创建 Agent

curl -X POST 'https://app.memorylake.cn/openapi/memorylake/api/v3/agents' \
  -H 'Authorization: Bearer sk_xxxxxx' \
  -H 'Content-Type: application/json' \
  -d '{
    "name": "Support Assistant",
    "custom_id": "support-assistant-v1",
    "description": "Handles customer support inquiries using knowledge base",
    "metadata": {
      "team": "customer-success",
      "tier": "production"
    },
    "model": "gpt-4o",
    "capabilities": {"web_search": {"open_datasets": ["research/academic"]}},
    "policies": {
      "max_turns": 20,
      "max_tool_calls": 10,
      "allow_tools": ["search_kb", "create_ticket"],
      "subagent_timeout": 300,
      "subagent_max_concurrency": 3
    },
    "output": {
      "mode": "text"
    },
    "subagents": [
      {
        "name": "ticket-creator",
        "mode": "fork",
        "context": "Create support tickets from conversation context",
        "inherit_tools": false
      }
    ],
    "system_prompt": "You are a helpful support assistant. Use the knowledge base to answer questions accurately.",
    "model_settings": {
      "temperature": 0.7,
      "max_tokens": 4096
    },
    "runtime_bindings": {
      "knowledge_base": "kb_prod_docs"
    }
  }'

{
  "success": true,
  "data": {
    "id": "agent-f8bf71ab17ed8eccad8c0e8ae708273f",
    "name": "Support Assistant",
    "description": "Handles customer support inquiries using knowledge base",
    "metadata": {
      "team": "customer-success",
      "tier": "production"
    },
    "version": 1,
    "model": "gpt-4o",
    "capabilities": {"web_search": {"max_results": 5, "search_depth": "basic", "chunks_per_source": 2, "open_datasets": ["research/academic"]}},
    "policies": {
      "max_turns": 20,
      "max_tool_calls": 10,
      "allow_tools": ["search_kb", "create_ticket"],
      "deny_tools": [],
      "subagent_timeout": 300,
      "subagent_max_concurrency": 3
    },
    "output": {
      "mode": "text",
      "json_schema": null
    },
    "subagents": [
      {
        "name": "ticket-creator",
        "mode": "fork",
        "context": "Create support tickets from conversation context",
        "inherit_tools": false
      }
    ],
    "skills": [],
    "custom_id": "support-assistant-v1",
    "actor_id": "actor-9ad7a492416c6a8197f42a64b0a4a2d2",
    "latest_version": 1,
    "system_prompt": "You are a helpful support assistant. Use the knowledge base to answer questions accurately.",
    "model_settings": {
      "temperature": 0.7,
      "max_tokens": 4096
    },
    "runtime_bindings": {
      "knowledge_base": "kb_prod_docs"
    },
    "created_at": "2025-03-10T09:00:00Z",
    "updated_at": "2025-03-10T09:00:00Z"
  }
}

curl -X POST 'https://app.memorylake.cn/openapi/memorylake/api/v3/agents' \
  -H 'Authorization: Bearer sk_xxxxxx' \
  -H 'Content-Type: application/json' \
  -d '{
    "name": "Support Assistant",
    "custom_id": "support-assistant-v1",
    "description": "Handles customer support inquiries using knowledge base",
    "metadata": {
      "team": "customer-success",
      "tier": "production"
    },
    "model": "gpt-4o",
    "capabilities": {"web_search": {"open_datasets": ["research/academic"]}},
    "policies": {
      "max_turns": 20,
      "max_tool_calls": 10,
      "allow_tools": ["search_kb", "create_ticket"],
      "subagent_timeout": 300,
      "subagent_max_concurrency": 3
    },
    "output": {
      "mode": "text"
    },
    "subagents": [
      {
        "name": "ticket-creator",
        "mode": "fork",
        "context": "Create support tickets from conversation context",
        "inherit_tools": false
      }
    ],
    "system_prompt": "You are a helpful support assistant. Use the knowledge base to answer questions accurately.",
    "model_settings": {
      "temperature": 0.7,
      "max_tokens": 4096
    },
    "runtime_bindings": {
      "knowledge_base": "kb_prod_docs"
    }
  }'

