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)
84 lines
3.0 KiB
Go
84 lines
3.0 KiB
Go
package main
|
|
|
|
import (
|
|
"net"
|
|
"net/url"
|
|
"strings"
|
|
|
|
"git.yeij.top/AskaEth/Cyrene_v2/internal/runtime/memory"
|
|
"git.yeij.top/AskaEth/Cyrene_v2/internal/runtime/model"
|
|
)
|
|
|
|
// memoryOwnerRoleID is the role id the composition root authors extracted
|
|
// memories with; the owner is the only actor whose private memory is written
|
|
// (group/stranger conversations never extract at all, design/11 §2).
|
|
const memoryOwnerRoleID = "owner"
|
|
|
|
// buildMemoryExtractor selects the memory extractor the loop runs after each
|
|
// settled turn (design/03 §1 LLM 异步自动提取):
|
|
//
|
|
// - the LLM extractor when the provider stack has a usable active group, so
|
|
// the model denoises and deduplicates what is worth remembering;
|
|
// - the deterministic ManualExtractor otherwise, so a keyless deployment still
|
|
// records the owner's explicit "记住 X" directives instead of failing a model
|
|
// call on every turn.
|
|
//
|
|
// The extractor deliberately holds NO persona: one process now answers as several
|
|
// roles (design/22 §4.2), so a persona baked in here would file every extracted
|
|
// memory under that one role. The loop stamps each item with the role that
|
|
// produced the turn when it persists it (loop.persistMemories).
|
|
//
|
|
// The loop detaches the LLM extractor to a background goroutine (ExtractAsync);
|
|
// see internal/runtime/loop/memory.go.
|
|
func (a *app) buildMemoryExtractor() memory.Extractor {
|
|
if llmGroupUsable(a.llm) {
|
|
a.log.Info("memory extractor: llm", "model", a.llm.Model(), "provider", a.llm.Provider())
|
|
return memory.NewLLMExtractor(a.llm, "", memoryOwnerRoleID)
|
|
}
|
|
a.log.Warn("memory extractor: manual (no usable llm model group)")
|
|
return memory.NewManualExtractor("", memoryOwnerRoleID)
|
|
}
|
|
|
|
// llmGroupUsable reports whether the active model group of the runtime stack can
|
|
// actually be called: a model and a base URL are mandatory, and either an API key
|
|
// is configured or the endpoint is a local/private address, where a keyless call
|
|
// is the normal case (a self-hosted OpenAI-compatible server).
|
|
//
|
|
// It is deliberately a cheap structural check, never a probe request: boot must
|
|
// not spend a model call. A group that looks usable but fails at call time only
|
|
// logs an extraction error (extraction failures are never fatal).
|
|
func llmGroupUsable(llm *model.RuntimeLLM) bool {
|
|
if llm == nil {
|
|
return false
|
|
}
|
|
group := llm.Group()
|
|
if strings.TrimSpace(group.Model) == "" || strings.TrimSpace(group.BaseURL) == "" {
|
|
return false
|
|
}
|
|
if strings.TrimSpace(group.APIKey) != "" {
|
|
return true
|
|
}
|
|
return keylessEndpoint(group.BaseURL)
|
|
}
|
|
|
|
// keylessEndpoint reports whether baseURL points at a loopback or private-network
|
|
// host, where a missing API key is expected rather than a misconfiguration.
|
|
func keylessEndpoint(baseURL string) bool {
|
|
parsed, err := url.Parse(strings.TrimSpace(baseURL))
|
|
if err != nil {
|
|
return false
|
|
}
|
|
host := parsed.Hostname()
|
|
if host == "" {
|
|
return false
|
|
}
|
|
if strings.EqualFold(host, "localhost") {
|
|
return true
|
|
}
|
|
ip := net.ParseIP(host)
|
|
if ip == nil {
|
|
return false
|
|
}
|
|
return ip.IsLoopback() || ip.IsPrivate()
|
|
}
|