Compare commits
2 Commits
355746f431
...
b62212d8bf
| Author | SHA1 | Date |
|---|---|---|
 朝夕
|
b62212d8bf | |
|
朝夕
|
a1adbd00af |
60
README.md
60
README.md
|
|
@ -5,8 +5,12 @@
|
|||
一款基于 Python 的分布式多 Agent 协作系统
|
||||
|
||||
[](https://www.python.org/)
|
||||
[](https://docs.ray.io/)
|
||||
[](https://ai.pydantic.dev/)
|
||||
[](LICENSE)
|
||||
|
||||
[**项目架构**](./docs/ARCHITECTURE.md) | [**更新日志**](./changelogs/CHANGELOG.md) | [**未来展望**](./changelogs/ROADMAP.md)
|
||||
|
||||
</div>
|
||||
|
||||
---
|
||||
|
|
@ -14,44 +18,30 @@
|
|||
|
||||
|
||||
---
|
||||
## 特别之处？
|
||||
## ✨ 核心特性
|
||||
|
||||
- 本项目通过 **Ray** 和 **vllm** 实现个人个性化助手的创建，你可以通过收集符合你自己偏好的数据，构建独属于自己的风格的人工智能助手。
|
||||
- 本项目通过多 Agent 协作，实现比起单 Agent 系统更强的性能，从而完成更复杂的任务。
|
||||
- 本项目通过调用本地模型，从而减少对于昂贵的API使用和一定程度上的安全保护。如果你是创作者，可以通过用自己的作品去训练属于自己的模型，提高工作效率。
|
||||
- （暂未实现）本项目适配多种消息平台，实现在外可通过多种方式给 **Pretor** 下达指令完成工作。
|
||||
- （暂未实现）本项目内置 **growth_node(生长节点)** ，实现傻瓜式微调模型操作，让你的 **Pretor** 自己学会一些独特的技能。
|
||||
### 🧠 异构协作体系
|
||||
- **多智能体集群**：内置主管 (Supervisory)、意识 (Consciousness)、控制 (Control) 三大核心节点，实现比单 Agent 系统更严谨的决策链。
|
||||
- **Worker 动态派生**：根据任务需求动态拉起 Ordinary 或 Skill 类型的 Worker Individual，实现资源的按需分配。
|
||||
|
||||
### 🚀 分布式性能保障
|
||||
- **Ray 驱动**：底层基于 Ray 构建，支持跨进程、跨机器的 Actor 通讯，轻松应对高并发任务流。
|
||||
- **本地化优先**：深度适配 **vLLM**，支持本地私有化模型部署，在保障隐私的同时大幅降低 API 调用成本。
|
||||
|
||||
### 🛠️ 工业级工程设计
|
||||
- **强类型契约**：基于 Pydantic-AI 实现 Tool 与 Agent 的接口定义，确保 AI 输出的确定性与安全性。
|
||||
- **自动化流**：内置工作流引擎 (Workflow Engine)，实现从需求发现到自动化执行的闭环。
|
||||
|
||||
### 📦 Pretor 生态子项目 (Sub-projects)
|
||||
|
||||
| 项目名称 | 代号 | 功能定位 | 当前状态 |
|
||||
|:-----------------------------------------------------------|:--------| :--- | :--- |
|
||||
| **[pretor-viceroy](https://github.com/zhaoxi826/viceroy)** | **总督** | **资源管理**：负责系统 Skill 的动态安装、元数据解析与全集群分发。 | ✅ 已发布 |
|
||||
| **pretor-stardomain** | **星域** | **安全沙箱**：为 Agent 自动生成的代码提供轻量化的隔离运行环境，防止逃逸。 | 📅 规划中 |
|
||||
| **pretor-explorer** | **探索者** | **网页感知**：自动化爬虫引擎，赋予智能体实时互联网信息搜索与内容抓取能力。 | 📅 规划中 |
|
||||
| **pretor-pioneer** | **先驱者** | **知识增强**：RAG 检索增强引擎，管理私有知识库的向量化、索引与精准检索。 | 📅 规划中 |
|
||||
|
||||
---
|
||||
## 快速开始
|
||||
本项目正在开发中...