{
  "success": true,
  "data": {
    "id": "agent-f8bf71ab17ed8eccad8c0e8ae708273f",
    "name": "Support Assistant",
    "description": "Handles customer support inquiries using knowledge base",
    "metadata": {
      "team": "customer-success",
      "tier": "production"
    },
    "version": 1,
    "model": "gpt-4o",
    "capabilities": {"web_search": {"max_results": 5, "search_depth": "basic", "chunks_per_source": 2, "open_datasets": ["research/academic"]}},
    "policies": {
      "max_turns": 20,
      "max_tool_calls": 10,
      "allow_tools": ["search_kb", "create_ticket"],
      "deny_tools": [],
      "subagent_timeout": 300,
      "subagent_max_concurrency": 3
    },
    "output": {
      "mode": "text",
      "json_schema": null
    },
    "subagents": [
      {
        "name": "ticket-creator",
        "mode": "fork",
        "context": "Create support tickets from conversation context",
        "inherit_tools": false
      }
    ],
    "skills": [],
    "custom_id": "support-assistant-v1",
    "actor_id": "actor-9ad7a492416c6a8197f42a64b0a4a2d2",
    "latest_version": 1,
    "system_prompt": "You are a helpful support assistant. Use the knowledge base to answer questions accurately.",
    "model_settings": {
      "temperature": 0.7,
      "max_tokens": 4096
    },
    "runtime_bindings": {
      "knowledge_base": "kb_prod_docs"
    },
    "created_at": "2025-03-10T09:00:00Z",
    "updated_at": "2025-03-10T09:00:00Z"
  }
}

POST /openapi/memorylake/api/v3/agents

使用指定配置创建一个新的 Agent。你需要提供名称和自定义标识符，并可选地附加描述、模型、能力、策略、子 Agent 和其他设置。Agent 创建时为版本 1。

所需权限： agent:create

请求体

string

required

Agent 名称

string

required

这个Agent在您自己系统里已有的标识符。在您的租户内必须唯一。后续请求带上 by_custom_id=true，即可用它代替 id 来访问。参见您自己的标识符。

string

Agent 描述

string

平台是否托管这个 Agent 的运行时。INTERNAL（默认）在平台内运行，支持版本、fork 与 A2A；EXTERNAL 只登记记忆——Agent 运行在您那边，因此下面所有运行时字段都不适用。

object

用于存储 Agent 附加信息的任意键值对

string

Agent 使用的模型（例如 gpt-4o、claude-sonnet-4-20250514）

object

Agent 在每次对话中都具备的能力，键是能力名，值是该能力的配置。这里可以设置 web_search 和 web_fetch，写法与 A2A 消息里的 capabilitiesOverride 相同；pii、vega 和 dataset 只能按消息开启。不认识的能力名会被丢弃，不会报错。不传此字段时，Agent 两者都没有。

object

约束 Agent 行为的执行策略

Show Policies 对象

integer

每次运行的最大对话轮次数

integer

每轮的最大工具调用次数

array

Agent 可调用的工具白名单

array

Agent 不可调用的工具黑名单

integer

子 Agent 在被取消前可运行的最大秒数

integer

可同时运行的最大子 Agent 数量

object

Agent 的输出配置

Show Output 对象

string

输出模式（text 或 json）

object

当模式为 json 时，Agent 输出必须符合的 JSON Schema

array

要附加到此 Agent 的子 Agent 定义

Show Subagent 对象

string

子 Agent 名称

string

执行模式：fork 或 swarm

string

传递给子 Agent 的上下文指令

boolean

子 Agent 是否继承父 Agent 的工具

array

要挂到这个 Agent 上的技能。所有技能都必须可用：只要有一个未通过安全审核、不存在或还没有技能包，请求就以 400 和 SKILL_NOT_USABLE 失败，并逐个列出技能及原因。

