Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
34 changes: 34 additions & 0 deletions .changeset/docs-align-retired-agents.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,34 @@
---
'hotcrm': patch
---

Docs alignment: correct the metadata counts in the README, rewrite the AI copilot
pages as skill docs, and disclaim the archived Salesforce comparison.

The README advertised "2 AI agents (sales-copilot, service-copilot), 20 flows,
5 sharing rules" three paragraphs above a repository-layout block that correctly
called the AI surface skills-only with 23 flows. The real numbers, taken from
`objectstack.config.ts`, are **0 app-owned agents, 6 AI skills, 23 flows and 9
sharing rules**; both README passages now say that, and the `8 datasets` count was
added so the two lists agree.

The published `ai-copilot/sales-copilot` and `ai-copilot/service-copilot` pages (en,
zh-Hans, zh-Hant) still described two AI personas that were retired in #512 — "it
lives in the right-side chat panel and as inline buttons" for capabilities that ship
today as six skills on the platform `ask` assistant. They are rewritten as skill docs
at their existing URLs, with the activation rules taken from each skill's real
`triggerConditions` and the write paths named as the actual HotCRM actions
(`convert_lead`, `schedule_followup`, `escalate_case`, `close_case`) the skills call.
The AI Copilot index, the Skills page, the glossary's "Agent (AI)" entry, the
customization guide (which told developers to edit a long-deleted
`src/agents/sales-copilot.agent.ts`) and every inbound link label were updated to
match.

`docs/archive/2026-02/SALESFORCE_FEATURE_COMPARISON.md` now opens with a
retired-architecture banner. It describes a ~148-object, 13-package product and marks
`sla_policy`, `queue`, `email_to_case` and `pricebook` as implemented — none exist —
so its "~95% Salesforce parity" headline was actively misleading anyone evaluating the
repo. Finally `docs/ARCHITECTURE.md` was corrected: the diagram no longer draws the
deleted `src/agents` or the never-created `src/cubes`, the manifest table reads
`2.2.2` instead of `1.0.5`, and `requires` no longer lists the `ai` capability that
was removed in 2.2.0.
4 changes: 2 additions & 2 deletions README.md
Original file line number Diff line number Diff line change
Expand Up @@ -5,7 +5,7 @@
# HotCRM

> **The reference app for AI-written enterprise software.** A complete CRM —
> 15 objects, 20 flows, 4 dashboards, 2 AI copilots, 4 languages — is roughly
> 15 objects, 23 flows, 4 dashboards, 6 AI skills, 4 languages — is roughly
> **170k tokens** of typed [ObjectStack](https://github.com/objectstack-ai/objectstack)
> metadata (~18,000 lines): the entire enterprise CRM fits in a single agent
> context window, so an AI can hold it whole, reason about it, and refactor it.
Expand Down Expand Up @@ -57,7 +57,7 @@ HotCRM is a complete, opinionated CRM built as the **first official application*
| `crm_product` | | | |
| `crm_forecast` | | | |

Plus **2 AI agents** (sales-copilot, service-copilot), **4 dashboards**, **20 flows**, **13 actions**, **6 AI skills**, **4 language bundles** (en, zh-CN, es-ES, ja-JP), **6 permission profiles**, **12 positions**, and **5 sharing rules**.
Plus **6 AI skills** (a skills-only surface — HotCRM defines no agents of its own; the skills attach to the platform `ask` assistant), **4 dashboards**, **23 flows**, **13 actions**, **8 datasets**, **4 language bundles** (en, zh-CN, es-ES, ja-JP), **6 permission profiles**, **12 positions**, and **9 sharing rules**.

> **Business reader?** The ObjectStack docs tour every one of these capabilities in plain business language — [What Can It Do?](https://objectstack.ai/docs/capabilities) — with HotCRM as the running example on every page.

Expand Down
49 changes: 30 additions & 19 deletions content/docs/ai-copilot/index.mdx
Original file line number Diff line number Diff line change
Expand Up @@ -5,47 +5,58 @@ description: The AI assistant embedded throughout HotCRM — what it can do, whe

# AI Copilot

HotCRM is **AI-native** — the assistant is built in, not bolted on. Two specialised copilots ship out of the box:
HotCRM is **AI-native** — the assistant is built in, not bolted on. But the assistant
itself is not something HotCRM ships: it is the **platform assistant** (`ask`) that
every ObjectStack environment already provides. What HotCRM contributes is **six
skills** that teach that assistant your CRM.

