- CollectionAgent Runtime & Infrastructure
- Classificationagent-native
- StatusOpen source · Apache-2.0 · self-hosted/local-first
- MCPAvailable
- Agent SkillUnknown
- InterfacesMCP · CLI · STDIO · http · API
What it does
Official README: “A local-first runtime for AI agents. Sessions, sandboxed tools, memory, credentials, audit trails, and a built-in Console — all running on your machine or in your own infrastructure.” — sandbaseai/sandbase-harness
SandBase Harness is a local-first runtime for operating multi-step AI-agent sessions. It combines persistent sessions and resumable event streams with sandboxed turns, MCP toolsets, credential scopes, permission policies, approvals, artifacts, memory, audit records, and replay. It can run on an operator’s machine or in self-hosted infrastructure; the exact isolation boundary depends on the selected and configured execution backend.
Agent-facing interfaces
- Catalog protocols: MCP, CLI, STDIO, http, API
- MCP: ✅ A native stdio MCP bridge is published.
- Agent Skill: ⚠️ No first-party Agent Skill package is published.
- Onboarding type:
mcp
Protocol surface
| Interface | Detail |
|---|---|
| HTTP/API | Local runtime and Console surface |
| MCP | Native stdio bridge; metadata in server.json |
| TypeScript | Runtime and CLI packages |
| Deployment | Local process, Docker, Kubernetes, and self-hosted worker paths |