Build your own agent orchestration platform: a team of specialist agents for budgeting, forecasting, variance, PVM, balance sheet, close and more, with a manager that routes the work and waits for your sign-off.
The org-chart model made popular by orchestrators like Paperclip, applied to finance. Works with Claude, Copilot, ChatGPT or Gemini. Press Run the close to watch a full month end, and approve the entries yourself.
Fourteen ready-made finance agents. Add up to eight to your team: the control room, every prompt, the test kits and the ROI estimate rebuild around your choice. Open any agent to see its output, instructions, log and test.
Every prompt on this page fills itself with these details. Nothing leaves your browser: values are saved only on this device. Leave a field blank and it stays as a red placeholder in the prompts.
Every tool gets you to level 2. Scheduling, approvals and code-level orchestration decide how far you go. Tap a column to pick your tool for the whole page.
✓ built in · ◐ possible with extra setup or code. Features change quickly and depend on your plan and admin settings, so check before you commit.
Each prompt rewrites itself for your tool, your company and your team. Copy it, paste it, and tick the step off when it's done. Your progress is saved on this device.
Want it all at once? Copy the seven prompts set up for Claude, or the instructions for every agent in your team.
A rough estimate for your team, based on the agents you picked. Baseline hours assume a finance team of 12. Change any agent's hours to match your own numbers.
Team size follows your setup (FP&A plus accounting) until you move the slider. Run cost adds each agent's monthly usage limit plus licences for the whole team.
Bake these into every agent's instructions in step 2, the orchestrator in step 4 and your tests in step 6.
Agents propose journal entries with support. A named person approves and posts them in the ERP.
Bridges sum to the variance, the trial balance nets to zero, and agents refuse to reply until the checks pass.
Each action records the agent, the data version it used, the output and who approved it.
Each agent passes its test pack before going live and again after every change to its instructions.