- The **Sales Copilot** helps reps qualify leads, draft emails, forecast revenue, and get a 360° view of accounts.
- The **Service Copilot** helps agents triage cases, see the full customer history, and draft customer responses.
> **No app-owned agents.** HotCRM used to define two agents of its own — a "Sales
> Copilot" and a "Service Copilot". They were retired, and their capability moved into
> skills that attach to the platform assistant. Everything on this page still works;
> only the wiring changed.

Both run on top of your CRM data — they read from the same objects you see in the UI, and they know your products, knowledge bases, and process rules.
The six skills read from the same objects you see in the UI, and they know your
products, knowledge bases, and process rules.

## Where you'll find it

- 🪄 **Copilot panel** — a chat panel that opens from the right edge of every page.
- **Inline actions** — buttons like *"Qualify this lead"*, *"Draft reply"*, *"Summarize case"* on record detail pages.
- 📋 **Suggestions** — automatic insights in dashboards (*"3 deals are at risk this week"*).
- 🪄 **Assistant panel** — the chat panel the platform opens from the right edge of every page.
- 🎯 **Context activation** — skills declare the record they belong to, so opening a lead and asking *"is this worth a call?"* runs Lead Qualification without you naming it.
- 📈 **Dashboard questions** — the forecasting skill also activates on dashboards, so you can ask *"what's at risk this week?"* where the numbers already are.

## How it works (in plain terms)

The Copilot has three ingredients:
The assistant has three ingredients:

1. **Skills** — bite-sized abilities like *Draft an email*, *Qualify this lead*, *Triage this case*. See [Skills](/docs/ai-copilot/skills).
2. **Knowledge bases** — searchable libraries of product info, sales playbooks, support articles, and competitive intel. See [Knowledge Bases](/docs/ai-copilot/knowledge-bases).
3. **Your live CRM data** — the Copilot reads from the same accounts, contacts, opportunities, and cases you do, respecting the same sharing rules.
3. **Your live CRM data** — the assistant reads from the same accounts, contacts, opportunities, and cases you do, respecting the same sharing rules.

When you ask a question, the Copilot picks the right skill, pulls relevant knowledge, and combines it with your CRM data to produce a grounded answer — with citations back to the source records and articles.
When you ask a question, the assistant picks the right skill, pulls relevant
knowledge, and combines it with your CRM data to produce a grounded answer — with
citations back to the source records and articles.

## What it won't do

- ❌ See records you can't see — it respects sharing and field-level security.
- ❌ Take actions without your confirmation — drafted emails and updates are previewed first.
- ❌ Fabricate data — every claim is traceable to a record or knowledge source.
- ❌ Reach for tools of its own — skills declare no bespoke tools, so every write goes through a real HotCRM action with the same permission checks as the button.

## Two flavours
## Two groupings of skills

| Copilot | Audience | Built-in skills |
| --- | --- | --- |
| [**Sales Copilot**](/docs/ai-copilot/sales-copilot) | Sales reps, sales managers | Lead qualification, email drafting, revenue forecasting, 360° customer view |
| [**Service Copilot**](/docs/ai-copilot/service-copilot) | Service agents, support managers | Case triage, 360° customer view, email drafting |
The six skills are documented in two role-shaped groups. This is a documentation
convenience, not a runtime boundary — one assistant runs all six, and it will use a
"sales" skill on a service question whenever that is the right answer.

Both copilots can call each other's skills when relevant — for example, the Service Copilot can use the Sales Copilot's *Customer 360°* skill to show what the customer has bought.
| Group | Audience | Skills |
| --- | --- | --- |
| [**Sales Skills**](/docs/ai-copilot/sales-copilot) | Sales reps, sales managers | Live Data, Lead Qualification, Email Drafting, Revenue Forecasting, Customer 360° |
| [**Service Skills**](/docs/ai-copilot/service-copilot) | Service agents, support managers | Live Data, Case Triage, Customer 360°, Email Drafting |

