Files
DevFlow/crates/df-ai-core/src/provider.rs

278 lines
10 KiB
Rust
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
//! LLM Provider trait — 统一的 LLM 调用抽象
//!
//! 支持 OpenAI 兼容 API覆盖 OpenAI / GLM / DeepSeek / Claude 兼容模式),
//! 含 function calling / tool use 能力。
//!
//! 本文件仅含 trait + 数据结构定义(零 IO。HTTP implOpenAICompatProvider /
//! AnthropicCompatProvider+ 业务逻辑ContextManager / AiToolRegistry /
//! build_provider 工厂)留在 df-ai crate。
use std::pin::Pin;
use async_trait::async_trait;
use futures::Stream;
use serde::{Deserialize, Serialize};
// ============================================================
// 核心数据结构
// ============================================================
/// LLM 调用请求
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct CompletionRequest {
/// 模型名称
pub model: String,
/// 提示消息列表
pub messages: Vec<ChatMessage>,
/// 温度0.0 ~ 2.0
pub temperature: Option<f32>,
/// 最大生成 token 数
pub max_tokens: Option<u32>,
/// 是否流式输出
pub stream: bool,
/// 可调用的工具定义
#[serde(skip_serializing_if = "Option::is_none")]
pub tools: Option<Vec<ToolDefinition>>,
/// 工具调用策略: "auto" | "none" | {"type":"function","name":"xxx"}
#[serde(skip_serializing_if = "Option::is_none")]
pub tool_choice: Option<serde_json::Value>,
}
/// 聊天消息
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct ChatMessage {
pub role: MessageRole,
pub content: String,
/// 工具调用 IDrole=Tool 时必填)
#[serde(skip_serializing_if = "Option::is_none")]
pub tool_call_id: Option<String>,
/// AI 发起的工具调用列表role=Assistant 时可能有)
#[serde(skip_serializing_if = "Option::is_none")]
pub tool_calls: Option<Vec<ToolCall>>,
/// 生成该消息的 model仅 assistant 消息有,消息级 model 追溯)
#[serde(default, skip_serializing_if = "Option::is_none")]
pub model: Option<String>,
/// 消息状态UX-09 编辑重生成None/"active" 正常可见;
/// "truncated" 软删(编辑某条 user 消息后其后续消息标记,保留 DB 可追溯但不进 LLM 上下文、前端视图过滤)
/// 默认 None向前兼容老 JSON 反序列化)。落库随 messages JSON 序列化,无需独立列。
#[serde(default, skip_serializing_if = "Option::is_none")]
pub status: Option<String>,
}
impl ChatMessage {
pub fn system(content: impl Into<String>) -> Self {
Self { role: MessageRole::System, content: content.into(), tool_call_id: None, tool_calls: None, model: None, status: None }
}
pub fn user(content: impl Into<String>) -> Self {
Self { role: MessageRole::User, content: content.into(), tool_call_id: None, tool_calls: None, model: None, status: None }
}
pub fn assistant(content: impl Into<String>) -> Self {
Self { role: MessageRole::Assistant, content: content.into(), tool_call_id: None, tool_calls: None, model: None, status: None }
}
pub fn assistant_with_tools(content: impl Into<String>, tool_calls: Vec<ToolCall>) -> Self {
Self { role: MessageRole::Assistant, content: content.into(), tool_call_id: None, tool_calls: Some(tool_calls), model: None, status: None }
}
pub fn tool_result(call_id: impl Into<String>, content: impl Into<String>) -> Self {
Self { role: MessageRole::Tool, content: content.into(), tool_call_id: Some(call_id.into()), tool_calls: None, model: None, status: None }
}
/// 是否处于 active 态status 为 None 或 "active")。其余状态一律 false。
///
/// 正面白名单F-15 §3.2):仅认 None / "active",新状态
/// 如阶段2 引入的 "archived_segment" / "compressed")自动落入不 active 分支,
/// 无需每加一个状态就来这里改。当前取值 None/Some("active")/Some("truncated")
/// 行为与旧反面排除完全等价None=true / "active"=true / "truncated"=false
pub fn is_active(&self) -> bool {
matches!(self.status.as_deref(), None | Some("active"))
}
}
/// 消息角色
#[derive(Debug, Clone, Serialize, Deserialize)]
#[serde(rename_all = "lowercase")]
pub enum MessageRole {
System,
User,
Assistant,
Tool,
}
/// 工具定义
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct ToolDefinition {
#[serde(rename = "type")]
pub tool_type: String,
pub function: ToolFunction,
}
impl ToolDefinition {
pub fn function(name: impl Into<String>, description: impl Into<String>, parameters: serde_json::Value) -> Self {
Self {
tool_type: "function".into(),
function: ToolFunction { name: name.into(), description: description.into(), parameters },
}
}
}
/// 函数定义
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct ToolFunction {