|
||||
|
||||
---
|
||||
## 项目子项目
|
||||
#### pretor-viceroy:
|
||||
**项目名称** : 总督
|
||||
**github网址** :https://github.com/zhaoxi826/viceroy
|
||||
**功能** :pretor的资源管理工具
|
||||
**目前实现** :
|
||||
- 对于skill的安装与解析
|
||||
|
||||
#### pretor-domain
|
||||
**项目名称** : 域
|
||||
**github网址** :（暂无）
|
||||
**功能** : 轻量化沙箱
|
||||
**目前实现** :
|
||||
（规划中）
|
||||
|
||||
#### pretor-explorer
|
||||
**项目名称** : 探索者
|
||||
**github网址** :（暂无）
|
||||
**功能** : 自动爬虫
|
||||
**目前实现** :
|
||||
（规划中）
|
||||
|
||||
#### pretor-pioneer
|
||||
**项目名称** : 先驱者
|
||||
**github网址** :（暂无）
|
||||
**功能** : RAG检索增强生成
|
||||
**目前实现** :
|
||||
（规划中）
|
||||
|
|
@ -0,0 +1,15 @@
|
|||
# ChangeLog
|
||||
|
||||
---
|
||||
## [v0.1.0Alpha] - 2026/4/28
|
||||
### 更新：
|
||||
#### 🚀 新增功能 (Added)
|
||||
- **分布式 Actor 骨架**：基于 Ray 框架构建了多智能体协作底座，支持节点跨进程通讯与资源调度。
|
||||
- **全局状态机 (GSM)**：实现了 `GlobalStateMachine` 模块，作为系统的“唯一真相来源”，管理所有 Individual、Skill 和 Provider 的注册信息。
|
||||
- **核心认知节点**：
|
||||
- `SupervisoryNode`：负责任务拆解与分发。
|
||||
- `ConsciousnessNode`：负责意图识别与语义理解。
|
||||
- `ControlNode`：负责工作流状态监控与逻辑卡点。
|
||||
- **异步工作流引擎**：实现 `WorkflowRunningEngine`，支持从数据库自动轮询并异步执行待办任务流。
|
||||
- **自适应适配器**：集成 `Pydantic-AI`，并封装了统一的 `AbstractAgent` 协议，支持 OpenAI、Gemini 和 Claude 等多模型后端。
|
||||
- **基础设施代理**：建立 `PostgresDatabase` Actor，提供分布式的数据库连接池支持。
|
||||
|
|
@ -0,0 +1,17 @@
|
|||
# Roadmap
|
||||
|
||||
---
|
||||
## [v0.1.0Alpha] - 2026/4/28
|
||||
### 未来展望：
|
||||
#### 功能增加
|
||||
- [ ] **完善系统插件**: 如 **RAG(检索增强生成)**，**沙箱**， **联网搜索** ，使agent拥有更多的能力适应多样化任务需求
|
||||
- [ ] **增加MCP功能**: 增加MCP，使得agent可以调用通用工具
|
||||
- [ ] **完善special_individual**： 使得`supervisory_node`等可以调用实现语言生成图像生成等功能
|
||||
- [ ] **完善supervisory_node**: 实现`supervisory_node`对于工作流状态的访问，实现更方便的检测
|
||||
- [ ] **对消息平台的对接**: 完善platform，实现对于更多消息平台的对接（如：钉钉微信等），实现在社交软件对`supervisory_node`下达命令
|
||||
|
||||
#### 系统优化
|
||||
- [ ] **优化workflow逻辑**: 通过**graph**等设计实现更优秀的工作流调度
|
||||
- [ ] **优化GSM设计**: 对于 **GSM（global_state_machine全局状态机）** 进行重构，实现更高的并发
|
||||
- [ ] **工具及skill优化**: 完善前端获取工具或skill的逻辑，实现对于skill或者tool的配置改写以及详细信息获取
|
||||
- [ ] **前端优化**: 完善前端设置逻辑（如：调节语言等），以及使前端更加灵活智能
|
||||
|
|
@ -8,5 +8,5 @@
|
|||
- **意识节点**：负责复杂任务的处理；
|
||||
- **生长节点**：负责获取资源并且将基础模型训练为特化模型；
|
||||
- **特殊子个体**：与外界交互的模型，如embedding模型，tts模型等；
|
||||
- **专家子个体**：；
|
||||
- **专家子个体**：携带有专业skill的agent对象；
|
||||
- **基础子个体**：普通的agent对象；
|
||||
|
|
@ -14,12 +14,15 @@ interface WorkerIndividual {
|
|||
output_template?: string; // Change to string for the form state
|
||||
bound_skill?: string; // Change to string for the form state
|
||||
workspace?: string; // Change to string for the form state
|
||||
tools?: string; // Form state for tools JSON array
|
||||
}
|
||||
|
||||
export function WorkerIndividualSettings() {
|
||||
const [providers, setProviders] = useState<Provider[]>([]);
|
||||
const [workers, setWorkers] = useState<WorkerIndividual[]>([]);
|
||||
const [systemNodes, setSystemNodes] = useState<any[]>([]);
|
||||
const [availableSkills, setAvailableSkills] = useState<string[]>([]);
|
||||
const [availableTools, setAvailableTools] = useState<string[]>([]);
|
||||
const [loading, setLoading] = useState(true);
|
||||
const [error, setError] = useState('');
|
||||
|
||||
|
|
@ -32,14 +35,20 @@ export function WorkerIndividualSettings() {
|
|||
const fetchData = async () => {
|
||||
setLoading(true);
|
||||
try {
|
||||
const [provRes, workRes, sysRes] = await Promise.all([
|
||||
const [provRes, workRes, sysRes, toolsRes, skillsRes] = await Promise.all([
|
||||
apiClient.get('/api/v1/provider/list'),
|
||||
apiClient.get('/api/v1/agent/worker'),
|
||||
apiClient.get('/api/v1/agent')
|
||||
apiClient.get('/api/v1/agent'),
|
||||
apiClient.get('/api/v1/resource/tool'),
|
||||
apiClient.get('/api/v1/resource/skill')
|
||||
]);
|
||||
setProviders(Object.values(provRes.data.provider_list || {}));
|
||||
setWorkers(workRes.data.workers || []);
|
||||
|
||||
const allTools = toolsRes.data.tools ? Object.values(toolsRes.data.tools).flatMap(tGroup => Object.keys(tGroup as any)) : [];
|
||||
setAvailableTools(allTools);
|
||||
setAvailableSkills(Object.keys(skillsRes.data.skills || {}));
|
||||
|
||||
const sysNodesData = sysRes.data.system_nodes || [];
|
||||
const defaultSysNodes = ['supervisory_node', 'consciousness_node', 'control_node'];
|
||||
|
||||
|
|
@ -54,6 +63,7 @@ export function WorkerIndividualSettings() {
|
|||
agent_type: 'System Node',
|
||||
provider_title: found && found.provider_title ? found.provider_title : defaultProvider,
|
||||