## Start here

- [Sales Copilot](/docs/ai-copilot/sales-copilot) — what reps and managers can do.
- [Service Copilot](/docs/ai-copilot/service-copilot) — what agents can do.
- [Sales Skills](/docs/ai-copilot/sales-copilot) — what reps and managers can do.
- [Service Skills](/docs/ai-copilot/service-copilot) — what agents can do.
- [Skills](/docs/ai-copilot/skills) — the full skill library.
- [Knowledge Bases](/docs/ai-copilot/knowledge-bases) — the RAG indexes the Copilot reads from.
- [Knowledge Bases](/docs/ai-copilot/knowledge-bases) — the RAG indexes the assistant reads from.
44 changes: 25 additions & 19 deletions content/docs/ai-copilot/index.zh-Hans.mdx
Original file line number Diff line number Diff line change
Expand Up @@ -5,47 +5,53 @@ description: 嵌入 HotCRM 各处的 AI 助手——它能做什么、在哪里

# AI Copilot

HotCRM 是 **AI 原生**的——助手是内置的,而非外挂的。开箱即用地提供两个专门的 Copilot:
HotCRM 是 **AI 原生**的——助手是内置的,而非外挂的。但助手本身并非由 HotCRM 提供:它
是每个 ObjectStack 环境已经自带的 **平台助手**(`ask`)。HotCRM 贡献的是 **六项技
能**,让这个助手学会你的 CRM。

- **销售 Copilot** 帮助销售代表判定线索资格、撰写邮件、预测收入,并获得客户的 360° 全景视图。
- **服务 Copilot** 帮助客服坐席分流工单、查看完整的客户历史,并撰写客户回复。
> **没有应用自有的智能体。** HotCRM 曾定义过两个自己的智能体——"销售 Copilot"和"服务
> Copilot"。它们已被退役,其能力迁移到了附着于平台助手的技能中。本页描述的一切依然有
> 效;变的只是接线方式。

两者都运行在你的 CRM 数据之上——它们从你在 UI 中看到的同一批对象读取数据,并且了解你的产品、知识库和流程规则。
这六项技能从你在 UI 中看到的同一批对象读取数据,并且了解你的产品、知识库和流程规则。

## 你会在哪里找到它

- 🪄 **Copilot 面板**——一个从每个页面右侧边缘打开的聊天面板
- **内联操作**——记录详情页上的按钮,如*"判定此线索资格"*、*"撰写回复"*、*"总结工单"*
- 📋 **建议**——仪表盘中的自动洞察(*"本周有 3 笔交易存在风险"*)
- 🪄 **助手面板**——平台从每个页面右侧边缘打开的聊天面板
- 🎯 **上下文激活**——技能声明了自己所属的记录,因此打开一条线索并问 *"这条值得打个电话吗?"*,无需点名就会运行线索资格判定
- 📈 **在仪表盘上提问**——收入预测技能也会在仪表盘上激活,你可以就地问 *"本周有什么风险?"*

## 它如何运作(通俗地说)

Copilot 有三大要素
助手有三大要素

1. **技能**——细分到点的能力,如*撰写邮件*、*判定此线索资格*、*分流此工单*。参见 [技能](/zh-Hans/docs/ai-copilot/skills)。
2. **知识库**——可搜索的产品信息、销售剧本、支持文章和竞争情报库。参见 [知识库](/zh-Hans/docs/ai-copilot/knowledge-bases)。
3. **你的实时 CRM 数据**——Copilot 从你看到的同一批客户、联系人、商机和工单读取数据,并遵循相同的共享规则。
3. **你的实时 CRM 数据**——助手从你看到的同一批客户、联系人、商机和工单读取数据,并遵循相同的共享规则。

当你提出问题时,Copilot 会选择合适的技能、拉取相关知识,并将其与你的 CRM 数据结合,生成有依据的答案——并附带回溯到源记录和文章的引用。
当你提出问题时,助手会选择合适的技能、拉取相关知识,并将其与你的 CRM 数据结合,生成有
依据的答案——并附带回溯到源记录和文章的引用。

## 它不会做什么

- ❌ 查看你看不到的记录——它遵循共享和字段级安全。
- ❌ 未经你确认就执行操作——撰写好的邮件和更新会先预览。
- ❌ 编造数据——每一条声明都可追溯到一条记录或一个知识来源。
- ❌ 使用自带的专属工具——技能不声明任何专属工具,每一次写入都经由真实的 HotCRM 操作完成,权限校验与点击按钮时完全一致。

