Files
Cyrene_v2/internal/runtime/memory/extract.go
T
AskaEth 0624d9ee36
CI / Backend (build / vet / test / cross-compile) (push) Has been cancelled
CI / Frontend (tsc + vite build) (push) Has been cancelled
feat(session,runtime): 多角色并存—按会话选角(迁移0011 + 按回合解析 + agent池 + 提取器修正)
按 DevDocs/design/22 批次 1 后端(§2/§3/§4.1-4.5/§4.7/§5 验收 1-6)。
- 迁移 0011:chat_sessions.persona_id + chat_messages.persona_id(均 DEFAULT '',无回填 → 升级零风险);同步包内幂等 DDL;**真实库升级验证**(拷贝 data/cyrene.db 到临时目录跑 MigrateUp:既有 1 会话/109 消息 persona 仍为 '',原库未被触碰)
- history:SessionMeta/Message 加 PersonaID + SetSessionPersona;ListSessions/List/scan/Append 带上
- runtime:SessionEnv.PersonaID + loop.personaFor(ctx) 替换固定 l.personaID(memory/request/turn)+ **agent 池**(Options.Agents,按角色取 instructions,未命中回落)+ **提取器不再把 persona 烤死在构造期**(多角色下记忆不再永远记在 cyrene 名下)+ **thinking 旋钮改为按会话角色解析**(否则角色 Y 会用角色 X 的预算/展示)
- gateway:会话 DTO/POST/PATCH 支持 personaId;ws/rest/voice 的 OwnerEnv 硬编改为**按会话解析角色**(actor 元数据语义不变);admin memories 搜索去掉硬编 cyrene
- Docs/api-client v0.42;**两处有意的行为收紧**已写明:POST /sessions body 非法 JSON → 400(原先完全忽略 body;空/null/{} 仍 201);POST /admin/memories/search 省略 personaId 且无默认角色 → 400(原先默认查 cyrene)
- 测试:design/22 §5 验收 1-6(记忆隔离/提示词首条 system/情绪按角色/提取 persona 归属/迁移兼容)+ 进程内 e2e(断言 adapter 收到的 model.Request.Messages[0] 即该角色的 systemPrompt)
- **已知未做(留批次 2 / 待确认)**:① IM 与渠道级未做 → **§7-3「owner 私聊镜像随主会话角色」尚未生效**;② 请求级覆盖未实现(本批只做会话级);③ **§7-5 的「上下文发送角色标记」只完成存储+DTO 半边**——消息 persona 已落库且 DTO 已带,但模型上下文不从 chat_messages 重建,「第 N 条是谁说的」尚未进上下文(需确认是否属批次 2);④「切角色重建摘要」当前是空操作(压缩未接线);⑤ thinking 旋钮仅在流式路径生效(非流式 agent.RunStep 不带,A 批次既有性质)
- 验收:go build/vet/test 全绿 + linux/amd64 与 darwin/arm64 交叉编译(既有 flake 本轮未触发;同批 session_orchestrator.go 一度 501 行 → 抽出 session_tail.go)
2026-10-03 20:12:29 +08:00

95 lines
3.2 KiB
Go

package memory
import (
"context"
"errors"
"strings"
"git.yeij.top/AskaEth/Cyrene_v2/internal/runtime/model"
)
var (
// ErrNilAdapter is returned when an LLMExtractor is built or used without a
// model.Adapter to drive extraction.
ErrNilAdapter = errors.New("memory: nil model adapter")
)
// Extractor derives candidate memory items from a conversation (design §1).
//
// Two implementations exist:
//
// - LLMExtractor: asks a model.Adapter to denoise and deduplicate each turn,
// running synchronously (Extract) or on a background goroutine
// (ExtractAsync).
// - ManualExtractor: deterministic, rule-based — it only records what the user
// explicitly asks to remember ("记住 X").
type Extractor interface {
// Extract returns memory items worth persisting from conversation. ctx
// controls cancellation; a canceled context returns an error.
Extract(ctx context.Context, conversation []model.Message) ([]MemoryItem, error)
}
// ManualExtractor implements the manual "记住 X" extraction path without an LLM.
// It scans user turns for a leading remember directive and produces a memory
// item for each, falling back to no item for ordinary dialogue. It is the
// deterministic counterpart to the M4 LLM extractor.
type ManualExtractor struct {
personaID string
roleID string
}
// NewManualExtractor builds a ManualExtractor whose extracted items carry
// personaID/roleID as their author. Like the LLM extractor, the persona is a
// default rather than the authority: a multi-role caller builds it with a blank
// persona and stamps each item with the turn's role when persisting it
// (design/22 §4.2), so pass "" for a role-agnostic extractor.
func NewManualExtractor(personaID, roleID string) *ManualExtractor {
return &ManualExtractor{personaID: personaID, roleID: roleID}
}
// Extract returns one memory item (kind relationship, persona scope) per user
// turn that asks the agent to remember something. A leading "记住"/"请记住"
// directive is stripped from the stored content; turns without one are skipped.
func (e *ManualExtractor) Extract(ctx context.Context, conversation []model.Message) ([]MemoryItem, error) {
if err := ctx.Err(); err != nil {
return nil, err
}
items := make([]MemoryItem, 0)
for _, msg := range conversation {
if msg.Role != model.RoleUser {
continue
}
text := strings.TrimSpace(messageText(msg))
content, ok := rememberDirective(text)
if !ok {
continue
}
items = append(items, MemoryItem{
PersonaID: e.personaID,
RoleID: e.roleID,
Scope: ScopePersona,
Kind: KindRelationship,
Content: content,
Confidence: 1.0,
DecayWeight: 1.0,
})
}
return items, nil
}
// rememberDirective strips the "记住" / "请记住" prefix when text is a remember
// instruction. It returns the remembered content and true, or ok=false when the
// text is not an explicit remember request.
func rememberDirective(text string) (string, bool) {
for _, prefix := range []string{"请记住", "记住"} {
if strings.HasPrefix(text, prefix) {
content := strings.TrimSpace(strings.TrimPrefix(text, prefix))
if content == "" {
return "", false
}
return content, true
}
}
return "", false
}