C
Callisto
Astra Cross-Dept AI WorkforceAstra 跨部门 · AI 数字员工
Astra Tech / Botim · Cross-department AI digital workforce · 2026 Astra Tech / Botim · Astra 跨部门的 AI 数字员工 · 2026

Callisto

Callisto — Jupiter's brightest moon木卫四 · 木星最亮的卫星

AI-led. Human-supervised.AI 为主力，人做监督。

Callisto is Astra's cross-department AI digital workforce — not just customer service. The core model: the AI works under an agent's supervision and grows into that agent's digital avatar, with every agent training their own AI. Phase 1 lands in phone support: it understands speech and text, reasons over enterprise data, and acts — answer, look up, raise tickets, place outbound calls — while people handle the exceptions. Callisto 是 Astra 的跨部门 AI 数字员工——不止于客服。核心模型:AI 在坐席监督下工作、逐步成为坐席的数字化身,每位坐席训练自己的 AI。第一期落在电话客服:它能听懂(语音/文字)、会思考(结合企业数据)、能动手(应答、查系统、建工单、外呼),人只处理需要人的例外。

North Star北极星 A single Agent definition — reused across voice, outbound, Botim Call and email, for both inbound and outbound. Define once, deploy everywhere: Quantix first, the same capability reused across Botim and other lines. 一次定义 Agent——跨语音/外呼/Botim Call/邮件复用,入呼与外呼两种方式规模化上岗。一次定义、跨渠道跨业务复用:一期在 Quantix 落地,同一套能力复用到 Botim 及其他业务线。

2.6–2.9s
Measured AI response latency实测 AI 应答延迟
7×24
Always-on, seconds to answer全天候,秒级响应
In + Out
Inbound & outbound verified E2E入呼+外呼 端到端验证
L1→L2
AI-autonomy maturity today当前 AI 自主度
Progress · de-risked当前进展 · 已降风险

A verified prototype, with Phase-2 capabilities already working已验证的原型,二期能力已提前跑通

The end-to-end loop — AI answers, understands, hands off to a human with live AI assist, and is configurable — is already proven. This cycle pulled several Phase-2 and agent-desk capabilities forward and live-tested them, so Phases 1–2 are demonstrated, not just planned.全链路已跑通:AI 接听 → 理解 → 转人工(人工有 AI 实时辅助)→ 可视化配置。本轮把多项二期与工作台能力提前落地并真机验证,一二期从"计划"变成"已见成效"。

AI outboundAI 外呼
Auto-dial and speak a message (collections / notifications), plus a full AI-conversation mode.按文本自动拨打并语音播报(催收/通知),另支持全 AI 对话外呼。
Phase 1
Outbound orchestration外呼编排
Batch campaigns, pacing & concurrency, automatic retry, Do-Not-Call list.批量任务、节流并发、失败自动重试、免打扰(DNC)名单。
Phase 2 →
Unified call history (CDR)统一通话历史(CDR)
Inbound and outbound in one auditable record — filter, search, export.入呼与外呼统一话单,可筛选、搜索、导出。
Desk
Post-call disposition话后处置
The AI auto-wraps contained calls; agents override and add a note.AI 自动结案,人工可改写并加备注。
Desk
Customer 360 / screen-pop客户 360 / 来电弹屏
Caller identity, tier, order history and past calls, surfaced by phone number.按号码弹出客户画像:等级、标签、历史订单与通话。
Desk
Ticket auto-raise工单自动创建
Every call produces a ticket (call record + outcome), idempotent by design.每通电话自动建工单(含通话记录与结果),设计上幂等去重。
Desk
Roadmap · four phases总体规划 · 四期

From voice, outward across channels and autonomy从语音起步,向渠道与自主度两端扩展

Each phase ships a usable business capability — widening channels (voice → Botim Call → email/Nora) and autonomy (answer under supervision → AI-drafts-actions-human-confirms → autonomous outbound → self-optimizing).每期都交付可用的业务能力——扩展渠道(语音 → Botim Call → 邮件/Nora)与自主度(监督下应答 → AI 生成操作·人确认 → 自主外呼 → 自我优化)。

