@ailib-official/ai-protocol 0.8.4

package/v1/providers/cohere.yaml

@@ -0,0 +1,175 @@
+$schema: "https://raw.githubusercontent.com/ailib-official/ai-protocol/main/schemas/v1.json"
+
+id: cohere
+protocol_version: "1.5"
+
+name: Cohere
+version: "v2"
+status: stable
+category: ai_provider
+official_url: "https://docs.cohere.com"
+support_contact: "https://cohere.com/contact"
+
+endpoint:
+  base_url: "https://api.cohere.com/v2"
+  protocol: https
+  timeout_ms: 60000
+
+auth:
+  type: bearer
+  token_env: "COHERE_API_KEY"
+
+payload_format: "cohere_v2"
+
+# Cohere V2 API
+api_families: ["chat"]
+default_api_family: "chat"
+
+endpoints:
+  chat:
+    path: "/chat"
+    method: "POST"
+    adapter: "cohere_v2"
+
+services:
+  list_models:
+    path: "/models"
+    method: "GET"
+    response_binding: "models"
+
+termination:
+  source_field: "finish_reason"
+  mapping:
+    COMPLETE: "end_turn"
+    MAX_TOKENS: "max_tokens"
+    STOP_SEQUENCE: "stop_sequence"
+    TOOL_CALL: "tool_use"
+    ERROR: "error"
+  notes:
+    - "Cohere uses uppercase finish reasons."
+    - "Known for Command R+ and enterprise-grade RAG capabilities."
+
+tooling:
+  source_model: "openai_tool_calls"
+  tool_use:
+    id_path: "id"
+    name_path: "name"
+    input_path: "parameters"
+    input_format: "json_object"
+  notes:
+    - "Cohere V2 uses 'tools' array similar to OpenAI."
+
+rate_limit_headers:
+  requests_limit: "x-ratelimit-limit"
+  requests_remaining: "x-ratelimit-remaining"
+  requests_reset: "x-ratelimit-reset"
+
+retry_policy:
+  strategy: "exponential_backoff"
+  max_retries: 3
+  min_delay_ms: 500
+  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"
+    "422": "invalid_request"
+    "429": "rate_limited"
+    "500": "server_error"
+    "503": "overloaded"
+
+parameter_mappings:
+  temperature: "temperature"
+  max_tokens: "max_tokens"
+  stream: "stream"
+  top_p: "p"
+  top_k: "k"
+  stop_sequences: "stop_sequences"
+  tools: "tools"
+  frequency_penalty: "frequency_penalty"
+  presence_penalty: "presence_penalty"
+  seed: "seed"
+
+response_format: "cohere_v2"
+
+response_paths:
+  content: "message.content[0].text"
+  tool_calls: "message.tool_calls"
+  usage: "usage"
+  finish_reason: "finish_reason"
+
+streaming:
+  event_format: "sse"
+  decoder:
+    format: "sse"
+    delimiter: "\n"
+    prefix: "data: "
+    done_signal: ""
+  content_path: "delta.message.content.text"
+  tool_call_path: "delta.message.tool_calls"
+  usage_path: "usage"
+  frame_selector: "exists($.type)"
+  event_map:
+    - match: "$.type == 'content-delta'"
+      emit: "PartialContentDelta"
+      fields:
+        content: "$.delta.message.content.text"
+    - match: "$.type == 'tool-call-start'"
+      emit: "ToolCallStarted"
+      fields:
+        tool_call_id: "$.delta.message.tool_calls.id"
+        tool_name: "$.delta.message.tool_calls.name"
+    - match: "$.type == 'tool-call-delta'"
+      emit: "PartialToolCall"
+      fields:
+        arguments: "$.delta.message.tool_calls.parameters"
+    - match: "$.type == 'message-end'"
+      emit: "FinalCandidate"
+      fields:
+        finish_reason: "$.delta.finish_reason"
+        usage: "$.delta.usage"
+  stop_condition: "$.type == 'message-end'"
+
+features:
+  multi_candidate:
+    support_type: "simulated"
+    max_concurrent: 1
+  response_mapping:
+    tool_calls:
+      path: "message.tool_calls"
+      fields:
+        id: "id"
+        name: "name"
+        args: "parameters"
+    error:
+      message_path: "message"
+      code_path: "code"
+
+capabilities:
+  streaming: true
+  tools: true
+  vision: false
+  agentic: true
+  parallel_tools: false
+  reasoning: false
+
+availability:
+  required: false
+  regions:
+    - global
+  check:
+    method: GET
+    path: "/models"
+    expected_status: [200, 401]
+    timeout_ms: 5000
+
+experimental_features:
+  - "rag"
+  - "connectors"
+  - "rerank"
+  - "embed"

