MELAYA EXCOM · 你的 AI 高管团队08 / 08

An AI executive team for solo founders.They prepare. You approve.

A CEO, CFO, CMO and up to eleven more roles read your mail, books, docs and dashboards, brief you every morning and draft the next move. They run on your own model, Claude Code included. Every send or change waits for your approval or a rule you accepted, and spending always asks.限量预览。我们正在评估容量,在扩大开放前仍限制访问。

What is an AI executive team?

Picture a small board of department heads who check the business every day and never act alone. Each one owns a part of your company and tells you what needs a decision.

An executive is not a chatbot you have to prompt. It has a role, goals, the accounts you let it read, its own memory and a daily review. It works on a schedule, reacts to events from your apps and asks its colleagues for help.

A title is a role, not a legal authority. An AI CFO cannot sign anything, and an AI CHRO never hires or dismisses anyone.

Fourteen roles. Start with three.
  1. CEODirection

    Turns the team's findings into the three decisions you must take this week.

  2. CFOFinance

    Explains cash, runway, collections and cost variances from your books.

  3. CMOMarketing

    Relates channel spend to qualified pipeline and keeps your positioning distinct.

  4. CTOEngineering

    Tracks delivery and reliability, diagnoses failed runs and proposes bounded repairs.

  5. COOOperations

    Finds stalled handoffs and duplicated work, then proposes the workflow that removes them.

  6. CROSales

    Moves qualified opportunities forward and keeps the pipeline honest.

Also on call: CPO, CXO, CHRO, CISO, CLO, CDO, CSCO and CQO. A role that sleeps costs nothing.

How does information flow from your tools to a decision?

Every piece of work follows the same path, left to right. Reading is free to happen. Anything that leaves stops at the gate.

  1. 01Read only

    Your sources

    Obsidian vault, Notion, Google Drive, Confluence and your website, plus mail, books and CRM through your own connections.

  2. 02Read only

    The executives

    Each one reads with its charter, its own memory and the team's shared knowledge, and cites every passage it uses.

  3. 03Read only

    Reads and proposals

    A daily brief, tasks with evidence, drafts and questions to colleagues. None of this needs a card.

    BriefTasksDraftsHandoffs
  4. 04Waits for you

    The approval gate

    Anything that sends, changes or spends stops here as a card in the app, by email, on Telegram or on WhatsApp.

    AppEmailTelegramWhatsApp
  5. 05Runs

    The action

    The exact call you approved runs, after grants, budget and the account are checked again.

    Exact callRe-checked
  6. 06Runs

    Audit and learning

    Every call leaves a receipt. Approve the same thing three times and the executive proposes a precise rule.

    ReceiptLessonRule
Lessons and rules feed the next turn

What does an AI CFO or an AI CMO actually do?

Two routines straight from the role templates. The names and numbers below are fictional.

Monday, 09:00 · weekly review

The CFO reconciles Stripe with the books

  1. Reads payouts, refunds and disputes from your Stripe account
  2. Matches them against the books and your plan in Google Sheets
  3. Flags every unmatched charge as a task with evidence
  4. Drafts reminders for overdue invoices, then asks you
2 unmatched charges flagged · 4 reminders drafted
Thursday · weekly campaign

The CMO drafts the newsletter, then waits

  1. Compares the last campaign's opens, clicks and unsubscribes
  2. Asks the CFO for spend by channel, aggregates only
  3. Drafts this week's newsletter for your review
  4. Holds the send until you approve it
Draft ready · send waiting for you

What can AI executives do on each Melaya plan?

Every plan runs the whole team end to end. The plan sets how far the executives may go. Inside that ceiling you still choose Observe, Prepare or Run within rules.

EXCOM is in limited preview. These ceilings apply as access opens. Your plan's pipeline limits stay as they are. Compare plans and prices

What can AI executives do on each Melaya plan?
PlansSandbox and OutpostForge Bastion and Citadel
Read assigned sources and accountsYesYesYes
Daily brief, tasks with evidence, questions to colleaguesYesYesYes
Draft messages and documentsNoInside MelayaYes
Send, post or change a recordNoNoYour approval or a rule you accepted
Spend, delete, deploy, legal and people mattersNoNoAlways asks you
Run scheduled pipelinesProposed as a taskProposed as a taskAfter a trial and two approvals
Change its own permissionsNeverNeverNever
True on every plan
  • Your own model, never one Melaya pays for
  • A read-only first pass after setup
  • Approval on anything that leaves
  • Every tool call checked against grants and budget
  • A receipt for every action in the audit trail

Can AI executives run on Claude Code or a local model?