model_id: found && found.model_id ? found.model_id : '',
|
||||
tools: found && found.tools ? JSON.stringify(found.tools) : '[]',
|
||||
is_system: true
|
||||
};
|
||||
});
|
||||
|
|
@ -75,7 +85,8 @@ export function WorkerIndividualSettings() {
|
|||
...worker,
|
||||
output_template: typeof worker.output_template === 'string' ? worker.output_template : JSON.stringify(worker.output_template || {}),
|
||||
bound_skill: typeof worker.bound_skill === 'string' ? worker.bound_skill : JSON.stringify(worker.bound_skill || {}),
|
||||
workspace: typeof worker.workspace === 'string' ? worker.workspace : JSON.stringify(worker.workspace || [])
|
||||
workspace: typeof worker.workspace === 'string' ? worker.workspace : JSON.stringify(worker.workspace || []),
|
||||
tools: typeof worker.tools === 'string' ? worker.tools : JSON.stringify(worker.tools || [])
|
||||
});
|
||||
setIsNew(false);
|
||||
setIsEditing(true);
|
||||
|
|
@ -92,7 +103,8 @@ export function WorkerIndividualSettings() {
|
|||
system_prompt: '',
|
||||
output_template: '{}',
|
||||
bound_skill: '{}',
|
||||
workspace: '[]'
|
||||
workspace: '[]',
|
||||
tools: '[]'
|
||||
});
|
||||
setIsNew(true);
|
||||
setIsEditing(true);
|
||||
|
|
@ -118,7 +130,8 @@ export function WorkerIndividualSettings() {
|
|||
const payload = {
|
||||
individual_name: editData.agent_name,
|
||||
provider_title: editData.provider_title,
|
||||
model_id: editData.model_id
|
||||
model_id: editData.model_id,
|
||||
tools: JSON.parse(editData.tools || '[]')
|
||||
};
|
||||
await apiClient.post('/api/v1/agent', payload);
|
||||
} else {
|
||||
|
|
@ -126,7 +139,8 @@ export function WorkerIndividualSettings() {
|
|||
...editData,
|
||||
output_template: JSON.parse(editData.output_template || '{}'),
|
||||
bound_skill: JSON.parse(editData.bound_skill || '{}'),
|
||||
workspace: JSON.parse(editData.workspace || '[]')
|
||||
workspace: JSON.parse(editData.workspace || '[]'),
|
||||
tools: JSON.parse(editData.tools || '[]')
|
||||
};
|
||||
|
||||
if (isNew) {
|
||||
|
|
@ -332,13 +346,27 @@ export function WorkerIndividualSettings() {
|
|||
/>
|
||||
</div>
|
||||
<div>
|
||||
<label className="block text-sm font-medium text-slate-700 mb-1">Bound Skill (JSON)</label>
|
||||
<textarea
|
||||
value={editData.bound_skill || '{}'}
|
||||
onChange={(e) => setEditData({...editData, bound_skill: e.target.value})}
|
||||
rows={3}
|
||||
className="w-full px-4 py-2 border border-slate-200 rounded-lg focus:ring-2 focus:ring-indigo-500 font-mono text-sm"
|
||||
/>
|
||||
<label className="block text-sm font-medium text-slate-700 mb-1">Bound Skill (Select)</label>
|
||||
<select
|
||||
value={(() => {
|
||||
try {
|
||||
const parsed = JSON.parse(editData.bound_skill || '{}');
|
||||
return Object.keys(parsed)[0] || '';
|
||||
} catch { return ''; }
|
||||
})()}
|
||||
onChange={(e) => {
|
||||
const val = e.target.value;
|
||||
const newSkill = val ? { [val]: [] } : {};
|
||||
setEditData({...editData, bound_skill: JSON.stringify(newSkill)});
|
||||
}}
|
||||
className="w-full px-4 py-2 border border-slate-200 rounded-lg focus:ring-2 focus:ring-indigo-500"
|
||||
disabled={editData.agent_type !== 'skill_individual'}
|
||||
>
|
||||
<option value="">No Skill Bound</option>
|
||||
{availableSkills.map(skill => (
|
||||
<option key={skill} value={skill}>{skill}</option>
|
||||
))}
|
||||
</select>
|
||||
</div>
|
||||
</div>
|
||||
|
||||
|
|
@ -354,6 +382,45 @@ export function WorkerIndividualSettings() {
|
|||
</>
|
||||
)}
|
||||
|
||||
<div>
|
||||
<label className="block text-sm font-medium text-slate-700 mb-1">Tools (Select Multiple)</label>
|
||||
<div className="flex flex-wrap gap-2 p-4 border border-slate-200 rounded-lg max-h-48 overflow-y-auto">
|
||||
{availableTools.map(tool => {
|
||||
let currentTools: string[] = [];
|
||||
try {
|
||||
currentTools = JSON.parse(editData.tools || '[]');
|
||||
} catch { currentTools = []; }
|
||||
|
||||
const isSelected = currentTools.includes(tool);
|
||||
return (
|
||||
<button
|
||||
key={tool}
|
||||
type="button"
|
||||
onClick={() => {
|
||||
let updatedTools = [...currentTools];
|
||||
if (isSelected) {
|
||||
updatedTools = updatedTools.filter(t => t !== tool);
|
||||
} else {
|
||||
updatedTools.push(tool);
|
||||
}
|
||||
setEditData({...editData, tools: JSON.stringify(updatedTools)});
|
||||
}}
|
||||
className={`px-3 py-1.5 text-sm rounded-full transition-colors ${
|
||||
isSelected
|
||||
? 'bg-indigo-100 text-indigo-700 border border-indigo-200'
