Name: BailingHub
Availability: InStock
Author: bailinghub

GitHub Website

Install

Updated10 hours ago

LanguageTypescript

Setup difficultyAdvanced

Stars: 127
Port: 18900 (BailingHub core HTTP API and console)

Docker

crpi-xm97pbcjrmf5in3s.cn-shanghai.personal.cr.aliyuncs.com/bailinghub/bailinghub

Install with Docker Compose

What is BailingHub?

BailingHub (百灵中枢) is an open-source agent-to-business (A2B) control plane that connects AI agents to existing business systems like e-commerce platforms, SaaS, CRM, and ERP. It governs what agents can do by managing trusted business identities, capability scopes, risk controls, and full audit trails — while the business systems retain final authorization. Users interact via embedded web chat, local agents (e.g., DeepSeek Harness), or automation platforms (Dify, n8n, MCP) using a shared routing, knowledge, and audit layer. The core is written in Node.js/TypeScript with a Vue/Element Plus console, and persists state in MySQL. It supports approval workflows for high-risk actions, idempotency, rate limiting, and tool governance via the Agent Capability Contract (ACC). Deployment options include Docker Compose, a one-click install script for Ubuntu/Debian, or manual Node.js 22+ with MySQL. The project is in early public validation (v0.9.x) under Apache-2.0.

Should you self-host BailingHub?

BailingHub scores 44/100 on selfhost.directory (how scoring works) and ranks #146 of 209 in AI Agents. It has 127 GitHub stars.

It shipped 9 stable releases in the past year. The latest, v0.10.0, came out on 22 October 2026. New versions land roughly every 4 days.

Setup is advanced: plan for several services and some ongoing maintenance. An official Docker image is available. It is Apache-2.0-licensed and written mostly in Typescript.

Its closest alternative on selfhost.directory is Orca (91/100, 108 446 stars). It scores higher overall, so compare the two before you pick. See BailingHub vs Orca side by side.

Features

Agent Client SDK for local agents
Agent-to-Business (A2B) control plane
Approval workflows for high-risk actions
Capability governance with ACC
Execution governance (allowlist, risk levels, rate limiting, confirmations, approvals, idempotency)
Full audit trail and traceability
Knowledge base with RAG
MCP server adapterSeparate repository: bailinghub/bailinghub-mcp-server
Multi-entry routing (web chat, local agents, API/automation)
Multi-modal input support
Notification durabilityAdded in v0.9.4
OpenAPI/ACC tool declaration with SDKs
Pause/resume control
Self-hosted console with RBAC
Trusted business identity binding

How to configure BailingHub

BAILING_ALERTS_TYPEAlert type (webhook or notify executor target)
Alert type (webhook or notify executor target)
values: webhook, <executor target name>
BAILING_ALERTS_URLWebhook URL for alerts when type=webhook
Webhook URL for alerts when type=webhook
BAILING_AUDIT_RETENTION_DAYS0Audit log retention in days (0 = no auto-delete)
Audit log retention in days (0 = no auto-delete)
BAILING_BOOTSTRAP_ADMIN_PASSWORDInitial admin password (only used if admin table is empty)
Initial admin password (only used if admin table is empty)
BAILING_BOOTSTRAP_ADMIN_USERNAMEInitial admin username (only used if admin table is empty)
Initial admin username (only used if admin table is empty)
BAILING_DISPLAY_TZAsia/ShanghaiDisplay timezone
Display timezone
BAILING_ENVEnvironment mode (production/development)
Environment mode (production/development)
values: production, development
BAILING_HOST127.0.0.1Bind address for the server
Bind address for the server
BAILING_LLM_CREDENTIALS_JSONLLM credentials as JSON object (e.g., OpenAI, DashScope, OpenRouter)
LLM credentials as JSON object (e.g., OpenAI, DashScope, OpenRouter)
BAILING_METRICS_ENABLEDfalseEnable OpenMetrics endpoint
Enable OpenMetrics endpoint
values: true, false
BAILING_METRICS_TOKENIndependent random token for metrics scraping
Independent random token for metrics scraping
BAILING_MYSQL_DATABASEbailinghubMySQL database name
MySQL database name
BAILING_MYSQL_HOSTMySQL host
MySQL host
BAILING_MYSQL_PASSWORDMySQL password
MySQL password
BAILING_MYSQL_PORT3306MySQL port
MySQL port
BAILING_MYSQL_USERMySQL username
MySQL username
BAILING_PORT18900Port the server listens on
Port the server listens on
BAILING_STATE_BACKENDmysqlState backend type
State backend type
values: mysql, jsonl
BAILING_TOKENRoot secret for machine-admin access and derived HMAC credentials (min 24 chars)
Root secret for machine-admin access and derived HMAC credentials (min 24 chars)
DEMO_BUSINESS_URLhttp://demo-business:19080Demo business system URL
Demo business system URL
DEMO_CLIENT_TOKENbailing-demo-client-tokenDemo client token for API access
Demo client token for API access
DEMO_PROFILEfull-localDemo profile mode
Demo profile mode
values: full-local, stateless-readonly
DEMO_TOOL_SECRETdemo-tool-secret-change-meSecret for demo business tool signing
Secret for demo business tool signing

FAQ

Can local agents use the hub's model packages?

Yes. After the host integrates model services, it can use the package's chat models and image generation tools with shared quota, or continue using its own configured models. Conversation and tool orchestration remain with the local agent.

Must I self-host a large language model locally?

Not required. The control plane and state store can be self-hosted; models can be cloud APIs or self-hosted based on cost, network, and data requirements. When calling external models, relevant inputs are sent per your configuration.

How does it differ from MCP, Dify, or n8n?

MCP solves tool protocols; Dify and n8n orchestrate applications or workflows. BailingHub focuses on trusted identity, capability scope, approval, signing, execution, and audit between agents and business systems. They can be combined.

Do all write operations require manual approval?

No. Whether approval is needed is determined by capability declarations and business rules; ordinary authorized operations can execute directly, while high-risk operations can be configured for approval-before-execution.

After integration, who decides the final permissions?

Your business system still decides. BailingHub controls which capabilities agents can discover and request; the business backend continues to execute or refuse based on current user, tenant, roles, and business rules.

Is BailingHub an RPA that simulates clicking web pages?

No. It does not rely on models arbitrarily clicking backend pages. Instead, it calls server-side capabilities explicitly declared and authorized by the business system, so parameters, identity, permissions, and results can be verified.

Do I need to rebuild my existing business backend?

No. You keep your existing backend, accounts, permission tables, and business processes. You declare selected actions via OpenAPI, ACC, or SDK and configure capability declarations, trusted identity, and call verification.