Phase 1 NOW当前8 / 31

AI Phone · QuantixAI 电话客服 · Quantix

  • Quantix Etisalat line connectedQuantix Etisalat 线路连通
  • AI answers + handles Quantix business (data integration; the human trains the AI's behavior)AI 接听 + Quantix 业务处理(数据接入;需 human 训练 AI 的行为)
  • AI-to-human transferAI 转人工接通
  • Agent monitors the AI live, can take over anytime人工实时监听 AI,可随时主动接管
  • Agent management console人工客服管理平台
  • Outbound API — text playback (collections / notifications)外呼接口 — 文本自动播报(催收/通知)
  • Customer 360 · screen-pop (tier, tags, order & call history)客户 360 · 来电弹屏(等级、标签、历史订单与通话)
AI answers with agents monitoring live and taking over anytime. 7×24.AI 接听,人工实时监督、随时接管 · 7×24。
Phase 29 / 30

Botim Call + AI takes actionBotim Call + AI 执行操作

  • Botim Call voice channel接入 Botim Call 通话渠道
  • Humanized voice interaction拟人化语音交互
  • Outbound orchestration — batch, pacing, retry, DNC外呼编排 — 批量、节流、重试、DNC
  • AI drafts action commands; human confirms, then executeAI 生成操作指令,human 确认后执行
  • Proactively surfaces relevant history to the caller主动提及历史数据,询问是否为此咨询
  • Above 95% accuracy → AI acts autonomously, human reviews after准确率 > 95% 时 AI 自主操作,human 事后监督
The AI becomes each agent's supervised digital avatar; every agent trains their own AI.AI 是坐席监督下的数字化身;每位坐席训练自己的 AI。
Phase 310 / 31

Pure AI outbound + memory纯 AI 外呼 + 记忆

  • Pure AI outbound: the AI converses to reach the goal纯 AI 外呼:AI 与用户对话拿到需要的结果
  • Cross-session, cross-system data sharing & memory跨会话、跨系统的数据共享与记忆
  • Reuse to Botim & other lines (boundary TBD)复用到 Botim 及其他业务线(边界待定)
AI serves and reaches out autonomously; human supervises.AI 自主服务与触达,人监督。
Phase 42027

Proactive + self-optimizing主动服务 + 自我优化

  • Proactive business insight; proactively contacts users主动业务洞察,主动联系用户
  • Self-built ticketing, fused with email (Nora)自建工单系统,和邮件 Nora 融合
  • Human sets goals; AI auto-QA, self-learning, self-orchestration人定目标愿景;AI 自动质检、自动学习、自主编排
Omnichannel loop + AI self-optimization.全渠道闭环 + AI 自我优化。
Growth · four axes成长 · 四条主线

How the AI Agent growsAI Agent 沿四条主线成长

Each axis is independently measurable and independently reportable.每条主线都可独立度量、独立汇报。

01Reach渠道广度

"Define once, deploy everywhere." Inbound voice → outbound → Botim Call → email (Nora) → (later) more business lines."一次定义,处处上岗。"语音入呼 → 外呼 → Botim Call → 邮件(Nora)→(远期)更多业务线。

02Autonomy自主深度

From "can talk" to "gets things done, anticipates": single reply → multi-turn task closure → cross-session memory + proactive service → goal-driven.从"会说"到"会办、会预判":单轮应答 → 多轮任务闭环 → 跨会话记忆 + 主动服务 → 目标驱动。

03Intelligence智能厚度

Rules + LLM → RAG (sourced answers) → real-time sentiment / intent → automated QA → self-curating knowledge → AI coach for human agents.规则+LLM → RAG 检索(答案有出处)→ 实时情绪/意图 → 自动质检 → 知识自动沉淀 → 给人工坐席的 AI 教练。

