@fleox-agent-crm/agentcrm

中文文档
0.12.2发布于 2026/11/07SEE LICENSE IN LICENSEES Module零依赖构建来源已验证近 30 天下载 3875

The static CRM your AI agent can run: a Twenty / HubSpot-style CRM that runs entirely in the browser or headless through the agentcrm CLI, with data on storage you own.

原始使用文档
文档目录16 节
  1. Why AgentCRM
  2. 🗂️ Nothing to run
  3. 🔐 Your contacts, encrypted, where you choose
  4. 🤖 Agent-native
  5. Quick start
  6. With an AI agent
  7. Without an agent
  8. Features
  9. Make it yours
  10. How it works
  11. How AgentCRM compares
  12. The agentcrm CLI
  13. MCP server
  14. Documentation
  15. Contributing
  16. License

AgentCRM

AgentCRM

The static CRM your AI agent can run.
Companies, people, deals, pipelines, tasks and reports in a browser admin.
Your contacts stay encrypted on storage you own. No server to run, ever.

npm version CI Node.js version Claude Code plugin MCP server License

Quick start · Features · Make it yours · How it works · CLI · MCP · Docs · Changelog

Watch the 2-minute demo: real Claude Code sessions install AgentCRM and run the daily sales work, then a tour of the admin
▶️ Watch the 2-minute demo: Claude Code installs the CRM, imports leads, moves a deal and plans the follow-ups, then a tour of the admin.


Why AgentCRM

🗂️ Nothing to run

The admin is a static web app: HTML and JavaScript you host anywhere, for free or nearly (GitHub, GitLab or Cloudflare Pages, an S3 bucket, a folder). There is no AgentCRM server, account or subscription, and nothing to patch.

🔐 Your contacts, encrypted, where you choose

Data lives in an encrypted journal (AES-256-GCM) on your S3-compatible bucket, private GitHub / GitLab repo or local folder, or in your own Firebase project. Several people can work at once without conflicts, and erasure is physical.

🤖 Agent-native

Every action of the admin is also a CLI command with JSON output, an MCP tool and a tool of the in-admin assistant, with ready-made skills for Claude Code and Hermes. Ask "move Globex to negotiation and plan the follow-ups": your agent runs the same engine, with the same roles and checks.

Quick start

With an AI agent

Claude Code

/plugin marketplace add fleox/agentcrm
/plugin install agentcrm@agentcrm

Then just ask:

Install AgentCRM for my company: data in my OVH bucket, admin on GitHub Pages.
Import the leads in ~/Downloads/leads.csv, skip duplicates, and create a follow-up task for each one.
Which deals have had no activity for two weeks? Draft a follow-up email for each.

The agent asks where the data lives and where the admin is hosted, then walks you through creating least-privilege keys. You type secrets in your own terminal, never in the chat.

A real Claude Code session: one sentence, and the agent installs AgentCRM with encrypted data and demo data on this computer, checks it with doctor and gives the admin URL; it ends on that CRM's dashboard
A real Claude Code session. The agent's work is sped up ×8 (labelled); typing and the answer play in real time. It ends on the admin opening that same encrypted data.

Once it is installed, the CRM is a conversation:

A real Claude Code session: the agent imports leads from a CSV file skipping duplicates, moves a deal to Negotiation, creates follow-up tasks for idle deals and shows the pipeline forecast, then the deals kanban in the admin
Same CRM, a second request: import, pipeline move, follow-up tasks and a forecast. Sped up ×16 while the agent works.

And a third one: "Make the CRM match our brand…"

A real Claude Code session: the agent sets a teal accent, a dark sidebar and the Inter font, and turns off the lead forms and email modules; it ends on the re-themed admin
Appearance and modules changed by the agent through the same engine as Settings. Sped up ×8.

AgentSetup
Claude Codethe plugin above: skills agentcrm-install, agentcrm-manage, agentcrm-admin, agentcrm-sales, agentcrm-extensions
Claude Desktop, Cursor, any MCP clientthe MCP server: agentcrm mcp --site <alias>
Hermes Agent / Python agentsgenerated skills with a Python wrapper, see hermes/
Codex, Windsurf, any other agentpoint it at AGENTS.md and the Markdown skills in plugin/skills/

Without an agent

npx -y @fleox-agent-crm/[email protected] serve

This starts the admin on your computer (Node.js 20+). Open the printed URL and follow the setup wizard: data → where → install. To look around first, pick This browser with demo data: nothing leaves your machine. The wizard can also put the admin on GitHub / GitLab / Cloudflare Pages or a bucket, so your team only needs a browser.