package/v1/providers/deepinfra.yaml

@@ -0,0 +1,156 @@
+$schema: ../../schemas/v1.json
+protocol_version: "1.5"
+id: "deepinfra"
+name: "DeepInfra"
+status: "stable"
+category: "model_provider"
+official_url: "https://deepinfra.com/"
+support_contact: "https://deepinfra.com/support"
+
+endpoint:
+  base_url: "https://api.deepinfra.com/v1/openai"
+  protocol: "https"
+  timeout_ms: 60000
+
+# Auth + payload format are required for online model fact-checking.
+auth:
+  type: bearer
+  token_env: "DEEPINFRA_API_KEY"
+payload_format: "openai_style"
+
+services:
+  list_models:
+    path: "/models"
+    method: "GET"
+    response_binding: "data"
+
+availability:
+  required: false
+  regions: ["global"]
+  check:
+    method: "GET"
+    path: "/models"
+    expected_status: [200]
+
+capabilities:
+  streaming: true
+  tools: true
+  vision: true
+  agentic: false
+  reasoning: false
+  parallel_tools: true
+
+# DeepInfra is OpenAI-compatible
+streaming:
+  event_format: "data_lines"
+  decoder:
+    format: "sse"
+    delimiter: "\n\n"
+    prefix: "data: "
+    done_signal: "[DONE]"
+  frame_selector: "exists($.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 (DeepInfra 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 (DeepInfra 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:
+    - "DeepInfra 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"
+
+response_paths:
+  content: "choices[0].message.content"
+  tool_calls: "choices[0].message.tool_calls"
+  usage: "usage"
+  finish_reason: "choices[0].finish_reason"
+
+parameter_mappings:
+  temperature: "temperature"
+  max_tokens: "max_tokens"
+  stream: "stream"
+  top_p: "top_p"
+  stop_sequences: "stop"
+  tools: "tools"
+  tool_choice: "tool_choice"
+
+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"
+    "422": "invalid_request"
+    "429": "rate_limited"
+    "500": "server_error"
+    "503": "overloaded"

package/v1/providers/deepseek.yaml

@@ -0,0 +1,189 @@
+$schema: "https://raw.githubusercontent.com/ailib-official/ai-protocol/main/schemas/v1.json"
+
+id: deepseek
+provider_id: deepseek
+protocol_version: "1.5"
+
+name: DeepSeek
+version: "v1"
+status: stable
+category: ai_provider
+official_url: "https://platform.deepseek.com/docs"
+support_contact: "https://support.deepseek.com"
+
+endpoint:
+  base_url: "https://api.deepseek.com/v1"
+  protocol: https
+  timeout_ms: 10000
+auth:
+  type: bearer
+  token_env: "DEEPSEEK_API_KEY"
+payload_format: "openai_style"
+
+# V1.5 Structured Endpoints
+endpoints:
+  chat:
+    path: "/chat/completions"
+    method: "POST"
+    adapter: "openai"
+  completions:
+    path: "/completions"
+    method: "POST"
+    adapter: "openai"
+
+# V1.5 Service Endpoints
+services:
+  list_models:
+    path: "/models"
+    method: "GET"
+    response_binding: "data"
+  get_balance:
+    path: "/user/balance"
+    method: "GET"
+
+# Legacy support
+api_families: ["chat_completions"]
+default_api_family: "chat_completions"
+
+termination:
+  source_field: "finish_reason"
+  mapping:
+    stop: "end_turn"
+    length: "max_tokens"
+    tool_calls: "tool_use"
+    content_filter: "refusal"
+
+tooling:
+  source_model: "openai_tool_calls"
+  tool_use:
+    id_path: "id"
+    name_path: "function.name"
+    input_path: "function.arguments"
+    input_format: "json_string"
+
+retry_policy:
+  strategy: "exponential_backoff"
+  max_retries: 2
+  min_delay_ms: 1000
+  max_delay_ms: 8000
+  jitter: "full"
+  retry_on_http_status: [429, 500, 502, 503]
+
+# Rate limit headers (OpenAI-compatible)
+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"
+
+error_classification:
+  by_http_status:
+    "400": "invalid_request"
+    "401": "authentication"
+    "402": "quota_exhausted"
+    "422": "invalid_request"
+    "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"
+
+response_format: "openai_style"
+response_paths:
+  content: "choices[0].message.content"
+  # Non-stream fallback when `content` is null (e.g. some reasoner / thinking payloads).
+  reasoning_content: "choices[0].message.reasoning_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]"
+    strategy: "sse_ignore_comments"
+  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"
+    # Stream end with finish reason
+    - match: "$.choices[*].finish_reason != null"
+      emit: "StreamEnd"
+      fields:
+        finish_reason: "$.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"
+      param_path: "error.param"
+
+capabilities:
+  streaming: true
+  tools: true
+  vision: false
+  agentic: true
+  parallel_tools: false
+  reasoning: true
+
+availability:
+  required: false
+  regions:
+    - cn
+    - global
+  check:
+    method: GET
+    path: "/models"
+    expected_status: [200, 401]
+    timeout_ms: 3000
+experimental_features:
+  - "deep_thinking"

package/v1/providers/doubao.yaml

@@ -0,0 +1,172 @@
+$schema: "https://raw.githubusercontent.com/ailib-official/ai-protocol/main/schemas/v1.json"
+
+id: doubao
+protocol_version: "1.5"
+
+name: 字节跳动豆包 (Doubao)
+version: "v1"
+status: stable
+category: ai_provider
+official_url: "https://www.volcengine.com/docs/82379"
+support_contact: "https://www.volcengine.com/contact"
+
+endpoint:
+  base_url: "https://ark.cn-beijing.volces.com/api/v3"
+  protocol: https
+  timeout_ms: 60000
+
+auth:
+  type: bearer
+  token_env: "VOLCENGINE_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"
+    content_filter: "refusal"
+  notes:
+    - "OpenAI-compatible finish_reason."
+    - "Doubao (豆包) is ByteDance's flagship LLM accessed via Volcengine."
+
+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"
+  frequency_penalty: "frequency_penalty"
+  presence_penalty: "presence_penalty"
+
+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: true
+
+availability:
+  required: false
+  regions:
+    - cn
+  check:
+    method: GET
+    path: "/models"
+    expected_status: [200, 401]
+    timeout_ms: 5000
+
+experimental_features:
+  - "bot_chat"
+  - "embeddings"
+  - "image_generation"