BotCom Workbench

BotCom Workbench is a local-first desktop operating cockpit for one-person companies.

It is designed for a founder/operator who uses AI workers for research, content, automation, delivery support, and review preparation, while keeping the human in the final approval loop.

It combines:

• a local file workbench;
• an embedded terminal for coding agents and command-line operators;
• a BotCom OS dashboard for one-person company operations;
• optional Feishu / WeCom status notifications;
• a safe localhost proxy for approve / reject / publish-prep actions.

The first packaged module is a content and acquisition operations layer for X, Douyin, Xiaohongshu, and other review-driven channels. The architecture is intentionally modular so operators can add CRM, delivery, revenue, asset-library, and automation adapters later.

What it is for

BotCom Workbench is designed for a workflow where AI workers prepare work and the human operator only performs final review:

strategy / research / content / delivery preparation
  -> local operating queues
  -> human approve / reject
  -> publish-prep / delivery / follow-up worker
  -> metrics and evidence collection
  -> growth and operating review

The repository contains the desktop cockpit and integration points. It does not include any private account data, platform credentials, content calendar, or generated media assets.

Core operating modules:

1. positioning and offers;
2. acquisition and growth;
3. content and distribution;
4. customer and community follow-up;
5. delivery and project execution;
6. revenue and monetization;
7. asset and knowledge management;
8. automation and review loops.

Features

• Desktop app built with Electron.
• Local-only backend bound to 127.0.0.1.
• File browser, previewer, editor, and embedded terminal.
• BotCom OS dashboard in the sidebar:
  • operating-state snapshot;
  • eight-module one-person company map;
  • local operating adapters for CRM, delivery, revenue, assets, and automation;
  • approval queues and mobile approval links;
  • service health, blockers, next actions, and review suggestions.
• AI beginner guide explaining agents, skills, models, API keys, context, and approval queues.
• Model / API profile panel for Claude Code, Codex, Anthropic, OpenAI-compatible, and DeepSeek-style setups.
• First-run setup wizard that creates a local one-person company workspace and example operating adapters without overwriting existing files.
• Review queue API integration:
  • GET /api/botcom/status
  • GET /api/botcom/review-state
  • POST /api/botcom/review-action
• Safe action allowlists for day, platform, action, and publisher.
• Feishu / Lark custom bot and WeCom webhook status bridge.
• Offline vendored frontend assets.
• macOS DMG build target.

Quick start

Requirements:

• macOS or a desktop OS supported by Electron;
• Node.js 18+;
• Python 3 for the optional notification bridge.

Install dependencies:

npm ci
npm run rebuild

Run the local web app:

npm start

Run the Electron app:

npm run app

Build a macOS DMG:

npm run dist

Run smoke checks:

npm run botcom:smoke

Run the full release gate:

npm run release:check

Publish to GitHub after authenticating gh:

gh auth login -h github.com -p https -w
npm run release:github

Configuration

BotCom Workbench is local-first. By default it uses:

BOTCOM_HOME=~/BotCom
BOTCOM_MEDIA_OPS_ROOT=~/BotCom/media-ops
BOTCOM_WORKBENCH_ROOT=~/BotCom/AI-Workbench
BOTCOM_ADAPTERS_ROOT=~/BotCom/adapters
BOTCOM_WORKBENCH_PORT=4570

You can override them:

BOTCOM_HOME="$HOME/BotCom" \
BOTCOM_MEDIA_OPS_ROOT="$HOME/BotCom/media-ops" \
BOTCOM_WORKBENCH_ROOT="$HOME/BotCom/AI-Workbench" \
npm run app

See .env.example and docs/CONFIGURATION.md.

First-run setup

Open the desktop app and click One-click setup. It creates:

• ~/BotCom/AI-Workbench with beginner-friendly strategy, content, customer, delivery, revenue, asset, and automation starter files;
• ~/BotCom/adapters with example CRM, delivery, and revenue adapters;
• no cloud account and no credential files.

Existing files are kept unchanged.

AI tools and model profiles

The in-app Model / API panel detects local claude and codex commands, stores local model profiles, and can load a selected profile into the embedded terminal.

Profiles are stored under ~/.botcom-workbench and raw API keys are not returned by the dashboard API.

Claude Code's official path uses Anthropic credentials. DeepSeek and other OpenAI-compatible providers can be configured as terminal profiles for tools or proxies that support OPENAI_API_KEY and OPENAI_BASE_URL.

Optional media-ops integration

The BotCom OS panel can integrate with a local media-ops workspace if it exposes a token-protected review server compatible with:

GET  /api/state
POST /api/approve
POST /api/reject
POST /api/publish
POST /api/retry

BotCom Workbench only proxies to 127.0.0.1; it does not accept arbitrary backend URLs from the browser.

See docs/INTEGRATION.md.

For the broader product model, see docs/ONE_PERSON_COMPANY_OS.md.

Notifications

Configure Feishu / Lark or WeCom webhooks in .env:

FEISHU_BOT_WEBHOOK=
FEISHU_BOT_SECRET=
WECHAT_WORK_BOT_WEBHOOK=

Preview without sending:

npm run botcom:status

Send:

npm run botcom:notify

Dry-run output redacts review tokens. Real notifications keep the review URL intact so an approved operator can open it from mobile.

Security model

• Backend binds to 127.0.0.1.
• Host header is restricted to localhost addresses.
• POST requests are protected by origin checks.
• Review actions are allowlisted.
• Review tokens are read locally and not exposed through the dashboard API.
• Webhook secrets are read from local environment files and not printed.
• Build artifacts, runtime files, .env, node_modules, and dist are ignored.
• GitHub publishing uses gh OAuth or GH_TOKEN; passwords are not supported and should not be used.

See docs/SECURITY.md.

Documentation

• docs/ARCHITECTURE.md
• docs/CONFIGURATION.md
• docs/INTEGRATION.md
• docs/ONE_PERSON_COMPANY_OS.md
• docs/PUBLISHING.md
• docs/SECURITY.md
• docs/ROADMAP.md
• CONTRIBUTING.md

Public repository hygiene

This repository intentionally excludes:

• personal workspace paths;
• private project files;
• platform credentials;
• generated review tokens;
• account-specific publishing queues;
• generated media outputs;
• packaged DMG files.

Attribution

BotCom Workbench is derived from FanBox by alchaincyf, licensed under MIT.

See NOTICE.md and LICENSE.