sandbase-harness plugin: Local-First Agent Runtime with DeepSeek Harness Bridge

An open-source CMA-compatible runtime for running local multi-agent systems with multiple model providers, MCP tools, scenarios, and a dashboard.

What this plugin does

The sandbase-harness plugin is a local-first agent runtime that provides persistent sessions, sandboxed tools, memory, credential management, and audit trails. It includes a built-in Console for human inspection and offers a DeepSeek Harness bridge over MCP stdio, enabling DSH to interact with agents and sessions.

The sandbase-harness plugin is a runtime layer for AI agents, not a visual workflow builder or model SDK. It handles persistent sessions, tool governance, sandbox boundaries, credential handling, memory, and auditability. The runtime is SQLite-backed by default, with support for local process, Docker, and Kubernetes sandboxes. It exposes a Claude Managed Agents-style /v1 API and a local Console. A notable feature is the DeepSeek Harness bridge over MCP stdio, which allows DSH to list agents, create and run sessions, inspect results, and cancel work through native mcpsandbase tools.

Documented capabilities

Session management

The sandbase-harness plugin provides persistent sessions with resumable Server-Sent Events for replay and debugging. Sessions can be created, run, and stopped through the API or the DeepSeek Harness bridge.

Sandboxed tools

Tools run in sandboxed environments with configurable backends: local process, Docker per-session containers, Kubernetes via kubectl exec/cp, or a self-hosted worker queue. This isolates tool execution.

DeepSeek Harness bridge

The plugin can run as a DSH plugin, starting a managed-agents MCP server over stdio. DSH can then list agents, create sessions, stream turns, and inspect artifacts through native mcpsandbase tools.

Practical use cases

Running multi-agent systems locally

Teams can run multi-agent systems on their own infrastructure with any model provider, using the sandbase-harness plugin's local-first runtime and sandboxed tool execution.

Integrating with DeepSeek Harness

Users can install the sandbase-harness plugin into a DSH profile, allowing DSH to manage agents and sessions through the MCP bridge, extending DSH's capabilities.

Auditing agent activity

The built-in Console and audit trails allow humans to inspect what agents did, including sessions, tool calls, and artifacts, supporting governance and debugging.

Review before you install

Read the source, recent releases, requested permissions, network access, and maintainer history before trusting a third-party plugin.

Frequently asked questions

How does the sandbase-harness plugin integrate with DeepSeek Harness?

The plugin can be installed as a DSH plugin, starting a managed-agents MCP server over stdio. DSH can then list agents, create sessions, stream turns, and inspect artifacts through native mcpsandbase tools.

What sandbox backends does the sandbase-harness support?

It supports local process, Docker per-session containers, Kubernetes via kubectl exec/cp, and a self-hosted worker queue. Docker is optional and only needed for Docker-backed sandboxes.

What are the requirements to run the sandbase-harness?

Node.js 22+, npm 10+, and a model provider API key (OpenAI, Anthropic, or OpenAI-compatible endpoint). Docker is optional for Docker-backed sandboxes.