Skip to main content

​
What is OpenSink?

OpenSink is a control layer for AI agents in production. Once agents move beyond prototypes, teams run into the same problems:
  • Agent outputs get scattered across Slack, email, logs, and dashboards
  • There’s no safe way to pause an agent, ask a human for approval, and resume
  • Changing agent behavior means redeploying code
  • Nobody can tell what an agent actually did, or why it failed
OpenSink solves these by giving your agents a central place to store results, track execution, request human input, and manage configuration — all through simple HTTP APIs. It works with any framework, any language, and any platform. OpenSink does not run your agents or care how they’re built. It just gives them the production infrastructure they’re missing.
Example: You build a trading agent that reads market news and proposes trades. With OpenSink, that agent can store its findings in a Sink, run inside a tracked Session, pause and ask a human to approve the trades via an Input Request, and resume execution automatically once approved — all without building any of that infrastructure yourself.

​
What OpenSink gives you

Sessions

Durable, resumable execution records. Track state, pause for human input, and resume without losing context.

Human-in-the-loop

Pause agents for approvals or input. Schema-driven forms render automatically. Execution resumes when the human responds.

Sinks & Items

Named containers for agent-produced information. Store results, decisions, and findings — not logs, not metrics.

Configurations

Versioned, schema-validated settings. Change agent behavior without redeploying. Full history and rollback.

Background Jobs

Scheduled HTTP triggers for periodic agent execution. No cron servers, no infrastructure to manage.

Visibility

Inspect, debug, and audit every agent execution. Activity timelines, state inspection, and full-text search.

​
How it works

OpenSink is a set of HTTP APIs. Your agent calls them during execution. 
Your agent runs → calls OpenSink APIs → OpenSink stores state, results, and activities
There is no SDK requirement, no runtime coupling, and no framework dependency. If your code can make an HTTP request, it can use OpenSink.

​
Get started

1

Create an account

Sign up at opensink.com. Create a workspace. You’ll land in your dashboard with a default Sink ready to go.
2

Send your first item

Copy the example command from your dashboard and run it. Your first item appears instantly — searchable and inspectable.
3

Register an agent

Create an agent, give it a configuration, and start running sessions. Log activities, request human input, and store results in sinks.

Quickstart

Follow the step-by-step guide to go from zero to your first item in under a minute.

​
Explore the platform

Agents

Register agents, track their status, and connect them to sessions, configs, and sinks.

Sessions

Durable execution records with resumable state and human-in-the-loop support.

Sinks & Items

Store and search the meaningful information your agents produce.

Configurations

Versioned settings with schema validation, variants, and rollback.

Input Requests

Pause agents for human approval with schema-driven forms.

Background Jobs

Schedule periodic agent execution with simple HTTP triggers.

Activities

A structured timeline of everything that happens during a session.

API Reference

Full HTTP API documentation with endpoint examples.