|
||||
: 'bg-slate-50 text-slate-600 border border-slate-200 hover:bg-slate-100'
|
||||
}`}
|
||||
>
|
||||
{tool}
|
||||
</button>
|
||||
);
|
||||
})}
|
||||
{availableTools.length === 0 && (
|
||||
<span className="text-sm text-slate-500">No tools available</span>
|
||||
)}
|
||||
</div>
|
||||
</div>
|
||||
|
||||
{modalMessage && (
|
||||
<div className="p-3 bg-red-50 text-red-700 text-sm rounded-lg">
|
||||
{modalMessage}
|
||||
|
|
|
|||
|
|
@ -33,7 +33,8 @@ class AgentFactory:
|
|||
output_type: ResponseModel,
|
||||
system_prompt: str,
|
||||
deps_type: DepsModel,
|
||||
agent_name: str) -> Agent:
|
||||
agent_name: str,
|
||||
tools: list = None) -> Agent:
|
||||
"""
|
||||
create_agent方法，将输入的provider对象实例化为一个pydantic-ai的agent对象
|
||||
|
||||
|
|
@ -58,5 +59,6 @@ class AgentFactory:
|
|||
name=agent_name,
|
||||
system_prompt=system_prompt,
|
||||
output_type=output_type,
|
||||
deps_type=deps_type)
|
||||
deps_type=deps_type,
|
||||
tools=tools)
|
||||
return agent
|
||||
|
|
@ -30,10 +30,12 @@ class AgentRegister(BaseModel):
|
|||
provider_title: str
|
||||
model_id: str
|
||||
individual_name: str
|
||||
tools: Optional[List[str]] = None
|
||||
|
||||
class AgentLocalRegister(BaseModel):
|
||||
path: str
|
||||
individual_name: str
|
||||
tools: Optional[List[str]] = None
|
||||
|
||||
@agent_router.get("")
|
||||
async def get_system_nodes(_: TokenData = Depends(RoleChecker(allowed_roles=UserAuthority.USER))):
|
||||
|
|
@ -56,20 +58,21 @@ async def load_agent(agent_register: Union[AgentRegister, AgentLocalRegister],
|
|||
await postgres_database.upsert_system_node_config.remote(
|
||||
agent_register.individual_name,
|
||||
agent_register.provider_title,
|
||||
agent_register.model_id
|
||||
agent_register.model_id,
|
||||
agent_register.tools
|
||||
)
|
||||
|
||||
# Load agent into state machine
|
||||
match agent_register.individual_name:
|
||||
case "supervisory_node":
|
||||
node = ray_actor_hook("supervisory_node").supervisory_node
|
||||
await node.create_agent.remote(global_state_machine,agent_register.provider_title,agent_register.model_id)
|
||||
await node.create_agent.remote(global_state_machine,agent_register.provider_title,agent_register.model_id, agent_register.tools)
|
||||
case "consciousness_node":
|
||||
node = ray_actor_hook("consciousness_node").consciousness_node
|
||||
await node.create_agent.remote(global_state_machine,agent_register.provider_title,agent_register.model_id)
|
||||
await node.create_agent.remote(global_state_machine,agent_register.provider_title,agent_register.model_id, agent_register.tools)
|
||||
case "control_node":
|
||||
node = ray_actor_hook("control_node").control_node
|
||||
await node.create_agent.remote(global_state_machine,agent_register.provider_title,agent_register.model_id)
|
||||
await node.create_agent.remote(global_state_machine,agent_register.provider_title,agent_register.model_id, agent_register.tools)
|
||||
case _:
|
||||
pass
|
||||
except Exception as e:
|
||||
|
|
@ -87,6 +90,7 @@ class WorkerIndividualCreate(BaseModel):
|
|||
output_template: dict
|
||||
bound_skill: Dict[str, List[str]]
|
||||
workspace: List[str]
|
||||
tools: Optional[List[str]] = None
|
||||
|
||||
|
||||
class WorkerIndividualUpdate(BaseModel):
|
||||
|
|
@ -99,6 +103,7 @@ class WorkerIndividualUpdate(BaseModel):
|
|||
output_template: Optional[dict] = None
|
||||
bound_skill: Optional[Dict[str, List[str]]] = None
|
||||
workspace: Optional[List[str]] = None
|
||||
tools: Optional[List[str]] = None
|
||||
|
||||
|
||||
@agent_router.post("/worker")
|
||||
|
|
@ -143,8 +148,29 @@ async def update_worker_individual(agent_id: str,
|
|||
|
||||
update_data = worker_data.model_dump(exclude_unset=True)
|
||||
updated_worker = await postgres_database.update_worker_individual.remote( agent_id=agent_id, **update_data)
|
||||
|
||||
global_state_machine = ray_actor_hook("global_state_machine").global_state_machine
|
||||
try:
|
||||
await global_state_machine.remove_individual.remote(agent_id)
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
return {"message": "success", "worker": updated_worker}
|
||||
|
||||
@agent_router.post("/worker/{agent_id}/reload")
|
||||
async def reload_worker_individual(agent_id: str, token_data: TokenData = Depends(Accessor.get_current_user)):
|
||||
postgres_database = ray_actor_hook("postgres_database").postgres_database
|
||||
worker = await postgres_database.get_worker_individual.remote(agent_id=agent_id)
|
||||
if not worker:
|
||||
raise HTTPException(status_code=404, detail="Agent not found")
|
||||
if worker.owner_id != token_data.user_id:
|
||||
raise HTTPException(status_code=403, detail="Forbidden: You do not own this agent")
|
||||
|
||||
global_state_machine = ray_actor_hook("global_state_machine").global_state_machine
|
||||
await global_state_machine.remove_individual.remote(agent_id)
|
||||
|
||||
return {"message": "Worker will be reloaded on next use"}
|
||||
|
||||
|
||||
@agent_router.delete("/worker/{agent_id}")
|
||||
async def delete_worker_individual(agent_id: str,
|
||||
|
|
|
|||
|
|
@ -22,7 +22,7 @@ class SystemNodeDatabase:
|
|||
self.async_session_maker = async_session_maker
|
||||
|
||||
@database_exception
|
||||
async def upsert_system_node_config(self, node_name: str, provider_title: str, model_id: str) -> SystemNodeConfig:
|
||||
async def upsert_system_node_config(self, node_name: str, provider_title: str, model_id: str, tools: Optional[List[str]] = None) -> SystemNodeConfig:
|
||||
async with self.async_session_maker() as session:
|
||||
statement = select(SystemNodeConfig).where(SystemNodeConfig.node_name == node_name)
|
||||
results = await session.execute(statement)
|
||||
|
|
@ -30,8 +30,10 @@ class SystemNodeDatabase:
|
|||
if config:
|
||||
config.provider_title = provider_title
|
||||
config.model_id = model_id
|
||||
if tools is not None:
|
||||
config.tools = tools
|
||||
else:
|
||||
config = SystemNodeConfig(node_name=node_name, provider_title=provider_title, model_id=model_id)
|
||||
config = SystemNodeConfig(node_name=node_name, provider_title=provider_title, model_id=model_id, tools=tools)
|
||||
session.add(config)
|
||||
await session.commit()
|
||||
await session.refresh(config)
|
||||
|
|
|
|||
|
|
@ -106,9 +106,9 @@ class PostgresDatabase:
|
|||
return await self._provider_database.update_provider(provider_id, **kwargs)
|
||||
|
||||
# System Node Database Methods
|
||||
async def upsert_system_node_config(self, node_name: str, provider_title: str, model_id: str):
|
||||
async def upsert_system_node_config(self, node_name: str, provider_title: str, model_id: str, tools: list[str] = None):
|
||||
await self.ready_event.wait()
|
||||
return await self._system_node_database.upsert_system_node_config(node_name, provider_title, model_id)
|
||||
return await self._system_node_database.upsert_system_node_config(node_name, provider_title, model_id, tools)
|
||||
|
||||
async def get_all_system_node_configs(self):
|
||||
await self.ready_event.wait()
|
||||
|
|
|
|||
|
|
@ -34,4 +34,5 @@ class WorkerIndividual(SQLModel, table=True):
|
|||
output_template: Optional[dict] = Field(sa_column=Column(JSON),description="输出模板标识")
|
||||
bound_skill: Optional[str] = Field(sa_column=Column(JSON))
|
||||
workspace: Optional[List[str]] = Field(sa_column=Column(JSON))
|
||||
tools: Optional[List[str]] = Field(sa_column=Column(JSON), default=None)
|
||||
owner_id: str
|
||||
|
|
@ -14,8 +14,12 @@
|
|||
|
||||
from sqlmodel import SQLModel, Field
|
||||
|
||||
from typing import List, Optional
|
||||
from sqlalchemy import Column, JSON
|
||||
|
||||
class SystemNodeConfig(SQLModel, table=True):
|
||||
__tablename__ = "system_node_config"
|
||||
node_name: str = Field(primary_key=True)
|
||||
provider_title: str
|
||||
model_id: str
|
||||
tools: Optional[List[str]] = Field(sa_column=Column(JSON), default=None)
|
||||
|
|
|
|||
|
|
@ -22,8 +22,6 @@ from pretor.core.global_state_machine.global_state_machine import GlobalStateMac
|
|||
from pretor.core.global_state_machine.model_provider.base_provider import Provider
|
||||
from pretor.adapter.model_adapter.agent_factory import AgentFactory
|
||||
|
||||
from pretor.utils.get_tool import get_tool
|
||||
|
||||
|
||||
@ray.remote
|
||||
class ConsciousnessNode:
|
||||
|
|
@ -33,7 +31,7 @@ class ConsciousnessNode:
|
|||
self.agent: None | Agent = None
|
||||
|
||||
|
||||
async def create_agent(self, global_state_machine: GlobalStateMachine, provider_title: str, model_id: str) -> None:
|
||||
async def create_agent(self, global_state_machine: GlobalStateMachine, provider_title: str, model_id: str, tools_list: list[str] = None) -> None:
|
||||
"""
|
||||
create_agent方法，将agent对象装配到ConsciousnessNode的属性内
|
||||
该方法通过provider_title从global_state_machine中获取provider对象，然后从provider对象中取出供应商形象，装配为pydantic_ai的
|
||||
|
|
@ -57,14 +55,18 @@ class ConsciousnessNode:
|
|||
"请确保所有的思考和生成过程符合逻辑，严密且高质量。"
|
||||
)
|
||||
output_type = Union[ForSupervisoryNode, ForWorkflow, ForWorkflowEngine]
