Some data can’t leave the building. Healthcare organizations processing patient records, financial institutions handling transaction data, government agencies working with classified information — they all face the same constraint. The AI automation is compelling, but the data residency requirements are non-negotiable.
JieGou’s hybrid deployment model solves this. Run execution agents inside your own VPC. Keep sensitive data in your infrastructure. Let JieGou manage the control plane — the UI, workflow orchestration, scheduling, and monitoring — in the cloud.
How it works
The architecture splits JieGou into two layers:
Control plane (JieGou cloud) — The web console, workflow editor, scheduling engine, monitoring dashboards, and user management. This is where you design, configure, and monitor your automations.
Execution agents (your VPC) — Lightweight agents that run recipe and workflow steps. They receive step execution requests from the control plane, execute them locally using your infrastructure and your API keys, and return results. The agents run as Docker containers or Kubernetes pods in your environment.
When a workflow runs, the control plane orchestrates the steps. For each step, it selects the best available agent based on capabilities, load, and scope, then dispatches the execution request. The agent processes the step — calling LLM providers with your API keys, accessing your databases, reading your files — and returns the output.
The control plane never sees the raw data. It receives execution metadata (duration, token count, success/failure) and whatever output fields the data residency policy allows.
Registering an agent
Deploy the JieGou execution agent container in your VPC. On first startup, it registers with the control plane and receives a bearer token (returned once in plaintext, stored as a SHA-256 hash). This token authenticates all subsequent communication.
Each agent declares its capabilities:
- Recipe execution
- LLM provider access (using your API keys)
- Browser automation
- MCP tool access
- Knowledge base access
- A2A outbound invocations
- Maximum concurrency
- Maximum step timeout
You can run multiple agents with different capability profiles. A “general purpose” agent might handle recipe execution and LLM calls, while a “browser” agent handles browser automation in an environment with GUI access. The control plane routes steps to the right agent automatically.
Health monitoring
Agents send heartbeats every 30 seconds (configurable). Each heartbeat reports:
- Status: healthy, degraded, or busy
- Resource utilization: CPU, memory, disk
- Active execution count
- Uptime
- Agent version
The control plane tracks agent status based on heartbeat timing:
- Online — Heartbeat received within expected interval
- Unhealthy — 3 consecutive missed heartbeats
- Offline — 10 consecutive missed heartbeats
Unhealthy and offline agents stop receiving new step dispatches. When they recover and resume heartbeats, they’re automatically brought back into the pool.
Data residency controls
Hybrid deployment isn’t just about where code runs — it’s about where data lives. JieGou’s data residency configuration lets you set per-category rules:
| Data Category | Options |
|---|---|
| Execution results | Local only, cloud sync, cloud redacted |
| Execution metadata | Local only, cloud sync |
| Audit logs | Local only, cloud sync |
| Knowledge base documents | Local only |
| LLM conversations | Local only |
| Credentials | Local only |
| Workflow definitions | Cloud only (needed for orchestration) |
| User data | Cloud only (needed for auth) |
“Local only” means the data never leaves your VPC. The control plane receives a success/failure status and timing metadata, but not the actual content.
“Cloud redacted” sends results to the control plane with PII automatically removed. JieGou can detect and redact emails, phone numbers, SSNs, credit card numbers, medical record numbers, and other sensitive patterns. You can add custom regex patterns for industry-specific data types.
“Cloud sync” sends full results to the control plane for features that need them (dashboards, version history, collaboration).
Compliance frameworks
When you enable data residency, you tag your configuration with the compliance frameworks you’re subject to: HIPAA, GDPR, PCI-DSS, SOX, FedRAMP, SOC 2, and others. This doesn’t enforce compliance automatically — that’s your team’s responsibility — but it establishes the policy context and ensures data residency rules are applied consistently.
The agent’s execution response includes a data residency report: which fields were retained locally, whether PII was redacted, and an output hash (SHA-256) for audit purposes — even when the content itself stays local.
Scoping
Agents can be scoped by department and environment. A production agent in the finance department only receives steps from finance workflows running in the production environment. A staging agent receives test executions without affecting production workloads.
Empty scope means the agent handles everything. Scoping is optional and additive — useful for organizations with strict separation requirements.
Deployment options
The execution agent runs anywhere containers run:
- Docker — Single container deployment for simple setups
- Kubernetes — Helm chart for orchestrated environments with auto-scaling
- ECS — AWS-native container service
- VM — Traditional virtual machine deployment
The agent reports its runtime environment (Docker, K8s, ECS, VM) and network context (internal IP, K8s namespace, container ID) to the control plane for operational visibility.
Availability
Hybrid/VPC deployment is available on Enterprise plans. Includes agent registration, health monitoring, data residency controls, PII redaction, and compliance framework tagging. Learn more about enterprise features or talk to sales.