sandbase-harness: A Local-First Agent Runtime for DeepSeek Harness
sandbaseai/sandbase-harness
Local-first agent runtime for DeepSeek Harness with sandboxing, audit, and replay.
Provides a production-grade agent runtime for DeepSeek Harness with persistent sessions, sandboxed tools, audit replay, and a local console.
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 Repository
sandbase-harness Repo Summary
SandBase Harness is a DeepSeek Harness (DSH) plugin (DSH plugin) for agent runtime, maintained by sandbaseai under the Apache-2.0 license. It is a local-first, CMA-compatible open-source agent runtime that addresses production-grade agent needs beyond the model loop: persistent sessions, tool governance, sandbox boundaries, credential handling, memory, auditability, and a UI for humans to inspect what happened. Core capabilities include SQLite-backed metadata storage, resumable SSE event streams, multiple sandbox backends (local process, Docker, Kubernetes, self-hosted worker queue), MCP toolsets, permission policies, credential vaults, skill packages, and a native DeepSeek Harness bridge over stdio MCP exposing agents, sessions, streamed turns, artifacts, and cancellation.
What core features does the SandBase Harness DSH plugin support?
SandBase Harness provides a Claude Managed Agents-style /v1 API and a local Console, with SQLite metadata, local file/skill bytes storage, resumable SSE session replay, Settings V2 configuration (model vendor, loop engine, storage, memory, sandbox), and MCP toolsets, permission policies, built-in tools, and skill packages. It also bridges DeepSeek Harness over MCP stdio, allowing DSH to natively list agents, create and run sessions, inspect results and artifacts, and stop work via mcp__sandbase__* tools.
How do I install the SandBase Harness DSH plugin?
To install SandBase Harness as a DSH plugin, use the following command:
dsh plugin --profile web add github:sandbaseai/sandbase-harness
After installation, set the MANAGED_AGENTS_URL environment variable to point to your local service, then start the DSH web interface.
Which DeepSeek Harness versions and platforms are compatible with the SandBase Harness DSH plugin?
SandBase Harness requires Node.js 22+ and npm 10+, and supports OpenAI, Anthropic, and OpenAI-compatible endpoints (including DeepSeek V4). Sandbox backends optionally include Docker (for Docker-backed sandboxes) and Kubernetes. The project was last updated in 2026-08, current version v0.3.2, and supports local-first SQLite and file storage with no required hosted control plane.
What is the license and community status of the SandBase Harness DSH plugin?
SandBase Harness is licensed under Apache-2.0, maintained by sandbaseai, with 597 stars and 57 forks on GitHub, indicating an active community. The project provides a TypeScript SDK (located at managed-agents/sdk) and includes a release gate check (npm run release:check).