Scripted installs (install --plan plan.json), hosting options and Firebase are in docs/hosting.md and docs/cli.md.

Features

The AgentCRM dashboard: open pipeline, weighted forecast, deals won this month, new leads, overdue tasks, pipeline by stage, won / lost, lead sources and activity by user

The deals pipeline as a kanban board, with amounts and weighted totals per stage

Pipelines: drag deals between stages, weighted forecast

A person's record page: portrait, fields, company link and a timeline with a task, a meeting, a note with an image and a PDF, and an email

Record pages: fields, relations, timeline, files

A people table in a saved view filtered on lifecycle stage, with avatars, company logos and coloured values

Tables: inline edit, filters, saved views, bulk actions

The Cmd+K command palette searching across companies, deals, people and activities

Cmd+K: search everything, jump anywhere

Sell

  • Companies, people, deals and custom objects with 20 field types (currency, relation one / many, select, user, rating, address, image, files…)
  • Several pipelines with probabilities and open / won / lost stages, kanban by stage, weighted forecast
  • Notes in Markdown, calls, emails, meetings and tasks with due dates and assignees; Gmail, Google Calendar and Outlook logging from the browser (OAuth, no server)
  • Contact photos, company logos and attachments on records and activities, stored with your data and encrypted with it

Organise

  • Table views with inline edit, filters, sorts, saved views and bulk actions; each user's layout follows them across devices
  • Drilldown and rollup fields: a market's company count links to those companies, a company's status is recomputed for the markets you filtered on
  • Duplicate detection and merge, CSV import with automatic mapping and a dry run, CSV / JSON export, full ZIP backup
  • Who changed what: every record keeps who last wrote each field and when, from the admin, the CLI, imports and syncs

Automate and report

  • Automations on create, update, field change, stage change, schedule or inactivity: create a task, update a field, add a note, assign an owner (round robin), call a webhook
  • Lead forms: a static form page that posts to an inbox you own; submissions become deduplicated records
  • Reports: pipeline by stage, forecast, won / lost, activity per user, lead sources, stale deals, conversion
  • Scheduled work runs while the admin is open, or from a GitHub Actions cron in a private repo of yours

Run it as code

  • Schema as code: objects, fields, pipelines and views in one JSON file, applied with a plan, idempotent, never destructive (guide)
  • Mirror another system: records sync keeps an object in step with an external source by key, archives what left, writes only the differences (guide)

Safe by design

  • Encrypted by default: passphrase → PBKDF2-SHA256 (600 000 iterations) → AES-256-GCM
  • Secrets (storage keys, git tokens, passphrases) stay in each user's browser or in the CLI's 0600 credentials file, never in config.js, the data or git
  • Admin, member and viewer roles, enforced by the engine itself (and by generated Firestore rules in Firebase mode)
  • GDPR: soft delete, then compact erases deleted records and files physically; export everything at any time
  • The admin is available in English and French

Make it yours

Everything below is in Settings, and an agent can do all of it too (CLI, MCP tools, skills), with a plan before any change.

Appearance. One accent colour generates the whole palette. Pick the grays, radius, density, font, sidebar style, dark or light by default, your logo and favicon, custom CSS, or one of the presets and theme extensions. Appearance guide

Default preset

Default

Ocean preset

Ocean

Forest preset

Forest

Sunset preset

Sunset

Graphite preset

Graphite

Rose preset

Rose

Midnight theme extension in dark mode

Midnight (theme extension)

Settings, Appearance: preset gallery, accent colour and live preview

Settings → Appearance

Modules. Switch off what you don't use (dashboard, deals and pipelines, reports, lead forms, tasks, automations, email and calendar, AI assistant). Pages, navigation and widgets disappear; no data is deleted. Modules guide

Extensions. Modules (pages, record tabs, dashboard widgets, settings pages, AI tools, automation actions, hooks) and themes are ZIP files installed at runtime, with no rebuild. They are stored in the CRM's own data, so every admin, the CLI and the MCP server load them. Installing shows who made it and what it adds, and safe mode starts without any. Extensions guide

npx -y @fleox-agent-crm/[email protected] ext new module deal-radar   # Vite project, typings, en / fr
cd deal-radar && npm install && npm run build
npx -y @fleox-agent-crm/[email protected] ext check .                 # validates it and loads it like the admin
npx -y @fleox-agent-crm/[email protected] ext install . --site acme   # plan; add --yes to install

Terminal: scaffolding an AgentCRM extension module, building and checking it, installing it on a CRM and calling its new AI tool
The module's AI tool is live right away in the CLI, the MCP server and the in-admin assistant.

