Agents & automationOpen source
Sluicegate
Open-source agent runner with approval gates, budgets and a full audit trail
About this tool
Sluicegate is a self-hosted runtime for agents that call tools: APIs, databases, browsers and internal scripts. Every tool is declared with permissions, and risky actions such as sending money or deleting records pause for a human approval in Slack or the web console. Each run has a spending limit and a replayable trace.
Platform engineers at companies that want agents in production without handing them unrestricted access use it as the layer between models and internal systems. It is model-agnostic and written in Go with Python and TypeScript SDKs. It is infrastructure, not a product: you will write code to define tools and workflows, and the web console is functional rather than polished.
Highlights
- 01Risky tool calls pause for human approval
- 02Per-run spending limits and replayable traces
- 03Model-agnostic, with Python and TypeScript SDKs