04Data flywheel数据飞轮

Every conversation → structured records → business insight (complaints, recovery, churn) → feeds product / risk / marketing → a better Agent. It gets smarter with use.每通对话 → 结构化数据 → 业务洞察(投诉、回款、流失)→ 反哺产品/风控/营销 → 更好的 Agent。越用越聪明。

Maturity · a yardstick for autonomy成熟度 · 自主度标尺

How much of the human's job the AI takes — L0 to L4AI 替人到什么程度 —— L0 到 L4

By analogy to self-driving levels. Phase 1 locks in a stable L2; Phases 2–3 push L3; L4 is the North Star.类比自动驾驶分级。一期锁定稳定 L2;二三期推进 L3;L4 为北极星。

L0

AI transcribes / suggests onlyAI 仅转录 / 提建议

Everything runs through people.一切经由人工。

Human: fully manual人:全人工Legacy传统客服
L1

AI answers simple FAQs, human backstopAI 应答简单 FAQ,人工兜底

The AI handles the easy questions; people catch the rest.AI 处理简单问题,其余由人接住。

Human: primary人:主力Phase 1 start一期起步
You are here — between L1 and L2当前位置 —— L1 与 L2 之间
L2

AI closes most transactions, human handles exceptionsAI 闭环多数事务,人处理例外

Lookup, ticketing, outbound playback — all handled end to end by the AI.查单、建单、外呼播报——AI 端到端完成。

Human: supervise + exceptions人:监督 + 例外Phase 1 target一期目标
L3

Autonomous across channels & sessions + proactive outreach跨渠道跨会话自主服务 + 主动触达

The AI serves and reaches out on its own; people supervise.AI 自主服务与触达,人监督。

Human: supervise人:监督Phase 2 / 3二 / 三期
L4

AI self-optimizes — auto-QA, self-learning, self-orchestrationAI 自我优化 —— 自动质检、自动学习、自主编排

People set goals and guardrails; the AI runs the rest.人只定目标与边界,其余交给 AI。

Human: set goals人:定目标Vision愿景
Value · what we report on价值 · 汇报指标

Quantifiable, commit-able可量化,可对赌

Instrumented from Phase 1 — call history, disposition and customer-360 already provide the data spine. Prove value with data, not anecdotes.一期即埋点——话单、处置、客户 360 已具备数据基础。用数据而非感觉证明价值。

Dimension维度 Metrics指标
Cost降本AI self-resolution rate · cost per call · agent-equivalents replacedAI 独立解决率 · 单通成本 · 等效替代坐席数
Efficiency提效Average handle time · first-contact resolution · queue / wait time平均处理时长 · 首解率 · 排队/等待时长
Revenue增收Collections recovery rate · renewal / upsell conversion外呼催收回收率 · 续费/营销转化率
Experience体验Seconds-to-answer · 7×24 coverage · CSAT秒级接通 · 7×24 覆盖率 · CSAT
Quality质量Automated-QA compliance · escalation rate · repeat-call rate自动质检合规率 · 升级率 · 复呼率
Decisions · we need leadership待决策 · 需管理层拍板

Four calls to make四个需要拍板的问题

A

Yeastar phone system — keep it?Yeastar 电话系统是否继续沿用

Whether the existing Yeastar system can continue to be used — gates the Phase-1 Etisalat line setup.现有 Yeastar 系统是否可以继续使用——一期前置项,影响 Etisalat 线路方案。

B

Cross-line reuse boundary跨业务复用边界

The boundary and priority of reusing the Quantix build for Botim and other lines (Phase 3, TBD).Quantix 先行,复用到 Botim 及其他业务线的边界与优先级(三期,待定)。

C

Data & compliance数据与合规

Access, masking and retention for order / collections / customer-360 data.订单/催收/客户 360 数据的接入权限、脱敏与留存策略。

D

AI autonomy thresholdAI 自主操作阈值

At >95% accuracy the AI acts on its own with human review after — which operations always need human sign-off first?准确率 > 95% 时 AI 自主操作、human 事后监督——哪些操作须始终人工先确认?