For AI agents: The documentation index is at https://docs.digitalocean.com/llms.txt. Markdown versions of pages use the same URL with index.html.md in place of the HTML page (for example, append index.html.md to the directory path instead of opening the HTML document).

Sourced Research Desk is a research agent for PMs, analysts, and operators who need answers they can defend. Ask a question in plain language; the agent plans queries, gathers public sources, builds a claim table with URL and date on every factual claim, and drafts a markdown brief with citations and conflict callouts when sources disagree. Use it when you need a decision-ready brief without babysitting browser tabs or risking an auto-post to Slack or email. Research-only runs keep the brief in artifacts by default; if you opt into outbound, a human approval gate fires before any send so you stay in control of what leaves the room. Ideal for competitive diligence, market sizing, policy research, and any brief that has to survive a skeptical review, when “the model said so” is not enough.

Why use it

• Citation-backed briefs with URL and date on every factual claim
• Conflict callouts when sources disagree, so you see the tension, not a smoothed narrative
• Human approval before any Slack or email send; deny keeps the brief local
• Research-only mode by default: no outbound noise, no surprise posts
• End-to-end flow from clarifying the question to a finished markdown brief
• Built for DigitalOcean Managed Agents (MARS) with a sandboxed LangGraph runtime
• Works from a clear question; no custom RAG stack or paid search API required

Requirements

• DigitalOcean Managed Agents / Harness Runtime (sandboxed LangGraph session)
• OpenAI-compatible inference provider (DigitalOcean Gradient / Inference recommended)
• Strong reasoning models work best (e.g. deepseek-v4-pro or comparable)
• Optional: Slack or email targets only if you enable outbound (approval still required)

Software Included

This Marketplace listing does not include a detailed software list.

Getting Started After Deploying Sourced Research Desk

1. Create the agent with your inference API key. No search API key is needed; the agent fetches public http and https sources directly.
2. Start a session and send a question, for example Research: What changed in managed agents pricing this quarter?. The agent replies with a short plan and asks **Start research run?** with the question, the angle it will take, and the outbound setting. Approve to start fetching. Deny stops without fetching anything.
3. Read the brief. Every factual claim carries a source URL and date, and disagreements between sources are called out. Sources that could not be fetched are listed instead of cited. If there is not enough sourced material, the agent says so rather than padding the brief.
4. Chat sessions are research-only, and the brief stays in the session. A run that sets outbound to slack or email with a destination stops at **Send research brief?** with a preview, claim count, and conflict tally. Choose **Send** or **Keep local only**. In this version, Send records the send in the run; it does not deliver a real Slack message or email yet.