OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
Agent Infrastructure

Sandbase Harness

Sandbase Harness is a self-hosted TypeScript runtime for AI agents, providing persistent sessions, sandboxed tools, MCP connectivity, memory, credential handling, audit and replay, and a local console.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Sandbase Harness — OpenViST project profile in Agent Infrastructure.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use Sandbase Harness when you need a runtime layer for persistent AI agent sessions and want control over sandboxing, tool access, runtime state and operational records.

Who it is for

  • Developers and teams operating AI agents on their own machines or infrastructure.
  • Operators who need to inspect, resume, replay or debug persistent agent sessions.
  • Teams that want to govern agent tool access through MCP toolsets, permissions and approvals.

Documented uses

  • Run agent sessions within local, Docker, Kubernetes or self-hosted worker sandboxes.
  • Connect agents to MCP-compatible tools and manage access with policies and approvals.
  • Keep session memory and credentials alongside agent runtime operations.
  • Inspect and replay agent activity through resumable event streams and the console.
Before you choose

Requirements & limits

The project is open source under the Apache-2.0 license. The documented local setup requires Node.js 22+, npm 10+ and a model-provider API key; Docker is optional and only required for Docker-backed sandboxes.

  • The project runs on local or self-managed infrastructure and does not require a hosted control plane.
  • Metadata uses SQLite by default; local file and skill data is stored in the workspace state directory.
  • Provider configuration changes require a runtime restart, and model IDs must match the selected provider.
  • The documented interfaces include a local console, a /v1 API and a TypeScript SDK.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Set up the runtime and configure a model

Build and start a local Sandbase Harness workspace, then configure a provider and model in the console.

Before you start

  • Node.js 22 or later
  • npm 10 or later
  • An API key for OpenAI, Anthropic, MiniMax, or an OpenAI-compatible endpoint
  • Docker only if using Docker-backed sandboxes

Steps

  1. Clone the v0.3.8 repository release, enter the project directory, install dependencies with npm ci, and build with npm run build.

  2. Create a workspace directory, run the built CLI's init command there, then start the runtime.

  3. In Settings > Setup, enter and save the provider API key; restart the runtime for the saved setting to take effect.

  4. In Agent models, set a model ID served by the selected provider.

Expected outcome

A workspace is initialized and the runtime serves its API and local console at 127.0.0.1:3000, with provider and model settings configured.

Guide limitations

  • A provider setting saved in the console does not become active until the runtime restarts.
  • The model ID must be one served by the selected provider; an unsupported ID causes a model_not_found error.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
13 October 2026
Original website title
sandbaseai/sandbase-harness · Official homepage title · retained in its source language
How we review sourcesSuggest a correction