Yes. EXCOM is the operating layer, not a model subscription. Each executive uses the model you pick: a cloud key, a router, or a CLI agent on your own machine through the Melaya runner.

With Claude Code, Codex or GitHub Copilot behind the runner, your executives think on the subscription you already pay for. The Melaya server still decides every tool call, so a runner model meets exactly the same approval gate as a cloud one. If the runner cannot get an answer, it runs nothing.

  • A model per executive, or one for the whole team
  • Ollama and LM Studio for fully local models
  • Setup never calls a model, so onboarding is free
  • Daily caps in money or tokens, per team and per executive
  • Claude Code
  • Codex
  • GitHub CopilotGitHub Copilot
  • OllamaOllama
  • LM StudioLM Studio
  • OpenRouterOpenRouter

Keep the runner up to date. An older runner keeps its executives in Observe.

Is it safe to give AI agents access to your company accounts?

Safety here is not a prompt. It is a check that runs on every tool call, before anything happens.

  1. 01

    Read-only to start

    Setup grants read access only, and every executive's first pass is read-only, whatever autonomy you chose.

  2. 02

    One account, one level

    Each account is read, draft or write, per executive. A read account never writes, even in Run within rules.

  3. 03

    Money always asks

    Payments, deletions, deploys, legal and people matters ask you every time. No rule can cover them.

  4. 04

    It cannot change itself

    Executives have no tool to widen their own grants, budget, rules, schedule or approvers.

  5. 05

    Private stays private

    Each executive reads only what it was granted. A colleague answers with its own access and never lends it.

Audit receiptFictional example
  1. CFO
    Read payoutsstripe · billing account
  2. CFO
    Proposed a task: 2 unmatched chargesevidence attached
  3. CFO
    Asked to send 4 reminderscard on Telegram
  4. You
    ApprovedTelegram
  5. CFO
    Sent 4 remindersgmail · billing account
  6. CQO
    Verified the taskresult read back

Checked before every call: pause, grants, account, caps, rules, autonomy.

先提供背景。再设权限。然后开始工作。

Six short steps in one window, about two minutes. The wizard saves after every step, so you can leave and come back.

  1. 1

    描述业务

    设定目标,从 Obsidian、Drive、Notion、Confluence 或你提供的文档中选择知识。

  2. 2

    分配正确账户

    财务使用账单邮箱,销售使用销售账户。为每位高管选择可访问的工具和资源。

  3. 3

    选择唤醒方式

    安排定期审查,或通过同事请求及已配置的应用事件唤醒角色。正在执行时,新请求会排队。

一支让你了解工作方式的团队。

EXCOM 提供什么、你需要准备什么,以及哪些事项由你掌控。

EXCOM 与 AI 聊天有什么区别?

聊天通常等待你的下一条提示。EXCOM 为每个角色提供连续对话、指定来源、工具,以及定时或事件驱动的工作。高管可以互相求助,并将行动建议汇集在同一处。

可以使用 Claude Code 或本地开放权重模型吗?

可以,需要已连接且兼容的 Melaya runner 以及受支持的模型配置。你提供服务商账户或本地算力。Runner 必须保持可用,相关高管才能工作。

AI 财务总监会使用销售的邮箱吗?

只有在你明确分配时才会。每位高管都有具体的账户和资源权限。如果连接器支持,可以分配多个账户。

EXCOM 会取代可视化 Agent Builder 吗?

不会。高管协调持续职责,pipeline 执行可重复工作流。托管 pipeline 仍遵守现有平台限额和审批控制。

现在所有用户都能使用 EXCOM 吗?

EXCOM 目前仅限预览。我们正准备更广泛的访问:Sandbox 观察、Forge 内部草稿、Bastion 受控自主行动。目前购买套餐不会解除预览访问限制。

发展你的企业。保留最终决定权。

了解持续工作的 AI 高管如何使用你的模型、账户和企业知识。

限量预览 · 你的模型 · 你的权限
// Cookie
Melaya 使用一小组第一方 cookie,仅用于身份验证、维持会话和保护平台。默认不使用广告 cookie、跨站追踪器或第三方分析。完整 cookie 清单见我们的 隐私政策.