**Every credential Activepieces touches (from API keys to OAuth tokens) can be routed to your own secret manager instead of a vendor database, a capability listed alongside SCIM and custom RBAC in its enterprise governance feature set.** By contrast, SaaS-only tools like Zapier and Make hold these credentials in their own cloud by design, creating a permanent interception point that cannot be audited or moved. ### Why SaaS convenience fails compliance audits During formal regulatory reviews, the ease of SaaS deployment creates significant friction because the customer lacks direct visibility into the vendor's backend operations. In regulated environments, "convenience" often translates to a loss of granular control over the data lifecycle. The following checklist summarizes this lack of transparency: * Data residency: Does the vendor store PII in a specific geographic region, or is it distributed across global data centers? * Sub-processor sprawl: Which third-party entities does the vendor share data with for logging, monitoring, or AI processing? * Audit trail gaps: Can the internal security team see every administrative action taken by the vendor’s employees on the underlying infrastructure? ## Technical requirements for private automation infrastructure A container-orchestration layer that mirrors the organization’s existing security standards is required for self-hosting an automation platform. By deploying within a Virtual Private Cloud (VPC), infrastructure teams ensure that sensitive data processing remains behind the corporate firewall, preventing the exposure of internal API keys or customer records to the public internet. ### Containerization with Docker and Kubernetes (K8s) To package the automation environment into immutable images, a production-grade self-hosted deployment relies on Docker. For high-availability requirements, Kubernetes (K8s) acts as the orchestration layer to manage these containers. The automation engine is a stateless container, allowing the system to scale horizontally. A Postgres database stores the permanent flow configurations and execution history. A Redis cache handles the task queue and temporary state, reducing the latency of high-frequency triggers. ### Hardware requirements by monthly execution volume Hardware requirements are determined by the volume of concurrent tasks and the complexity of the payloads being processed. For a baseline of 100,000 monthly executions, a single virtual machine with 2 vCPUs and 4GB RAM is sufficient, meaning that scaling beyond this volume will necessitate an immediate infrastructure upgrade. These figures represent estimates based on average task complexity, such as simple webhook triggers or basic data routing. Actual capacity fluctuates significantly based on the nature of the workflows. A heavy AI data processing flow or a script performing deep document analysis consumes far more memory and CPU cycles than a standard API call. Organizations reaching 1,000,000 monthly executions require a multi-node Kubernetes cluster. This volume necessitates redundant workers to prevent task queuing and ensure immediate processing of time-sensitive webhooks. ### Self-hosted secrets and environment variable management Security is maintained in a self-hosted environment by keeping all credentials within the organization’s own secret management system. The system injects environment variables at the container level, so developers never hardcode credentials into the automation logic itself. Integration with an internal HashiCorp Vault or a similar secrets manager rotates keys according to corporate policy without manual intervention. ## Deploying Activepieces in secure and private environments **Activepieces executes workflows across 739+ integrations while keeping all data processing and API secrets within the organization’s private network.** ### Setting up the MIT-licensed Community Edition on Docker Using a single Docker Compose file, a team can instantly deploy the Community Edition, which serves as an open-source foundation for workflow automation. Because the core is MIT-licensed, organizations retain full ownership of their automation logic without being subject to sudden licensing changes. The following table outlines the hardware requirements for these deployments, showing that even entry-level virtual machines can support a functional automation stack. | Environment | vCPU | RAM | Storage | | :--- | :--- | :--- | :--- | | Development | 1 | 2GB | 20GB | | Production (Standard) | 2 | 4GB | 50GB | | Production (High) | 4 | 8GB | 100GB | Once the containers are active, the platform can communicate with local Large Language Models, such as Mistral Large 4 or Gemini 3.8 Flash. This ensures that no prompt data leaves the facility. ### Enterprise governance: SSO, RBAC, and audit logs Centralized control over who can modify critical business processes is a requirement for enterprise-grade deployments. **Activepieces addresses this through integration with identity providers via SSO and SCIM, allowing teams to manage access for up to 5 users on the free cloud plan or at enterprise scale in private infrastructure.** The system automatically revokes a user’s access to automation tools the moment they are offboarded from the corporate directory.  Role-Based Access Control (RBAC) allows administrators to restrict specific departments to their own workspaces. Detailed audit logs record every change to a workflow and every execution attempt, providing the necessary paper trail for compliance officers. ### Air-gapped automation deployment for regulated industries When the orchestration engine, the execution environment, and the target applications all reside within a private network, fully air-gapped automation is possible. This requires the manual transfer of container images and dependencies into a private registry. The automation platform can then function even if the physical network cable is pulled. **Regulated organizations like Alan and Moneypenny run these air-gapped builds in production today, ensuring that governance tools like secret manager integration and audit logs remain available without an external connection.