DEDUPED Hot AI repos, turned into use cases Get the Monday 5

How to run a multi-step data task with approval checks with Nexus Pilot

Delegate a multi-step database task to the Agent mode and review each write before it runs.

Credentials. Agent writes run against the connected database with the permissions of the connection, so use an account with only the rights the task needs.

Who it is forA data engineer or developer who needs the agent to explore a database and make a sequence of changes, and who wants to review risky writes first.

The problemMulti-step database tasks are error-prone when done by hand, and an automated agent could make accidental writes without a clear review step.

The outcomeThe agent carries out the task through restricted workbench tools, and SQL or Redis writes stop for risk analysis and approval.

What you need first

  • A machine that can run the NexusPilot desktop installer, or Bun 1.3 or later and a Rust toolchain for development
  • Connection details for at least one supported database that you are allowed to access
  • An API key or account with a supported LLM provider, if you want to use the AI features
  • Familiarity with the database you connect to, so you can review agent queries before acting on them

Steps

  1. Install Bun 1.3 or later and a Rust toolchain, then run bun install in the cloned repository.
  2. Run bun run dev:all to start the desktop app and the AI Runtime together.
  3. Add a SQL or Redis connection in the app and open the agent in Agent mode.
  4. Describe the multi-step task in natural language and let the agent explore the objects it needs.
  5. Review the risk analysis and approval prompt for each SQL or Redis write, and confirm only the changes you expect.

Prompt for your agent

Using the NexusPilot repo at https://github.com/GuqierMcl/NexusPilot, run `bun install` and `bun run dev:all`, then give the Agent mode this task on <connection name>: <task description>. Expected output: a summary of each step and each approved write. Verify that every write was shown with a risk analysis and approved by a person before it ran.

Caveat. The README says SQL and Redis writes currently use risk analysis and approval flows, so other write paths may not have the same review step.

Use this in your AI agent

Ask your agent

Using the Deduped MCP, get the setup steps for GuqierMcl/NexusPilot and "Run a multi-step data task with approval checks"

Add Deduped to Claude Code once

claude mcp add --transport http deduped https://dedupedai.com/mcp

Cursor, ChatGPT, others

Share: X LinkedIn Hacker News Email Copy link

Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer