Use Cases

​

What people build with OpenSink

OpenSink is general-purpose — any agent that needs structure, visibility, or human interaction can use it. But certain patterns come up repeatedly.

---

​

News & research monitoring

An agent that runs periodically, collects information from external sources, and stores the results for later review. How it works:

1. A Background Job triggers the agent on a schedule
2. The agent fetches data from APIs, RSS feeds, or web scraping
3. An LLM filters, summarizes, or categorizes the results
4. Important findings are stored as Items in a Sink
5. Activities log what the agent did during each run

OpenSink primitives used:

---

​

Trading & financial agents

An agent that analyzes market data, proposes actions, and waits for human approval before executing. How it works:

1. The agent analyzes market conditions and proposes trades
2. Proposals are stored in the Session state
3. An Input Request pauses execution and asks for approval
4. A human reviews and approves or rejects
5. The agent resumes via its Execution Endpoint and executes approved actions
6. Results are stored in a Sink

OpenSink primitives used:

---

​

Content & report generation

An agent that generates content — reports, summaries, emails — and optionally asks for review before publishing. How it works:

1. The agent gathers source data and generates a draft
2. The draft is stored in the Session state
3. An Input Request asks a human to review and edit
4. The agent publishes the final version and records it in a Sink

OpenSink primitives used:

---

​

Competitor & market analysis

An agent that tracks competitor activity and surfaces changes worth knowing about. How it works:

1. A Background Job triggers the agent daily or weekly
2. The agent checks competitor websites, social media, or product pages
3. An LLM identifies meaningful changes and generates summaries
4. Results are written to a Sink, searchable and organized by competitor
5. Configurations control which competitors to track and what to look for

OpenSink primitives used:

---

​

Alert & escalation agents

An agent that monitors systems and escalates issues that require human attention. How it works:

1. The agent monitors metrics, logs, or external services
2. When something requires attention, it creates an Input Request
3. A human triages the alert and provides instructions
4. The agent acts on the response — acknowledges, escalates, or remediates
5. All actions are logged as Activities for audit

OpenSink primitives used:

---

​

Combining patterns

Most real agents combine several of these patterns. A single agent might:

• run on a schedule (Background Job)
• fetch and analyze data (Session)
• store results (Sink)
• ask for approval on high-stakes actions (Input Request)
• resume and execute (Execution Endpoint)

OpenSink’s primitives are composable. Use what you need, ignore what you don’t.