FollowAgents AI AGENTS DIRECTORY
Home / All Agents / Enterprise AI Agent Platform
Dev & Engineering langgraphfastapivue3mcphuman-in-the-looplong-term-memorysandbox-workspace

Enterprise AI Agent Platform

A self-hostable foundation for teams building governed, extensible Agent platforms with a web console and separate runtime.

View on GitHub ↗ Quick start ↓
Source repo
ljxpython/ai-agent-platform
Stars
★ 152
Last updated
today
Primary language
Python
FA score
57/100 · Major gaps

At a glance

How it runs
Self-hosted serviceWeb app
Works with
Portable with changes
Setup effort
High · needs real infrastructure
You'll need
Python 3.11+Node.js 18+pnpm 9+PostgreSQL 14+RedisDockerShell / CLINetwork accessLocal filesystemMCP Server
Typical use
An enterprise platform team needs a private Agent control plane with user and project isolation plus audit logging.
Not a fit if
  • Individuals who only need a local chat UI, not a self-hosted platform
  • Teams seeking a plug-in they can install directly in Codex or Claude Code
Source review
57/100 · Major gaps

What does this agent do, and when should you use it?

This repository provides an AI Agent platform foundation for customization and private enterprise deployment, split into Platform API, Platform Web, and Runtime Service. The FastAPI, SQLAlchemy, and PostgreSQL control plane handles authentication, project isolation, audit logs, model catalog management, and forwarding of managed contracts; the console is built with Vue 3. The Runtime Service uses Python 3.11+, LangGraph, GraphHarbor, and Redis for graph orchestration, tool and MCP assembly, session scheduling, and SSE event delivery. Included examples are the introductory showcase_demo and a production-oriented DeerFlow Agent, with documented examples of human approval, sandbox workspaces, long-term memory, and subagent collaboration. It can be deployed with local multi-process scripts or Docker Compose, and is best suited to teams prepared to adapt a platform and Agent implementation rather than users seeking a ready-made chatbot.

A user starts a request in the Vue 3 workbench. Platform API reads the project's enabled model parameters, tool allowlist, and system prompt, assembles a managed contract, and forwards it to Runtime. A Runtime Worker drives a LangGraph state machine to call an LLM and, when requested, execute Python Tools or MCP tools; sensitive actions can enter a HITL approval flow. Execution events stream to the frontend over SSE for live chat and trace inspection. The DeerFlow Agent example also schedules Research, Coding, and Planning modes, manages short-term conversation context and cross-session personal or project memories, and can produce files in a sandbox workspace for preview and archive download.

  1. An enterprise platform team needs a private Agent control plane with user and project isolation plus audit logging.
  2. A LangGraph engineering team wants a starter platform with a runtime, tool registration, and session scheduling to customize.
  3. A team needs standard MCP-connected tools and human approval for sensitive tool calls.
  4. A coding or research team wants to evaluate a complex Agent template with a sandbox file workspace, interactive terminal, and archived outputs.
  5. An operations team needs to deploy the API, runtime, and console through Docker Compose or local multi-process scripts.

How do you install or deploy this agent?

The documented local environment requires Python 3.11+, Node.js 18+, pnpm 9+, PostgreSQL (14+ recommended), and Redis. The README does not provide a complete dependency-installation and first-time configuration sequence; configuration initialization, database creation, and password setup are covered by the deployment guide. Once the environment is configured, check and start the local stack with:

source "apps/runtime-service/.venv/bin/activate"
bash "scripts/local-stack.sh" doctor
bash "scripts/local-stack.sh" start

A full-stack Docker Compose option is also documented:

docker compose -f deploy/docker-compose.stack.yml --env-file deploy/.env.stack up -d

How do you use this agent?

After startup, open Platform Web and check the Platform API and Runtime Service. The documented default endpoints are:

Platform Web:     http://127.0.0.1:3000
Platform API:     http://127.0.0.1:2142
Runtime Service:  http://127.0.0.1:8123

Run the health checks with:

curl -fsS "http://127.0.0.1:2142/_system/health"
curl -fsS "http://127.0.0.1:8123/ready"

In the console, select or create a project and configure a model before chatting with an included Agent. Add custom Agent graphs under apps/runtime-service/src/runtime_service/graphs/, Python tools under apps/runtime-service/src/runtime_service/tools/, and configure external MCP services in Runtime. The README does not specify model credential names or show an API invocation example.

What are this agent's strengths and limitations?

Pros
  • Separates platform governance from Agent execution; the control plane includes authentication, project isolation, auditing, and a model catalog.
  • Built on LangGraph with explicit extension points for custom graphs, Python tools, MCP services, and management APIs.
  • Includes complex Agent examples covering HITL, sandbox workspaces, subagents, and memory workflows.
  • Documents both local multi-process and Docker Compose deployment for a web console, API, and separate runtime.
Limitations
  • Deployment involves several services and infrastructure components, including PostgreSQL, Redis, Python, and Node.js.
  • The project targets customization; the README does not give a complete clone-to-installed dependency and configuration command sequence.
  • The documented material does not identify specific model providers, credential settings, usage costs, or a directly verifiable model API call.
  • Execution is centered on LangGraph, so moving graph, tool, and session behavior to another Agent runtime may require adaptation.