|
||||
from pretor.utils.get_tool import load_tools_from_list
|
||||
provider: Provider = await global_state_machine.get_provider.remote( provider_title)
|
||||
agent_factory = AgentFactory()
|
||||
|
||||
callables = load_tools_from_list(tools_list)
|
||||
self.agent = agent_factory.create_agent(provider=provider,
|
||||
model_id=model_id,
|
||||
output_type=output_type,
|
||||
system_prompt=system_prompt,
|
||||
deps_type=ConsciousnessNodeDeps,
|
||||
agent_name="consciousness_node")
|
||||
agent_name="consciousness_node",
|
||||
tools=callables)
|
||||
|
||||
@self.agent.system_prompt
|
||||
async def dynamic_prompt(ctx: RunContext[ConsciousnessNodeDeps]):
|
||||
|
|
@ -134,7 +136,6 @@ class ConsciousnessNode:
|
|||
async def _run(self, payload: Union[ForSupervisoryInput, ForWorkflowInput, ForWorkflowEngineInput]) -> Union[ForSupervisoryNode, ForWorkflow, ForWorkflowEngine]:
|
||||
try:
|
||||
self.agent.retries = 3
|
||||
tool = await get_tool("control_node")
|
||||
if isinstance(payload, ForWorkflowEngineInput):
|
||||
deps = ConsciousnessNodeDeps(
|
||||
original_command=payload.original_command,
|
||||
|
|
@ -144,8 +145,7 @@ class ConsciousnessNode:
|
|||
self.logger.debug("ConsciousnessNode: 开始生成工作流 (原生重试开启)")
|
||||
result = await self.agent.run(
|
||||
"根据original_command制定严密的可执行workflow，可以学习并参考workflow_template的设计理念",
|
||||
deps=deps,
|
||||
tools=tool)
|
||||
deps=deps)
|
||||
return result.output
|
||||
|
||||
elif isinstance(payload, ForWorkflowInput):
|
||||
|
|
@ -155,8 +155,7 @@ class ConsciousnessNode:
|
|||
)
|
||||
self.logger.debug("ConsciousnessNode: 开始处理工作流节点任务 (原生重试开启)")
|
||||
result = await self.agent.run(f"处理此工作流步骤信息：\n{payload.workflow_step.model_dump_json()}",
|
||||
deps=deps,
|
||||
tools=tool)
|
||||
deps=deps)
|
||||
return result.output
|
||||
|
||||
elif isinstance(payload, ForSupervisoryInput):
|
||||
|
|
@ -166,8 +165,7 @@ class ConsciousnessNode:
|
|||
)
|
||||
self.logger.debug("ConsciousnessNode: 开始生成技术总结报告 (原生重试开启)")
|
||||
result = await self.agent.run(f"基于以下工作流的执行记录，生成技术报告：\n{payload.workflow.model_dump_json()}",
|
||||
deps=deps,
|
||||
tools=tool)
|
||||
deps=deps)
|
||||
return result.output
|
||||
except Exception as e:
|
||||
self.logger.exception(f"ConsciousnessNode 模型生成最终失败: {str(e)}")
|
||||
|
|
|
|||
|
|
@ -18,7 +18,6 @@ from pretor.core.global_state_machine.global_state_machine import GlobalStateMac
|
|||
from pretor.core.global_state_machine.model_provider.base_provider import Provider
|
||||
from pretor.adapter.model_adapter.agent_factory import AgentFactory
|
||||
from pretor.core.individual.control_node.template import ForWorkflow, ForWorkflowInput, ControlNodeDeps
|
||||
from pretor.utils.get_tool import get_tool
|
||||
|
||||
|
||||
|
||||
|
|
@ -30,7 +29,7 @@ class ControlNode:
|
|||
self.agent: Agent | None = None
|
||||
|
||||
|
||||
async def create_agent(self, global_state_machine: GlobalStateMachine, provider_title: str, model_id: str) -> None:
|
||||
async def create_agent(self, global_state_machine: GlobalStateMachine, provider_title: str, model_id: str, tools_list: list[str] = None) -> None:
|
||||
"""
|
||||
create_agent方法，将agent对象装配到Control的属性内
|
||||
该方法通过provider_title从global_state_machine中获取provider对象，然后从provider对象中取出供应商形象，装配为pydantic_ai的
|
||||
|
|
@ -54,14 +53,18 @@ class ControlNode:
|
|||
"请注意：你的输出应当具体、实用，直接提供任务所要求的结果，不要做过多无关的寒暄。"
|
||||
)
|
||||
output_type = ForWorkflow
|
||||
from pretor.utils.get_tool import load_tools_from_list
|
||||
provider: Provider = await global_state_machine.get_provider.remote( provider_title)
|
||||
agent_factory = AgentFactory()
|
||||
|
||||
callables = load_tools_from_list(tools_list)
|
||||
self.agent = agent_factory.create_agent(provider=provider,
|
||||
model_id=model_id,
|
||||
output_type=output_type,
|
||||
system_prompt=system_prompt,
|
||||
deps_type=ControlNodeDeps,
|
||||
agent_name="control_node")
|
||||
agent_name="control_node",
|
||||
tools=callables)
|
||||
@self.agent.system_prompt
|
||||
async def dynamic_prompt(ctx: RunContext[ControlNodeDeps]):
|
||||
prompt = system_prompt + "\n\n"
|
||||
|
|
@ -89,12 +92,9 @@ class ControlNode:
|
|||
)
|
||||
self.logger.debug(f"ControlNode: 开始执行工作流节点 [{payload.workflow_step.name}] (原生重试开启)")
|
||||
|
||||
tool = await get_tool("control_node")
|
||||
|
||||
result = await self.agent.run(
|
||||
f"请根据提供的 workflow_step 上下文，执行此步骤并输出结果。\n详细指令或附加数据：{payload.workflow_step.model_dump_json()}",
|
||||
deps=deps,
|
||||
tools=tool
|
||||
