Firedrill

Firedrill

Firedrill is a simulation platform for AI agents. It gives the agent you are building a practice ground: a synthetic world of users, stateful tools, data and scenarios where it can act without touching production. You see what your agent does before your customers do.

To set up Firedrill with a coding agent, give it this prompt:

Set up Firedrill for this project. Fetch https://firedrill.run/agent-setup.md and follow it.

Who it is for

Teams building AI agents that call tools, change state or talk to people: support agents, operations agents, sales and finance agents, coding agents and internal copilots. Firedrill is operated by your coding agent (Claude Code, Codex, Cursor or any MCP client) or from the web app.

How it works

  1. Connect. Your coding agent connects over the remote MCP server at https://mcp.firedrill.run or the firedrill-run CLI. You approve access in your browser.
  2. Describe your agent. The coding agent reads your repository and registers how your agent starts, which tools it calls and what it needs. Source code is not uploaded.
  3. Build its world. It proposes synthetic users, scenarios, tools and evaluations. You review the plan and the credit estimate before anything runs.
  4. Run and improve. Your agent runs on your machine against its synthetic world. Firedrill records what happened and scores it. Change the agent and run the same scenario again to compare.

What a world contains

  • Synthetic users that hold a conversation with your agent, with goals, personas and changing behavior.
  • Stateful synthetic tools that stand in for the services your agent calls. They keep state across calls and can be set to fail, slow down or return edge cases.
  • Scenarios that combine users, tools and starting data into repeatable situations, including the rare and risky ones.
  • Evaluations that score outcomes and check hard rules, with the evidence for each result.

Where things run

Your agent, its code and its model calls stay on your machine. Firedrill keeps the worlds, scenarios, synthetic tools and reports in the cloud. Real accounts are never used unless you approve a specific one.

Pricing

Start free with 500 credits a month and no payment card. Paid plans are Starter ($29 a month), Team ($99) and Scale ($299). Credits pay for synthetic users, scoring, synthetic tools and world preparation. Your own model provider bills you separately. See pricing and credits. For a synthetic version of your business with your systems, data and policies, see Firedrill Enterprise.

Questions

Is Firedrill a sandbox for my agent's code? No. Your agent runs on your machine with your own permissions. Firedrill replaces the outside world it talks to: users, tools and data.

Do I need to upload my code? No. Registration sends names, hashes and interface details only.

Which coding agents work with Firedrill? Claude Code, Codex, Cursor, VS Code and any client that supports remote MCP servers over HTTP. The CLI works in any terminal with Node.js 22, 24 or 26.

Does Firedrill call real services? Not by default. Every dependency is synthetic unless you approve a real account for it.

Links