How does this agent compare with similar options?

The README positions the project as a LangGraph/LangChain-based platform foundation and says its design draws on open-swe, deepagents, and deer-flow. These are engineering references and sources of ideas, not documented compatibility claims or drop-in deployment alternatives.

Key facts side by side with the most closely related agents.

Agent Source review Form / cost Stars Updated Language Full support on
Enterprise AI Agent Platform This agent 57 · Major gaps Self-hosted service ★ 152 today Python —
Deuz SDK 76 · Good Library / SDKFree + model costs ★ 508 22d ago TypeScript OpenAI API · Claude API
Email Agents From Scratch 46 · Major gaps Library / SDKFree + model costs ★ 2.4k 2mo ago Jupyter Notebook OpenAI API
Open Swarm Insufficient evidence — ★ 826 1mo ago JavaScript Claude API

How does FollowAgents rate this agent?

FollowAgents source review · FARS-2.1
Major gaps
57/ 100 5-point scale 2.9 / 5
Trust 17/29
Reliability 6/14
Adaptability 12/18
Convention 10/18
Effectiveness 7/13
Verifiability 5/8
Why each dimension lost points
Trust17 / 29 · 2.9/5

The runbook specifies a non-root sandbox, no network, resource limits, credential-file permissions, mTLS, precise cleanup, and rollback; cleanup requires human verification and a candidate token. These support confirmation and recovery. Deductions: the README claims broad capabilities, but this sample lacks the complete authorization implementation or data-flow code; local mode explicitly uses a trusted host shell and the remote node account has Docker-group access, so least privilege is not fully established. Dependency security is described through frozen, hashed installation, but no vulnerability review evidence is shown.

Reliability6 / 14 · 2.1/5

The runbook describes reconciliation by original ID, avoiding resubmission of unknown commands, denying admission on failures, and excluding disk loss from recovery commitments. CI lists some lint, compilation, test, and build steps. Deductions: the README's end-to-end contract and quality claims cannot be independently confirmed from this sample. pyproject.toml instead has empty dependencies and a placeholder description, with version 0.1.2 conflicting with the README's v0.5.0 release claim; the sample provides no dependency lock or availability guarantee.

Adaptability12 / 18 · 3.3/5

The README targets teams doing secondary development and lists service responsibilities, built-in agents, and extension points. The academic-paper skill states its trigger and output scope, while the runbook distinguishes local, local-Docker, and remote-Docker modes. Deductions: the sample does not fully cover product boundaries, resource requirements, or deployment audiences, and it lacks a complete product-level capability inventory.

Convention10 / 18 · 2.8/5

The README provides a system overview, architecture-diagram links, a secondary-development guide, quick-start commands, and changelog navigation. AGENTS.md organizes documentation, and the runbook covers configuration, lifecycle, rollback, and failure boundaries. Deductions: license metadata is unknown, and the attribution text supplied is not a license; the root AGENTS.md is explicitly an AI development and contribution workflow, not a user-facing product behavior contract; package metadata still says aitestlab with a placeholder description; no clear maintenance owner or support channel is identified.

Effectiveness7 / 13 · 2.7/5

The platform combines a control plane, runtime, frontend, MCP, skills, memory, and sandbox as a secondary-development foundation; if implemented as described, this could reduce integration setup. The README includes quick-start commands. Deductions: this sample cannot verify the breadth or completeness of those capabilities, the setup instructions assume an existing virtual environment without explaining its creation here, and the deployment cost of maintaining multiple services and a remote sandbox is not quantified.

Verifiability5 / 8 · 3.1/5

The README's architecture and feature claims can be partly compared with the runbook, CI configuration, frontend entry point, and test fixtures. The runbook points to formal acceptance records and identifies some items as deferred. Deductions: several quality and release claims appear only in the README, with no supporting results in this sample; fixture code describes isolated scenarios but does not prove they ran; there is no complete evidence map linking each product claim to implementation or verification material.

