailib-official/ai-protocol 0.8.4

package/v1/providers/azure.yaml

@@ -0,0 +1,155 @@
+# AI-Protocol Provider Manifest for Azure OpenAI
+$schema: "https://raw.githubusercontent.com/ailib-official/ai-protocol/main/schemas/v1.json"
+id: azure
+protocol_version: "1.5"
+
+# Azure URLs are highly dynamic: https://{resource}.openai.azure.com/openai/deployments/{deployment}/chat/completions?api-version={version}
+# This manifest serves as a template. The runtime's `HttpTransport` should support variable injection
+# or the user should configure the `base_url` explicitly in their overlay config.
+endpoint:
+  base_url: "https://${AZURE_RESOURCE_NAME}.openai.azure.com/openai/deployments/${AZURE_DEPLOYMENT_ID}"
+  protocol: "https"
+  timeout_ms: 30000
+
+availability:
+  required: true
+  regions: ["global"]
+  check:
+    method: "GET"
+    path: "/models?api-version=2024-02-01"
+    expected_status: [200]
+
+capabilities:
+  streaming: true
+  tools: true
+  vision: true
+
+auth:
+  type: "api_key"
+  key_env: "AZURE_OPENAI_KEY"
+  header_name: "api-key"
+  # Azure also supports Active Directory via "Authorization: Bearer <token>", 
+  # but api-key is the standard for simple integrations.
+
+parameter_mappings:
+  # Model is technically the "deployment_id" in Azure URL, 
+  # but often passed in body as well for validation.
+  model: model 
+  messages: messages
+  max_tokens: max_tokens
+  temperature: temperature
+  stream: stream
+  tools: tools
+  tool_choice: tool_choice
+
+streaming:
+  event_format: "data_lines"
+  decoder:
+    format: "sse"
+    delimiter: "\n\n"
+    prefix: "data: "
+    done_signal: "[DONE]"
+  frame_selector: "exists($.choices) || exists($.error)"
+  candidate:
+    candidate_id_path: "$.choices[*].index"
+    fan_out: true
+  event_map:
+    - match: "exists($.error)"
+      emit: "StreamError"
+      fields:
+        error: "$.error"
+    - 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: "$.choices[0].finish_reason != null"
+  extra_metadata_path: "$.usage"
+  content_path: "$.choices[0].delta.content"
+  tool_call_path: "$.choices[0].delta.tool_calls"
+
+# Termination reason normalization (Azure OpenAI uses OpenAI-compatible finish_reason)
+termination:
+  source_field: "finish_reason"
+  mapping:
+    stop: "end_turn"
+    length: "max_tokens"
+    tool_calls: "tool_use"
+    content_filter: "refusal"
+  notes:
+    - "finish_reason is reported per candidate."
+
+# Tool invocation normalization (Azure OpenAI is OpenAI-compatible)
+tooling:
+  source_model: "openai_tool_calls"
+  tool_use:
+    id_path: "id"
+    name_path: "function.name"
+    input_path: "function.arguments"
+    input_format: "json_string"
+  notes:
+    - "Azure OpenAI tool_calls.function.arguments is typically a JSON string."
+
+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"
+
+response_paths:
+  content: "choices[0].message.content"
+  tool_calls: "choices[0].message.tool_calls"
+  usage: "usage"
+  finish_reason: "choices[0].finish_reason"
+
+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]
+
+error_classification:
+  by_http_status:
+    "400": "invalid_request"
+    "401": "authentication"
+    "403": "permission_denied"
+    "404": "not_found"
+    "429": "rate_limited"
+    "500": "server_error"
+
+rate_limit_headers:
+  requests_limit: "x-ratelimit-limit-requests"
+  requests_remaining: "x-ratelimit-remaining-requests"
+  requests_reset: "x-ratelimit-reset-requests"  

package/v1/providers/baichuan.yaml

@@ -0,0 +1,168 @@
+$schema: "https://raw.githubusercontent.com/ailib-official/ai-protocol/main/schemas/v1.json"
+
+id: baichuan
+protocol_version: "1.5"
+
+name: 百川智能 (Baichuan)
+version: "v1"
+status: stable
+category: ai_provider
+official_url: "https://platform.baichuan-ai.com/docs/api"
+support_contact: "https://www.baichuan-ai.com/contact"
+
+endpoint:
+  base_url: "https://api.baichuan-ai.com/v1"
+  protocol: https
+  timeout_ms: 60000
+
+auth:
+  type: bearer
+  token_env: "BAICHUAN_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"
+
+termination:
+  source_field: "finish_reason"
+  mapping:
+    stop: "end_turn"
+    length: "max_tokens"
+    tool_calls: "tool_use"
+  notes:
+    - "OpenAI-compatible finish_reason."
+    - "Baichuan is a leading Chinese AI company founded by former Sogou CEO."
+
+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"
+  top_k: "top_k"
+  stop_sequences: "stop"
+  tools: "tools"
+  tool_choice: "tool_choice"
+
+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: false
+  agentic: true
+  parallel_tools: false
+  reasoning: false
+
+availability:
+  required: false
+  regions:
+    - cn
+  check:
+    method: GET
+    path: "/models"
+    expected_status: [200, 401]
+    timeout_ms: 5000
+
+experimental_features:
+  - "knowledge_base"
+  - "web_search"

package/v1/providers/baidu.yaml

@@ -0,0 +1,173 @@
+$schema: "https://raw.githubusercontent.com/ailib-official/ai-protocol/main/schemas/v1.json"
+
+id: baidu
+protocol_version: "1.5"
+
+name: 百度文心 (Baidu ERNIE)
+version: "v1"
+status: stable
+category: ai_provider
+official_url: "https://cloud.baidu.com/doc/WENXINWORKSHOP/index.html"
+support_contact: "https://cloud.baidu.com/support"
+
+endpoint:
+  base_url: "https://aip.baidubce.com/rpc/2.0/ai_custom/v1/wenxinworkshop"
+  protocol: https
+  timeout_ms: 60000
+
+# Baidu uses access_token obtained via API Key + Secret Key
+auth:
+  type: bearer
+  token_env: "BAIDU_ACCESS_TOKEN"
+
+payload_format: "baidu_style"
+
+# Baidu ERNIE API families
+api_families: ["chat"]
+default_api_family: "chat"
+
+endpoints:
+  # ERNIE 4.0
+  chat_ernie4:
+    path: "/chat/completions_pro"
+    method: "POST"
+    adapter: "baidu"
+  # ERNIE 3.5
+  chat_ernie35:
+    path: "/chat/completions"
+    method: "POST"
+    adapter: "baidu"
+  # ERNIE Speed
+  chat_ernie_speed:
+    path: "/chat/ernie_speed"
+    method: "POST"
+    adapter: "baidu"
+  # ERNIE Lite
+  chat_ernie_lite:
+    path: "/chat/ernie-lite-8k"
+    method: "POST"
+    adapter: "baidu"
+
+termination:
+  source_field: "finish_reason"
+  mapping:
+    normal: "end_turn"
+    stop: "end_turn"
+    length: "max_tokens"
+    content_filter: "refusal"
+    function_call: "tool_use"
+  notes:
+    - "Baidu ERNIE uses finish_reason in response."
+    - "文心一言 (ERNIE Bot) is Baidu's flagship conversational AI."
+
+tooling:
+  source_model: "openai_tool_calls"
+  tool_use:
+    id_path: "id"
+    name_path: "name"
+    input_path: "arguments"
+    input_format: "json_string"
+  notes:
+    - "ERNIE supports function calling with similar semantics to OpenAI."
+
+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_output_tokens"
+  stream: "stream"
+  top_p: "top_p"
+  stop_sequences: "stop"
+  tools: "functions"
+  system_message: "system"
+
+response_format: "baidu_style"
+
+response_paths:
+  content: "result"
+  function_call: "function_call"
+  usage: "usage"
+  finish_reason: "finish_reason"
+
+streaming:
+  event_format: "data_lines"
+  decoder:
+    format: "sse"
+    delimiter: "\n\n"
+    prefix: "data: "
+    done_signal: ""
+  content_path: "result"
+  usage_path: "usage"
+  frame_selector: "exists($.result) || exists($.error_code)"
+  event_map:
+    - match: "exists($.result)"
+      emit: "PartialContentDelta"
+      fields:
+        content: "$.result"
+    - match: "exists($.function_call)"
+      emit: "ToolCallStarted"
+      fields:
+        tool_name: "$.function_call.name"
+        arguments: "$.function_call.arguments"
+    - match: "exists($.usage)"
+      emit: "Metadata"
+      fields:
+        usage: "$.usage"
+    - match: "$.is_end == true"
+      emit: "FinalCandidate"
+      fields:
+        finish_reason: "$.finish_reason"
+  stop_condition: "$.is_end == true"
+
+features:
+  multi_candidate:
+    support_type: "simulated"
+    max_concurrent: 1
+  response_mapping:
+    function_call:
+      path: "function_call"
+      fields:
+        name: "name"
+        args: "arguments"
+    error:
+      message_path: "error_msg"
+      code_path: "error_code"
+
+capabilities:
+  streaming: true
+  tools: true
+  vision: true
+  agentic: true
+  parallel_tools: false
+  reasoning: false
+
+availability:
+  required: false
+  regions:
+    - cn
+  check:
+    method: GET
+    path: "/chat/completions"
+    expected_status: [200, 401, 403]
+    timeout_ms: 5000
+
+experimental_features:
+  - "knowledge_base"
+  - "plugins"
+  - "image_understanding"