pub name: String,
pub description: String,
pub parameters: serde_json::Value,
}
/// 工具调用AI 发起)
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct ToolCall {
pub id: String,
#[serde(rename = "type")]
pub call_type: String,
pub function: ToolCallFunction,
}
impl ToolCall {
pub fn new(id: impl Into<String>, name: impl Into<String>, arguments: impl Into<String>) -> Self {
Self {
id: id.into(),
call_type: "function".into(),
function: ToolCallFunction { name: name.into(), arguments: arguments.into() },
}
}
}
/// 工具调用函数部分
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct ToolCallFunction {
pub name: String,
pub arguments: String,
}
/// LLM 调用响应
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct CompletionResponse {
/// 生成的文本
pub text: String,
/// 使用的模型
pub model: String,
/// 消耗的 token 数
pub usage: TokenUsage,
/// AI 发起的工具调用(如有)
#[serde(skip_serializing_if = "Option::is_none")]
pub tool_calls: Option<Vec<ToolCall>>,
}
/// Token 用量
#[derive(Debug, Clone, Default, Serialize, Deserialize)]
pub struct TokenUsage {
pub prompt_tokens: u32,
pub completion_tokens: u32,
pub total_tokens: u32,
}
/// 流式输出的 chunk
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct StreamChunk {
/// 增量文本
pub delta: String,
/// 是否结束
pub finished: bool,
/// 工具调用增量(如有)
#[serde(skip_serializing_if = "Option::is_none")]
pub tool_calls: Option<Vec<ToolCallDelta>>,
/// Token 用量(流末 chunk 携带,由 provider 解析自 SSE usage 事件)
#[serde(skip_serializing_if = "Option::is_none")]
pub usage: Option<TokenUsage>,
/// provider 流式错误事件(如 Anthropic SSE `type=="error"`)。
/// 非空表示流中途出错不应视为正常完成finished 路径),由 stream_llm 转 AiError。
#[serde(skip)]
pub error: Option<String>,
}
/// 工具调用增量(流式中的片段)
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct ToolCallDelta {
/// 索引
pub index: u32,
/// 工具调用 ID仅第一个 chunk 有)
#[serde(skip_serializing_if = "Option::is_none")]
pub id: Option<String>,
/// 函数名片段
#[serde(skip_serializing_if = "Option::is_none")]
pub function_name: Option<String>,
/// 函数参数片段
#[serde(skip_serializing_if = "Option::is_none")]
pub function_arguments: Option<String>,
}
/// 异步流类型别名
pub type StreamResult = Pin<Box<dyn Stream<Item = anyhow::Result<StreamChunk>> + Send>>;
/// LLM Provider trait
#[async_trait]
pub trait LlmProvider: Send + Sync {
/// 同步调用
async fn complete(&self, request: CompletionRequest) -> anyhow::Result<CompletionResponse>;
/// 流式调用(返回异步流)
async fn stream(
&self,
request: CompletionRequest,
) -> anyhow::Result<StreamResult>;
/// 文本嵌入:批量文本 → 语义向量(供知识库向量检索)
///
/// 默认实现返回 Err(协议不支持)。OpenAI 兼容协议覆盖实现(/v1/embeddings);
/// Anthropic 无 embedding API,保持默认。
async fn embed(&self, _model: &str, _texts: Vec<String>) -> anyhow::Result<Vec<Vec<f32>>> {
anyhow::bail!("该 Provider 不支持 embedding({})", self.name())
}
/// Provider 名称
fn name(&self) -> &str;
/// 实际请求端点(含 base_url + 关键路径,如 chat completions / messages
/// 默认回落 `name()`provider 实现覆盖返真实 URL供 401/网络错误诊断打印
/// —— 旧路径只能近似打印 provider_type看不到实际请求端点。
fn endpoint(&self) -> String {
self.name().to_string()
}
}
#[cfg(test)]
mod tests {
use super::*;
#[test]
fn is_active_whitelist() {
// F-15 §3.2 正面白名单:仅 None / "active" 为 true其余一律 false。
// 零行为变化None / "active" / "truncated" 与旧反面排除完全等价;
// "archived_segment" / "compressed"阶段2 引入,当前代码未赋值)由白名单
// matches! 只认 None/active 自动落入 false面向未来验证。
// None构造默认值向前兼容老 JSON
let m = ChatMessage::user("hi");
assert!(m.is_active(), "None 应 active");
// "active"
let mut m = ChatMessage::user("hi");
m.status = Some("active".to_string());
assert!(m.is_active(), "Some(active) 应 active");
// "truncated" — 当前取值与旧实现等价false
let mut m = ChatMessage::user("hi");
m.status = Some("truncated".to_string());
assert!(!m.is_active(), "truncated 应不 active");
// "archived_segment" — 阶段2 待引入,白名单自动隔离
let mut m = ChatMessage::user("hi");
m.status = Some("archived_segment".to_string());
assert!(!m.is_active(), "archived_segment 应不 active白名单隔离");
// "compressed" — 阶段2 待引入,白名单自动隔离
let mut m = ChatMessage::user("hi");
m.status = Some("compressed".to_string());
assert!(!m.is_active(), "compressed 应不 active白名单隔离");
}
}