Risks and how to mitigate them
  • Before deployment, verify the actual license, dependency security status, and consistency between the README release version and package metadata.
  • Include trusted-host local execution and Docker-group access in the threat model; the remote sandbox's single-node persistent disk does not cover recovery from disk damage or total host loss.
  • This assessment uses only the supplied file sample; README claims about tests, releases, and capabilities were not executed or independently verified.
Evidence confidence: Low Reviewed Nov 07, 2026 Reviewed revision 9fca1f9aa43b New commits since this review; the score may not cover them
Review evidence README.en.mdAGENTS.mdimplement-feature/SKILL.mdplan-project/SKILL.mdverify-change/SKILL.mdacademic-paper-review/SKILL.mdrunbooks/runtime-sandbox-node.mdsrc/main.tspyproject.tomlworkflows/ci.ymlworkflows/release.ymlfixtures/error_contract_server.pyfixtures/governance_server.pyfixtures/runtime_delegation_verifier.py
See the full review method →
Maintain this project? Add the FARS badge to your README

The badge shows the current review score and verdict and links back to this page. It refreshes about once an hour after a re-review.

FARS

[![FARS review](https://followagents.com/badge/ai-agent-platform.svg)](https://followagents.com/en/agents/ai-agent-platform)
<a href="https://followagents.com/en/agents/ai-agent-platform"><img src="https://followagents.com/badge/ai-agent-platform.svg" alt="FARS review"></a>

Recent releases

Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.

  1. v0.5.0 v0.5.0 - 企业级 AI Agent 平台架构重构与生产级智能体里程碑
    Release notes excerpt
    v0.5.0 - 企业级 AI Agent 平台架构重构与生产级智能体里程碑 > 正式发布说明，将自动同步为 GitHub Release Notes。 Summary v0.5.0 是平台的重大里程碑版本：项目全面确立“面向二次开发与企业落地的 AI Agent 平台底座”定位，解耦平台治理层（FastAPI + PostgreSQL）与 Agent 执行层（LangGraph + GraphHarbor + Redis）；引入 3 套 Archify 2K 架构/时序/扩展点全景可视化系统；实装以 open-swe、deepagents 与 deer-flow 为核心工程支柱的生产级智能体 DeerFlow Agent，支持多模式工作流、长期记忆闭环引擎、安全沙箱 Workspace 与真实 PTY 交互终端。 Highlights - 企业级分层解耦架构：平台治理层专职鉴权、多租户、审计与 Catalog 治理；Agent 运行时专职图编排与状态流转，两层通过受管契约通信，二开零侵入。 - Archify 2K 架构与时序全景：发布系统总览、Agent 请求执行链路、二次开发扩展点 3 套交互式可视化系统与 100% 完整长图。 …
    View on GitHub ↗
  2. v0.4.1
    Release notes excerpt
    Full Changelog: https://github.com/ljxpython/ai-agent-platform/compare/v0.4.0...v0.4.1
    View on GitHub ↗
  3. v0.4.0 v0.4.0 - 可信运行时与聊天 v2 工作台
    Release notes excerpt
    v0.4.0 - 可信运行时与聊天 v2 工作台 > 正式发布说明，可直接用于 GitHub Release。 Summary v0.4.0 将平台聊天主线迁移到受管的 v2 运行时契约：服务端生成可信身份和项目上下文，前端以流式 SDK 状态作为唯一运行真相，同时提供独立调试工作台。 Highlights - 正式聊天使用 v2 流式运行时，覆盖续接、中断、工具调用和历史消息展示 - 模型、工具和 prompt 配置通过受管运行时契约进入每次运行 - 身份与项目上下文由服务端生成，不再依赖前端注入可信字段 - 调试能力独立为工作台，避免污染正式聊天交互 Added - 聊天 v2 状态适配、运行元数据展示和独立 Debug 工作台 - 运行时委派、平台认证和运行上下文解析的服务端实现与回归测试 - 运行级模型、工具和 prompt 配置入口 Changed - 正式聊天收敛到 LangChain 流式 SDK 的运行状态 - 前端聊天依赖升级并收敛消息、工具调用和中断交互 - 运行时请求使用可信服务端上下文，部署配置同步更新 Fixed - 修复 runtime_delegation_not_configured 导致聊天无法运行的问题 - 修复发送消息后的页面闪烁、线程目标恢复和线程搜索兼容性问题 …
    View on GitHub ↗

FAQ

What services are needed to deploy it?
The local development environment lists Python 3.11+, Node.js 18+, pnpm 9+, PostgreSQL (14+ recommended), and Redis. Docker Compose files are also provided.
Which model providers can it use?
The README describes a model catalog and multi-provider management, but names no specific providers or credential variables. Confirm those in the project configuration and implementation.
Can an Agent require human approval before a tool runs?
The included examples document HITL approval interruptions and approval UI for sensitive actions. Enabling it for a custom tool depends on the Agent graph and platform configuration.
Is this available as a hosted service?
The documented deployment paths are local and Docker Compose self-hosting; the README does not describe a vendor-hosted SaaS offering.
Was this page helpful?
View on GitHub ↗ Install ↓

Compare agents like this one

The same FARS review applied across the shortlist this agent qualifies for.

MCP Agents, Compared

MCP agents ranked by FARS review — agents that ship or consume Model Context Protocol servers, with license, maintenance and trust scores side by side.