## 两种形态
## 技能的两种分组

| Copilot | 受众 | 内置技能 |
| --- | --- | --- |
| [**销售 Copilot**](/zh-Hans/docs/ai-copilot/sales-copilot) | 销售代表、销售经理 | 线索资格判定、邮件撰写、收入预测、360° 客户视图 |
| [**服务 Copilot**](/zh-Hans/docs/ai-copilot/service-copilot) | 服务坐席、支持经理 | 工单分流、360° 客户视图、邮件撰写 |
六项技能按角色分成两组来编写文档。这只是文档上的便利,而非运行时边界——同一个助手运行
全部六项技能,只要"销售"技能是正确答案,它也会用它来回答服务问题。

两个 Copilot 在相关时可以互相调用对方的技能——例如,服务 Copilot 可以使用销售 Copilot 的 *Customer 360°* 技能来展示客户购买过什么。
| 分组 | 受众 | 技能 |
| --- | --- | --- |
| [**销售技能**](/zh-Hans/docs/ai-copilot/sales-copilot) | 销售代表、销售经理 | 实时数据、线索资格判定、邮件撰写、收入预测、Customer 360° |
| [**服务技能**](/zh-Hans/docs/ai-copilot/service-copilot) | 服务坐席、支持经理 | 实时数据、工单分流、Customer 360°、邮件撰写 |

## 从这里开始

- [销售 Copilot](/zh-Hans/docs/ai-copilot/sales-copilot)——销售代表和经理能做什么。
- [服务 Copilot](/zh-Hans/docs/ai-copilot/service-copilot)——坐席能做什么。
- [销售技能](/zh-Hans/docs/ai-copilot/sales-copilot)——销售代表和经理能做什么。
- [服务技能](/zh-Hans/docs/ai-copilot/service-copilot)——坐席能做什么。
- [技能](/zh-Hans/docs/ai-copilot/skills)——完整的技能库。
- [知识库](/zh-Hans/docs/ai-copilot/knowledge-bases)——Copilot 读取的 RAG 索引。
- [知识库](/zh-Hans/docs/ai-copilot/knowledge-bases)——助手读取的 RAG 索引。
44 changes: 25 additions & 19 deletions content/docs/ai-copilot/index.zh-Hant.mdx
Original file line number Diff line number Diff line change
Expand Up @@ -5,47 +5,53 @@ description: 嵌入 HotCRM 各處的 AI 助手——它能做什麼、在哪裡

# AI Copilot

HotCRM 是 **AI 原生**的——助手是內建的,而非外掛的。開箱即用地提供兩個專門的 Copilot:
HotCRM 是 **AI 原生**的——助手是內建的,而非外掛的。但助手本身並非由 HotCRM 提供:它
是每個 ObjectStack 環境已經自帶的 **平台助手**(`ask`)。HotCRM 貢獻的是 **六項技
能**,讓這個助手學會你的 CRM。

- **銷售 Copilot** 幫助銷售代表判定潛在客戶資格、撰寫郵件、預測收入,並獲得客戶的 360° 全景視圖。
- **服務 Copilot** 幫助客服坐席分流工單、查看完整的客戶歷史,並撰寫客戶回覆。
> **沒有應用自有的智能體。** HotCRM 曾定義過兩個自己的智能體——「銷售 Copilot」和「服
> 務 Copilot」。它們已被退役,其能力遷移到了附著於平台助手的技能中。本頁描述的一切依
> 然有效;變的只是接線方式。

兩者都執行在你的 CRM 資料之上——它們從你在 UI 中看到的同一批物件讀取資料,並且了解你的產品、知識庫和流程規則。
這六項技能從你在 UI 中看到的同一批物件讀取資料,並且了解你的產品、知識庫和流程規則。

## 你會在哪裡找到它

- 🪄 **Copilot 面板**——一個從每個頁面右側邊緣開啟的聊天面板
- **內聯操作**——記錄詳情頁上的按鈕,如*「判定此潛在客戶資格」*、*「撰寫回覆」*、*「總結工單」*
- 📋 **建議**——儀表板中的自動洞察(*「本週有 3 筆交易存在風險」*)
- 🪄 **助手面板**——平台從每個頁面右側邊緣開啟的聊天面板
- 🎯 **情境啟用**——技能宣告了自己所屬的記錄,因此打開一筆潛在客戶並問 *「這筆值得打個電話嗎?」*,無需點名就會執行潛在客戶資格判定
- 📈 **在儀表板上提問**——收入預測技能也會在儀表板上啟用,你可以就地問 *「本週有什麼風險?」*

