SandBase Harness
SandBase Harness is a local-first, self-hosted AI agent runtime and MCP bridge that runs sandboxed agent sessions, memory, credential vaults and audit trails on the user's own machine or infrastructure, with a built-in web console.
SandBase Harness runs AI agent sessions with sandboxed tool execution in local, Docker, Kubernetes or other self-hosted workers, so generated code does not run directly on the host. It keeps persistent sessions with resumable event streams, audit logging and replay for inspecting long-running agents, and controls tool access through MCP toolsets, credential vaults, permission policies and approval steps.
It works with OpenAI, Anthropic, MiniMax and other OpenAI-compatible providers, including DeepSeek, and stores its state locally in SQLite and the filesystem with no hosted control plane required. It installs by cloning the repository, building it with npm, and initializing a new agents directory, after which a local dashboard is available in the browser.
It is written in TypeScript and licensed under Apache-2.0. A separate, lighter-weight project by the same team, SandBase CLI, offers a local MCP bridge without the full runtime.