package/v1/providers/cerebras.yaml

@@ -0,0 +1,178 @@
+$schema: "https://raw.githubusercontent.com/ailib-official/ai-protocol/main/schemas/v1.json"
+
+id: cerebras
+protocol_version: "1.5"
+
+name: Cerebras
+version: "v1"
+status: stable
+category: ai_provider
+official_url: "https://inference-docs.cerebras.ai"
+support_contact: "https://cerebras.ai/contact"
+
+endpoint:
+  base_url: "https://api.cerebras.ai/v1"
+  protocol: https
+  timeout_ms: 10000
+
+auth:
+  type: bearer
+  token_env: "CEREBRAS_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"
+
+termination:
+  source_field: "finish_reason"
+  mapping:
+    stop: "end_turn"
+    length: "max_tokens"
+    tool_calls: "tool_use"
+  notes:
+    - "OpenAI-compatible finish_reason."
+    - "Cerebras is known for extremely fast inference (thousands of tokens/second)."
+
+tooling:
+  source_model: "openai_tool_calls"
+  tool_use:
+    id_path: "id"
+    name_path: "function.name"
+    input_path: "function.arguments"
+    input_format: "json_string"
+  notes:
+    - "Standard OpenAI-compatible tool calling."
+
+rate_limit_headers:
+  requests_limit: "x-ratelimit-limit-requests"
+  requests_remaining: "x-ratelimit-remaining-requests"
+  tokens_limit: "x-ratelimit-limit-tokens"
+  tokens_remaining: "x-ratelimit-remaining-tokens"
+
+retry_policy:
+  strategy: "exponential_backoff"
+  max_retries: 3
+  min_delay_ms: 500
+  max_delay_ms: 10000
+  jitter: "full"
+  retry_on_http_status: [429, 500, 502, 503]
+  notes:
+    - "Cerebras inference is extremely fast; shorter retry delays are appropriate."
+
+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"
+  seed: "seed"
+
+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"
+
+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
+
+experimental_features:
+  - "ultra_fast_inference"
+  - "wafer_scale_engine"