## 它如何運作(通俗地說)

Copilot 有三大要素
助手有三大要素

1. **技能**——細分到點的能力,如*撰寫郵件*、*判定此潛在客戶資格*、*分流此工單*。參見 [技能](/zh-Hant/docs/ai-copilot/skills)。
2. **知識庫**——可搜尋的產品資訊、銷售劇本、支援文章和競爭情報庫。參見 [知識庫](/zh-Hant/docs/ai-copilot/knowledge-bases)。
3. **你的即時 CRM 資料**——Copilot 從你看到的同一批客戶、聯絡人、商機和工單讀取資料,並遵循相同的共享規則。
3. **你的即時 CRM 資料**——助手從你看到的同一批客戶、聯絡人、商機和工單讀取資料,並遵循相同的共享規則。

當你提出問題時,Copilot 會選擇合適的技能、拉取相關知識,並將其與你的 CRM 資料結合,生成有依據的答案——並附帶回溯到來源記錄和文章的引用。
當你提出問題時,助手會選擇合適的技能、拉取相關知識,並將其與你的 CRM 資料結合,生成有
依據的答案——並附帶回溯到來源記錄和文章的引用。

## 它不會做什麼

- ❌ 查看你看不到的記錄——它遵循共享和欄位級安全。
- ❌ 未經你確認就執行操作——撰寫好的郵件和更新會先預覽。
- ❌ 編造資料——每一條聲明都可追溯到一條記錄或一個知識來源。
- ❌ 使用自帶的專屬工具——技能不宣告任何專屬工具,每一次寫入都經由真實的 HotCRM 操作完成,權限校驗與點擊按鈕時完全一致。

## 兩種形態
## 技能的兩種分組

| Copilot | 受眾 | 內建技能 |
| --- | --- | --- |
| [**銷售 Copilot**](/zh-Hant/docs/ai-copilot/sales-copilot) | 銷售代表、銷售經理 | 潛在客戶資格判定、郵件撰寫、收入預測、360° 客戶視圖 |
| [**服務 Copilot**](/zh-Hant/docs/ai-copilot/service-copilot) | 服務坐席、支援經理 | 工單分流、360° 客戶視圖、郵件撰寫 |
六項技能按角色分成兩組來編寫文件。這只是文件上的便利,而非執行時邊界——同一個助手執行
全部六項技能,只要「銷售」技能是正確答案,它也會用它來回答服務問題。

兩個 Copilot 在相關時可以互相呼叫對方的技能——例如,服務 Copilot 可以使用銷售 Copilot 的 *Customer 360°* 技能來展示客戶購買過什麼。
| 分組 | 受眾 | 技能 |
| --- | --- | --- |
| [**銷售技能**](/zh-Hant/docs/ai-copilot/sales-copilot) | 銷售代表、銷售經理 | 即時資料、潛在客戶資格判定、郵件撰寫、收入預測、Customer 360° |
| [**服務技能**](/zh-Hant/docs/ai-copilot/service-copilot) | 服務坐席、支援經理 | 即時資料、工單分流、Customer 360°、郵件撰寫 |

## 從這裡開始

- [銷售 Copilot](/zh-Hant/docs/ai-copilot/sales-copilot)——銷售代表和經理能做什麼。
- [服務 Copilot](/zh-Hant/docs/ai-copilot/service-copilot)——坐席能做什麼。
- [銷售技能](/zh-Hant/docs/ai-copilot/sales-copilot)——銷售代表和經理能做什麼。
- [服務技能](/zh-Hant/docs/ai-copilot/service-copilot)——坐席能做什麼。
- [技能](/zh-Hant/docs/ai-copilot/skills)——完整的技能庫。
- [知識庫](/zh-Hant/docs/ai-copilot/knowledge-bases)——Copilot 讀取的 RAG 索引。
- [知識庫](/zh-Hant/docs/ai-copilot/knowledge-bases)——助手讀取的 RAG 索引。
Loading
Loading