Show Skill 引用

string

技能 ID

integer

把引用固定到这个版本。不传则跟随最新版本。固定版本的引用按该版本自己的审核结果判断。

string

用于指示 Agent 的系统提示词

object

模型特定设置，如温度和最大 token 数

object

工具和数据源的运行时绑定配置

curl -X POST 'https://app.memorylake.cn/openapi/memorylake/api/v3/agents' \
  -H 'Authorization: Bearer sk_xxxxxx' \
  -H 'Content-Type: application/json' \
  -d '{
    "name": "Support Assistant",
    "custom_id": "support-assistant-v1",
    "description": "Handles customer support inquiries using knowledge base",
    "metadata": {
      "team": "customer-success",
      "tier": "production"
    },
    "model": "gpt-4o",
    "capabilities": {"web_search": {"open_datasets": ["research/academic"]}},
    "policies": {
      "max_turns": 20,
      "max_tool_calls": 10,
      "allow_tools": ["search_kb", "create_ticket"],
      "subagent_timeout": 300,
      "subagent_max_concurrency": 3
    },
    "output": {
      "mode": "text"
    },
    "subagents": [
      {
        "name": "ticket-creator",
        "mode": "fork",
        "context": "Create support tickets from conversation context",
        "inherit_tools": false
      }
    ],
    "system_prompt": "You are a helpful support assistant. Use the knowledge base to answer questions accurately.",
    "model_settings": {
      "temperature": 0.7,
      "max_tokens": 4096
    },
    "runtime_bindings": {
      "knowledge_base": "kb_prod_docs"
    }
  }'

响应

object

Show Agent 对象

string

Agent 唯一标识符

string

Agent 名称

string

Agent 描述

object

用于存储 Agent 附加信息的任意键值对

integer

当前活跃版本号

string

Agent 使用的模型

object

Agent 具备的能力，键是能力名。平台会为您未设置的字段补上默认值，所以配置里可能出现本文档没有说明的字段。

object

约束 Agent 行为的执行策略

Show Policies 对象

integer

每次运行的最大对话轮次数

integer

每轮的最大工具调用次数

array

Agent 可调用的工具白名单

array

Agent 不可调用的工具黑名单

integer

子 Agent 在被取消前可运行的最大秒数

integer

可同时运行的最大子 Agent 数量

object

Agent 的输出配置

Show Output 对象

string

输出模式

object

结构化输出的 JSON Schema

array

附加到此 Agent 的子 Agent 定义

Show Subagent 对象

string

子 Agent 名称

string

执行模式：fork 或 swarm

string

传递给子 Agent 的上下文指令

boolean

子 Agent 是否继承父 Agent 的工具

array

附加到此 Agent 的技能

Show Skill 引用

string

技能标识符

integer

技能版本号

string

用于外部系统集成的自定义标识符

string

与此 Agent 关联的 Actor 标识符

integer

此 Agent 可用的最新版本号

string

用于指示 Agent 的系统提示词

object

模型特定设置，如温度和最大 token 数

object

工具和数据源的运行时绑定配置

string

创建时间戳

string

最后更新时间戳

{
  "success": true,
  "data": {
    "id": "agent-f8bf71ab17ed8eccad8c0e8ae708273f",
    "name": "Support Assistant",
    "description": "Handles customer support inquiries using knowledge base",
    "metadata": {
      "team": "customer-success",
      "tier": "production"
    },
    "version": 1,
    "model": "gpt-4o",
    "capabilities": {"web_search": {"max_results": 5, "search_depth": "basic", "chunks_per_source": 2, "open_datasets": ["research/academic"]}},
    "policies": {
      "max_turns": 20,
      "max_tool_calls": 10,
      "allow_tools": ["search_kb", "create_ticket"],
      "deny_tools": [],
      "subagent_timeout": 300,
      "subagent_max_concurrency": 3
    },
    "output": {
      "mode": "text",
      "json_schema": null
    },
    "subagents": [
      {
        "name": "ticket-creator",
        "mode": "fork",
        "context": "Create support tickets from conversation context",
        "inherit_tools": false
      }
    ],
    "skills": [],
    "custom_id": "support-assistant-v1",
    "actor_id": "actor-9ad7a492416c6a8197f42a64b0a4a2d2",
    "latest_version": 1,
    "system_prompt": "You are a helpful support assistant. Use the knowledge base to answer questions accurately.",
    "model_settings": {
      "temperature": 0.7,
      "max_tokens": 4096
    },
    "runtime_bindings": {
      "knowledge_base": "kb_prod_docs"
    },
    "created_at": "2025-03-10T09:00:00Z",
    "updated_at": "2025-03-10T09:00:00Z"
  }
}

MemoryLake API

Memory Router API

请求体

响应

​请求体

​响应

请求体

响应