deps=deps
|
||||
)
|
||||
return result.output
|
||||
except Exception as e:
|
||||
|
|
|
|||
|
|
@ -22,7 +22,6 @@ from pretor.core.global_state_machine.model_provider import Provider
|
|||
from pretor.core.individual.supervisory_node.template import ForConsciousnessNode, ForUser, SupervisoryNodeDeps, TerminationMessage
|
||||
from pydantic_ai import RunContext, Agent
|
||||
from pretor.utils.ray_hook import ray_actor_hook
|
||||
from pretor.utils.get_tool import get_tool
|
||||
|
||||
|
||||
@ray.remote
|
||||
|
|
@ -33,7 +32,7 @@ class SupervisoryNode:
|
|||
self.agent: None | Agent = None
|
||||
|
||||
|
||||
async def create_agent(self, global_state_machine: GlobalStateMachine, provider_title: str, model_id: str) -> None:
|
||||
async def create_agent(self, global_state_machine: GlobalStateMachine, provider_title: str, model_id: str, tools_list: list[str] = None) -> None:
|
||||
"""
|
||||
create_agent方法，将agent对象装配到SupervisoryNode的属性内
|
||||
该方法通过provider_title从global_state_machine中获取provider对象，然后从provider对象中取出供应商形象，装配为pydantic_ai的Agent实例，
|
||||
|
|
@ -57,14 +56,18 @@ class SupervisoryNode:
|
|||
"请保持冷静、专业，并严格遵循上述路由规则。"
|
||||
)
|
||||
output_type = Union[ForConsciousnessNode, ForUser]
|
||||
from pretor.utils.get_tool import load_tools_from_list
|
||||
provider: Provider = await global_state_machine.get_provider.remote( provider_title)
|
||||
agent_factory = AgentFactory()
|
||||
|
||||
callables = load_tools_from_list(tools_list)
|
||||
self.agent = agent_factory.create_agent(provider=provider,
|
||||
model_id=model_id,
|
||||
output_type=output_type,
|
||||
system_prompt=system_prompt,
|
||||
deps_type=SupervisoryNodeDeps,
|
||||
agent_name="supervisory_node")
|
||||
agent_name="supervisory_node",
|
||||
tools=callables)
|
||||
|
||||
@self.agent.system_prompt
|
||||
async def dynamic_prompt(ctx: RunContext[SupervisoryNodeDeps]):
|
||||
|
|
@ -76,6 +79,13 @@ class SupervisoryNode:
|
|||
f"- 当前时间 (Time): {ctx.deps.time}\n"
|
||||
f"- 可用工作流模板 (Available Templates): {ctx.deps.available_templates}\n"
|
||||
)
|
||||
# 修改 system_prompt 变量
|
||||
prompt += (
|
||||
"\n\n注意：你必须调用且只能调用一个函数（工具）来输出结果。"
|
||||
"如果你想直接回复用户，请调用 ForUser；"
|
||||
"如果你想移交给工作流，请调用 ForConsciousnessNode。"
|
||||
"严禁返回纯文本，必须使用工具格式！"
|
||||
)
|
||||
if ctx.deps.error_history:
|
||||
prompt += (
|
||||
f"\n=== 错误重试指示 ===\n"
|
||||
|
|
@ -172,10 +182,8 @@ class SupervisoryNode:
|
|||
if isinstance(payload, TerminationMessage):
|
||||
prompt_message = f"【工作流执行结束报告】\n请将以下技术报告转化为对用户的友好回复：\n{message}"
|
||||
self.agent.retries = 3
|
||||
tool = await get_tool("supervisory_node")
|
||||
result = await self.agent.run(prompt_message,
|
||||
deps=deps,
|
||||
tools=tool)
|
||||
deps=deps)
|
||||
return result.output
|
||||
except Exception as e:
|
||||
self.logger.exception(f"SupervisoryNode 模型生成或解析最终失败: {str(e)}")
|
||||
|
|
|
|||
|
|
@ -12,7 +12,9 @@
|
|||
# See the License for the specific language governing permissions and
|
||||
# limitations under the License.
|
||||
|
||||
import importlib
|
||||
import importlib.util
|
||||
import os
|
||||
import sys
|
||||
from typing import Callable, Dict, List
|
||||
import pathlib
|
||||
from pretor.utils.ray_hook import ray_actor_hook
|
||||
|
|
@ -20,50 +22,59 @@ from pretor.utils.ray_hook import ray_actor_hook
|
|||
from pretor.utils.logger import get_logger
|
||||
logger = get_logger('get_tool')
|
||||
_tool_cache: Dict[str, Callable] = {}
|
||||
_agent_tool_result_cache: Dict[str, List[Callable]] = {}
|
||||
|
||||
|
||||
def _get_tool_func(tool_name: str) -> Callable | None:
|
||||
func = _tool_cache.get(tool_name, None)
|
||||
if func:
|
||||
return func
|
||||
tool_plugin_dir = pathlib.Path(__file__).parent.parent.parent / "plugin" / "tool_plugin" / tool_name
|
||||
if not tool_plugin_dir.exists() or not tool_plugin_dir.is_dir():
|
||||
|
||||
app_root = "/app"
|
||||
tool_plugin_dir = os.path.join(app_root, "pretor", "plugin", "tool_plugin", tool_name)
|
||||
|
||||
if not os.path.exists(tool_plugin_dir) or not os.path.isdir(tool_plugin_dir):
|
||||
logger.error(f"Tool directory not found: {tool_plugin_dir}")
|
||||
return None
|
||||
module_name = f"pretor.plugin.tool_plugin.{tool_name}"
|
||||
|
||||
init_file = os.path.join(tool_plugin_dir, "__init__.py")
|
||||
if not os.path.exists(init_file):
|
||||
logger.error(f"Tool init file not found: {init_file}")
|
||||
return None
|
||||
|
||||
try:
|
||||
module = importlib.import_module(module_name)
|
||||
func = getattr(module, tool_name)
|
||||
