VaultRun
VaultRun is a self-hosted secure runtime for AI agents that provides isolated Docker sandboxes for executing untrusted code and commands. It exposes a REST API, a 53-tool MCP server, CLI, and SDKs for Go and Python, allowing agents to run operations with strict security controls. The platform is designed around local-first workflow assets called "missions" with verification checkpoints, memory, and swarm graphs. Every action is recorded in an HMAC-signed audit trail, and the system enforces resource limits, disables network by default, and prevents path traversal and shell injection. VaultRun is open source under Apache-2.0, with no SaaS or telemetry. It includes a management dashboard, a CI runner for GitHub webhooks, and enterprise features like OIDC and SAML SSO. The project is early-stage but fully self-hosted, giving organizations complete control over their agent infrastructure.
- Category
- Apis
- Pricing
- Free
Problem solved
VaultRun solves the problem of safely executing AI agent commands and code by providing isolated, sandboxed environments with policy checks, audit trails, and resource constraints, reducing the risk of malicious or accidental harm while maintaining full self-hosting and data privacy.
Use cases
- Run untrusted code from AI agents in isolated Docker sandboxes
- Integrate a 53-tool MCP server with Claude, OpenAI, or custom agents
- Track and audit all actions performed by AI agents with signed logs
- Define and execute reusable workflow assets (missions) with verification
- Trigger sandboxed CI pipelines via GitHub webhooks
- Manage AI agent infrastructure on your own servers without SaaS dependency
Audience
- Developers
- ML Engineers
- Enterprise
- Startups
- Hobbyists
Pros
- Fully self-hosted with no SaaS or telemetry
- Comprehensive security: Docker isolation, exec API only, signed audit trail
- Rich integration: 53-tool MCP server, Go/Python SDKs, CLI, dashboard
- Open source under Apache-2.0
- Local-first workflow assets with missions, verify checkpoints, and memory
Cons
- Early-stage open source project with limited community support
- Requires technical expertise to deploy and operate (Docker, Go)
- No managed cloud offering; all infrastructure must be self-managed
- Documentation may be incomplete for all features
Official website: https://www.vaultrun.dev/
Canonical LambdaBase page: https://www.lambdabase.com/tools/vaultrun