** ## Total cost of ownership for self-hosted automation systems By decoupling the execution volume from the monthly subscription price, self-hosting automation systems provides a predictable cost structure. While a managed service provider charges for the convenience of their perimeter, an internal deployment shifts the financial burden toward fixed infrastructure and the internal labor required to maintain it. ### License fees vs. infrastructure overhead costs Direct hardware utilization costs replace recurring per-user or per-connector fees in a self-hosted deployment. In a sovereign environment, an organization pays for the virtual machines or containers required to run the automation engine. The budget is tied to actual compute consumption rather than a vendor’s arbitrary feature gating. ### Calculating the 'maintenance tax' for self-managed updates Engineering time allocated to the internal procurement and update cycle is the primary cost driver for private infrastructure. 1. Mirroring the container image to a private registry ensures no unauthorized code enters the production environment. 2. Running the update in a staging environment for 48 hours allows the team to catch breaking changes. 3. Performing a manual database backup before applying migrations prevents data loss. This process ensures that the system remains stable and compliant, though it requires a dedicated percentage of an engineer's weekly hours. ### Scaling costs: SaaS task tiers vs. self-hosted compute power Once a task limit is reached, scaling a SaaS solution typically involves moving to a higher pricing tier. Scaling a self-hosted instance involves increasing the CPU and RAM allocated to the existing nodes. This scales linearly with the organization's standard cloud or hardware rates. | Scaling Dimension | SaaS Provider Model | Self-Hosted Model | | :--- | :--- | :--- | | Volume Limit | Hard caps on monthly tasks; exceeding these triggers overage fees or service suspension. | Limited only by available hardware; adding a node increases capacity without changing the license. | | Model Access | Often restricted to specific versions; requires a higher tier for frontier models like Gemini 3.1 Pro. | Direct connection to any internal model endpoint; no tier-based restrictions on which LLMs can be called. | | Data Egress | Charges applied for moving large datasets out of the provider's cloud into local storage. | Zero egress cost; data moves over the internal backbone between the automation engine and local databases. |  ## The Monday morning plan for private automation deployment ### Audit your most sensitive data pathways first The security team must map every integration that touches restricted data before any infrastructure is provisioned to ensure no "shadow" cloud calls are made. Customer PII stored in a local PostgreSQL database requires strict encryption at rest. API Secrets stored in a centralized HashiCorp Vault ensure no plaintext keys reside in the automation engine's environment variables. Proprietary Code hosted on an internal GitLab instance must be reachable only via the internal virtual private cloud. ### Select a pilot workflow for the local runner environment To prove the reliability of the private stack, the initial deployment should focus on a high-frequency, low-complexity task that utilizes the organization’s most capable local models. For advanced reasoning and agentic coding tasks, the team should target workflows that can leverage GPT-6 Astra via a private Azure OpenAI endpoint or a locally hosted Mistral Large 4. 1. Provision VPC and private subnet. 2. Deploy Postgres and Redis instances. 3. Run Docker Compose with the automation engine image. 4. Configure OIDC/SSO for team access. 5. Validate local model connectivity. ### Backup and disaster recovery for automation engines Internal DevOps teams are responsible for resilience in a self-hosted environment, which necessitates a rigorous schedule for state and configuration backups. The internal DevOps team must automate hourly snapshots of the persistent database to an off-site, encrypted S3-compatible storage bucket. Integrity checks of the Redis cache ensure that long-running tasks can resume after a service restart. ## Frequently asked questions about self-hosted automation ### Can I run self-hosted automation without an internet connection? When the orchestration engine, the execution environment, and the target applications all reside within a private network, fully air-gapped automation is possible. This network must have no outbound route to the public internet. This architecture ensures that sensitive credentials and internal data payloads never leave the physical or virtual perimeter of the organization. * Mistral Large 4 is a flagship multimodal model for complex reasoning and agentic tasks within the local perimeter. * Z.ai GLM 5.3 provides a deep context window for processing large internal document repositories without external data egress. * Ministral 3 8B offers an efficient option for high-volume text and vision processing on constrained local compute resources. ### How do I handle updates without breaking existing workflows? A staged deployment strategy is required for managing updates in a self-hosted environment, where the new version of the automation binary is verified against a mirrored database before being promoted to production. 1. Clone the production database to a staging environment to ensure the update script handles existing flow logic correctly. 2. Deploy the new container image to a test cluster and trigger a suite of smoke tests against internal service mocks. 3. Perform a rolling update of the production nodes, allowing you to revert to the previous image immediately if telemetry shows a spike in execution errors. ### Does self-hosting improve workflow execution latency? By eliminating the multiple network hops and TLS handshakes required to send data from your internal systems to a public cloud endpoint and back, self-hosting significantly reduces execution latency. This proximity means a high-frequency synchronization task between a local database and an internal ERP system completes in milliseconds rather than seconds. The data packets never traverse the public backbone or compete for bandwidth with external tenants. ## Related reading - [Self-Hosted vs Cloud Security Automation 2026: Cost & Data Residency](https://www.activepieces.com/blog/self-hosted-vs-cloud-security-automation-for-2026) - [AI Data Sovereignty for Insurance Agents: 2026 Guide](https://www.activepieces.com/blog/best-ai-tools-for-insurance-agents-2026-privacy-guide) - [Migrating Your Prompt Library Out of Vellum: Self-Hosted Automation](https://www.activepieces.com/blog/migrating-your-prompt-library-out-of-vellum-with-automation) ## References - [Activepieces](https://github.com/activepieces/activepieces/tree/main/packages/pieces/community)By relying on a provider's cloud perimeter, an organization is forced to accept a generic security architecture that may not align with internal hardening standards.