Guides: creating a module · creating a theme · runtime API · examples. With Claude Code, the agentcrm-extensions skill drives the whole loop.

How it works

You and your AI agent use four front doors (the admin, the in-admin assistant, the CLI, the MCP server). All of them call one engine (roles, validation, automations), which reads and writes your data where you choose: encrypted journal on a bucket, a git repo or a local folder, Firebase, a private git repo as JSON, or this browser. The admin is hosted on any static host. No server of ours.

All business rules live in one engine. The admin, the CLI, the MCP server and the assistant are four front doors to it, so an agent can do exactly what a user can do: no more (roles apply), no less. The data backend (where the CRM lives) and the hosting (where the admin is served) are independent choices.

Data backendWhereGood forEncryptedSeveral users
Encrypted journal (default)S3-compatible bucket (AWS, R2, OVH, Scaleway…), private GitHub / GitLab repo, local folderMost teams: personal data under your controlYesYes, conflict-free
FirebaseFirestore + Auth in your own projectReal-time teams with logins; incremental loading fits the free planAt rest by GoogleYes, roles in generated rules
Private git repo (JSON)One readable JSON file per recordReadable history, per-user tokensNo (history keeps old values)Yes
This browserIndexedDBTrying it out—No

The journal appends one encrypted segment per write, with no lock, and merges them field by field with hybrid logical clocks; compact folds them into a snapshot and erases what was deleted. The admin itself goes on GitHub Pages, GitLab Pages, Cloudflare Pages, an S3-compatible bucket, a local folder or a ZIP; its config.js never holds a secret. See architecture, hosting and Firebase.

How AgentCRM compares

AgentCRMTwentyHubSpotA spreadsheet
Server or database to run❌ none✅ Postgres + Node (or their cloud)their cloud❌ none
Where the data livesyour bucket / repo / Firebase, encryptedyour database or their cloudtheir clouda file
Costyour storage, often centshosting or seatsseats and tiersfree
Pipelines, automations, forms, reports✅✅✅manual
Fully drivable by an AI agent✅ CLI, MCP, skills, assistantAPIAPI, paid AIad hoc
Scalesmall and medium teams (whole dataset in memory)largelargesmall

The agentcrm CLI

The same engine the agents use, by hand: npx -y @fleox-agent-crm/[email protected] help. Every command accepts --json and prints exactly one { ok, data } or { ok: false, error } object, and commands that change the workspace print a plan until you add --yes.

Terminal running the agentcrm CLI: a filtered deals list, creating a person, moving a deal to Won, adding a task and a pipeline report
Run the CRM from a terminal or a script, on the same data as the admin.

Describe the data model in a file and keep an object in step with another system:

Terminal applying a JSON schema to the CRM (plan, apply, already up to date), then mirroring a list of markets with records sync: created, updated and archived records
schema apply and records sync: plan first, idempotent, nothing deleted.

All commands
AreaCommands
Setupstatus, options, credentials set|list|remove|copy|verify|types, install --plan, site list|show|add|use|rename|forget|credentials|runtime, doctor, serve, deploy-admin
Recordsrecords list|get|create|update|delete|restore|search|duplicates|merge|image|sync, files list|upload|get|delete|usage
Schemaschema apply, objects list|get|create|update|delete, fields add|update|delete
Workactivities add|list|done, notes add, tasks list|add|done|reassign, pipelines list|move
Adminviews list|create|delete, users list|add|role|disable, settings get|set
Dataimport csv|zip, export csv|json|zip, report <name>, stats, compact
Automationautomations list|enable|disable|run, forms list|publish|ingest
Customisationmodules list|get|enable|disable|config, appearance get|presets|set|preset|logo|favicon|css|reset, ext new|check|pack|install|list|enable|disable|uninstall|export
AItools list|call, mcp

Full reference: docs/cli.md.

MCP server

claude mcp add agentcrm -- npx -y @fleox-agent-crm/[email protected] mcp --site acme

Terminal adding the AgentCRM MCP server to Claude Code, the server reported as connected, and a search tool call returning JSON

Tools are the shared AI catalog: search, list, get, create, update, delete and merge records, notes, activities, tasks, deal stages, reports, CSV import, automations, modules and appearance, plus the tools of the site's enabled extension modules. Resources are the schema, pipelines, modules and appearance. Prompts cover a pipeline review, a follow-up email and lead qualification. --read-only hides every tool that writes. Claude Desktop and Cursor setup: docs/mcp.md.

The admin also has an AI assistant panel that calls the Claude API from your browser with your own key, with the same tools; tools that write ask for confirmation.

Documentation

HostingData backends, storage options, hosting the admin, protecting it
FirebaseConsole setup, generated rules, invitations, CLI sign-in, costs
Data modelObjects, fields, relations, drilldowns, rollups, pipelines
Schema as code · SyncThe data model in a file; mirroring another system
CLI · MCPEvery command; MCP clients
Automations · Import / export · FilesTriggers and scheduled runs; CSV, JSON, ZIP; images and attachments
Modules · Appearance · ExtensionsCustomising the workspace
ArchitectureThe engine, the journal, the code layout
AGENTS.mdInstructions for AI agents working on this repository

Contributing

Bug reports and ideas are welcome: open an issue. Read CONTRIBUTING.md before opening a pull request. For security issues, follow SECURITY.md instead of opening a public issue.

npm install --legacy-peer-deps
npm run dev        # admin at http://localhost:5180
npm test           # Vitest
npm run build      # admin + CLI

License

Source-available under the PolyForm Noncommercial License 1.0.0. Personal, hobbyist, educational, research and other noncommercial use is free, and you can fork, modify and redistribute under the same terms.

Commercial use requires written authorization. This includes selling the software or hosting it as a paid service. For a commercial license, open an issue.

版本与发布记录

@fleox-agent-crm/agentcrm 共发布 20 个版本,首次发布于 2026/11/04;过去 12 个月发布了 20 个版本。最新版本附带 npm provenance 构建来源证明,可追溯到公开的 CI 构建记录。

版本发布日期距今
0.12.2latest2026/11/079 天前
0.12.12026/11/069 天前
0.12.02026/11/069 天前
0.11.02026/11/0610 天前
0.10.12026/11/0510 天前
0.10.02026/11/0510 天前

关于 @fleox-agent-crm/agentcrm 的常见问题

@fleox-agent-crm/agentcrm 是什么?

The static CRM your AI agent can run: a Twenty / HubSpot-style CRM that runs entirely in the browser or headless through the agentcrm CLI, with data on storage you own.

如何安装 @fleox-agent-crm/agentcrm?

在项目目录运行 npm install @fleox-agent-crm/agentcrm,或使用 pnpm add / yarn add / bun add。它还提供命令行工具 agentcrm,可以用 npx agentcrm 直接运行。

@fleox-agent-crm/agentcrm 支持 TypeScript 吗?

该包的 package.json 没有声明类型入口。需要类型提示时,可以在 npm 上查找社区维护的 @types/fleox-agent-crm__agentcrm 是否存在。

@fleox-agent-crm/agentcrm 支持 ESM 还是 CommonJS?

以 ES Module 形式发布,推荐使用 import 引入。在 CommonJS 项目中,Node.js 20.19+ / 22.12+ 可以直接 require() 不含顶层 await 的 ES Module,更早的版本需要改用动态 import()。

@fleox-agent-crm/agentcrm 需要什么版本的 Node.js?

package.json 的 engines 字段要求 Node.js >=20。版本不符时 npm 默认只给出警告,开启 engine-strict 后会拒绝安装。

@fleox-agent-crm/agentcrm 可以商用吗?

@fleox-agent-crm/agentcrm 使用 SEE LICENSE IN LICENSE 许可。许可条款写在包内的独立文件中,请阅读该文件确认可否商用。以上为概括,不构成法律意见,请以许可原文为准。

@fleox-agent-crm/agentcrm 的下载量有多少?

根据 npm 官方下载统计,@fleox-agent-crm/agentcrm 在 2026/10/15 – 2026/11/13 这 30 天内被下载 3,875 次,日均约 129 次。下载量会把 CI 构建、镜像同步和作为其他包依赖的安装一并计入,适合横向比较热度,不能直接等同于使用人数。

@fleox-agent-crm/agentcrm 还在维护吗?

最新版本 0.12.2 发布于 2026/11/07(9 天前);至今共发布 20 个版本,其中过去 12 个月发布了 20 个;首次发布于 2026/11/04。发布频率只是参考信号:功能稳定的工具可能很少需要更新,建议结合源码仓库的 issue 和提交记录判断。