module_name = f"pretor.plugin.tool_plugin.{tool_name}"
|
||||
spec = importlib.util.spec_from_file_location(module_name, init_file)
|
||||
if spec is None or spec.loader is None:
|
||||
logger.error(f"Failed to create spec for {module_name}")
|
||||
return None
|
||||
|
||||
module = importlib.util.module_from_spec(spec)
|
||||
sys.modules[module_name] = module
|
||||
spec.loader.exec_module(module)
|
||||
|
||||
func = getattr(module, tool_name, None)
|
||||
|
||||
if not callable(func):
|
||||
logger.error(f"Tool function '{tool_name}' not found or not callable in {module_name}")
|
||||
return None
|
||||
_tool_cache[tool_name] = func
|
||||
return func
|
||||
except ModuleNotFoundError:
|
||||
logger.error(f"Module {module_name} not found")
|
||||
except Exception as e:
|
||||
logger.error(f"Failed to load module {module_name}: {e}")
|
||||
return None
|
||||
|
||||
def del_tool_cache(tool_name: str) -> None:
|
||||
if tool_name in _tool_cache:
|
||||
del _tool_cache[tool_name]
|
||||
refresh_agent_tools()
|
||||
|
||||
def load_tools_from_list(tool_names: List[str] | None) -> List[Callable]:
|
||||
if not tool_names:
|
||||
return []
|
||||
|
||||
async def get_tool(agent_name: str) -> List[Callable]:
|
||||
cached = _agent_tool_result_cache.get(agent_name)
|
||||
if cached is not None:
|
||||
return cached
|
||||
global_state_machine = ray_actor_hook("global_state_machine").global_state_machine
|
||||
_tool_list = await global_state_machine.get_tool_list.remote(agent_name)
|
||||
tool_list = []
|
||||
for tool_name in _tool_list.keys():
|
||||
for tool_name in tool_names:
|
||||
tool_func = _get_tool_func(tool_name)
|
||||
if tool_func:
|
||||
tool_list.append(tool_func)
|
||||
else:
|
||||
continue
|
||||
_agent_tool_result_cache[agent_name] = tool_list
|
||||
return tool_list
|
||||
|
||||
def refresh_agent_tools() -> None:
|
||||
_agent_tool_result_cache.clear()
|
||||
return tool_list
|
||||
|
|
@ -42,19 +42,25 @@ class BaseIndividual:
|
|||
self.agent: Agent | None = None
|
||||
|
||||
async def _init_agent(self, agent_name: str, system_prompt: str):
|
||||
from pretor.utils.get_tool import load_tools_from_list
|
||||
global_state_machine = ray_actor_hook("global_state_machine").global_state_machine
|
||||
provider_title = self.agent_config.get("provider_title", "openai") # default fallback
|
||||
model_id = self.agent_config.get("model_id", "gpt-4o") # default fallback
|
||||
tools_list = self.agent_config.get("tools", None)
|
||||
|
||||
provider: Provider = await global_state_machine.get_provider.remote( provider_title)
|
||||
agent_factory = AgentFactory()
|
||||
|
||||
callables = load_tools_from_list(tools_list)
|
||||
|
||||
self.agent = agent_factory.create_agent(
|
||||
provider=provider,
|
||||
model_id=model_id,
|
||||
output_type=WorkerIndividualResponse,
|
||||
system_prompt=system_prompt,
|
||||
deps_type=WorkerIndividualDeps,
|
||||
agent_name=agent_name
|
||||
agent_name=agent_name,
|
||||
tools=callables
|
||||
)
|
||||
|
||||
@self.agent.system_prompt
|
||||
|
|
|
|||
|
|
@ -30,7 +30,8 @@ def test_create_agent_success_real():
|
|||
name="myagent",
|
||||
system_prompt="You are an AI",
|
||||
output_type=str,
|
||||
deps_type=dict
|
||||
deps_type=dict,
|
||||
tools=None
|
||||
)
|
||||
assert agent == mock_agent_cls.return_value
|
||||
|
||||
|
|
|
|||
|
|
@ -57,8 +57,10 @@ async def test_change_password_success(mock_session_maker, mock_dependencies):
|
|||
mock_statement = MagicMock()
|
||||
mock_select.return_value.where.return_value = mock_statement
|
||||
|
||||
from pretor.utils.access import Accessor
|
||||
|
||||
mock_user = MagicMock()
|
||||
mock_user.hashed_password = "old_password"
|
||||
mock_user.hashed_password = Accessor.hash_password("old_password")
|
||||
|
||||
mock_exec_result = MagicMock()
|
||||
mock_exec_result.scalar_one_or_none.return_value = mock_user
|
||||
|
|
@ -95,8 +97,9 @@ async def test_change_password_wrong_password(mock_session_maker, mock_dependenc
|
|||
maker, session = mock_session_maker
|
||||
db = AuthDatabase(maker)
|
||||
|
||||
from pretor.utils.access import Accessor
|
||||
mock_user = MagicMock()
|
||||
mock_user.hashed_password = "actual_password"
|
||||
mock_user.hashed_password = Accessor.hash_password("actual_password")
|
||||
mock_exec_result = MagicMock()
|
||||
mock_exec_result.scalar_one_or_none.return_value = mock_user
|
||||
session.execute = AsyncMock(return_value=mock_exec_result)
|
||||
|
|
|
|||
Loading…
Reference in New Issue