sandbase-harness: A Local-First Agent Runtime Plugin for DeepSeek Harness
sandbaseai/sandbase-harness
SandBase Harness is a local-first AI agent runtime with sandboxed sessions, MCP tools, memory, credentials, and audit/replay, plus a built-in console, runnable as a DeepSeek Harness (DSH) plugin.
sandbase-harness is a local-first AI agent runtime plugin for DeepSeek Harness that extends the model loop into a full production-grade agent infrastructure. It solves the problem of long-running agents lacking persistent sessions, tool governance, sandbox boundaries, credential management, and auditability, letting you run and inspect agents safely on your own machine or infrastructure. Core capabilities include sandboxed sessions, MCP tools, memory, credential vaults, audit/replay, and a built-in console, supporting models like OpenAI, Anthropic, MiniMax, and DeepSeek V4.
How to Install
dsh plugin --profile web add github:sandbaseai/sandbase-harness- Category
- Skills & Agents
- Platform
- DSH Plugin
- Author
- sandbaseai
- Distribution
- Plugin
sandbase-harness Key Features
sandbase-harness Repo Summary
What Does It Do?
SandBase Harness is a local-first AI agent runtime DSH plugin for DeepSeek Harness that extends the model loop into a full production-grade agent infrastructure. It solves the problem of long-running agents lacking persistent sessions, tool governance, sandbox boundaries, credential management, and auditability, letting you run and inspect agents safely on your own machine or infrastructure. Core capabilities include sandboxed sessions, MCP tools, memory, credential vaults, audit/replay, and a built-in console, supporting models like OpenAI, Anthropic, MiniMax, and DeepSeek V4.
Core Features
- Sandboxed execution: supports local processes, Docker (per-session containers), Kubernetes, and self-hosted worker queues for safely running generated code.
- Persistent sessions and replay: SQLite-backed sessions with resumable Server-Sent Events for audit and debugging.
- Tool governance: MCP toolsets, permission policies, built-in tools, and skill packages to control agent tool access.
- Credential vault: centralizes API keys and sensitive information with approval workflows.
- Multi-model support: connects to OpenAI, Anthropic, MiniMax, and DeepSeek V4 via OpenAI-compatible endpoints.
- Local-first storage: uses SQLite and file storage by default with no required hosted control plane.
How to Use This Plugin?
After enabling it in DSH, manage agents through the built-in console (default port 3000). Configure a model provider in Settings > Models, then create sessions and run agents. To adjust sandbox or storage behavior, modify settings in Settings V2, which supports form and JSON modes with automatic restart.
This page is an independent rewrite of the plugin's official README — for authoritative documentation and the latest changes, refer to the source: sandbaseai/sandbase-harness. The plugin is third-party code that runs on your machine once installed; inclusion does not imply endorsement — please review the source before installing.