Skip to main content
Setting up an AI Agent in Bold takes three steps: write instructions in plain language, define what the agent is allowed to touch, and decide when it runs. You do not need a technical background — if you can explain a task to a colleague, you can configure an agent. This guide walks you through each step and shows you how to monitor your agents once they are running.

​
Step-by-Step Configuration

1

Write Natural-Language Instructions

Open the agent configuration panel in Bold and type your instructions in the same way you would brief a new team member.What good instructions look like:
  • Be specific about the trigger or timing: “Every morning at 07:00, look at yesterday’s production data…”
  • Describe the desired outcome clearly: “…and send an email to the plant manager summarising actual vs. planned output, any anomalies, and productivity by operator.”
  • Mention any edge cases you care about: “If no production data exists for the previous day, skip the email.”
Treat the instructions field as a handover note. The more context you give — what data to look at, what decision to make, what to do with the result — the more reliably the agent will perform.
You do not need to describe how the agent accesses Bold’s data — it handles that automatically based on the goal you describe.
2

Set Scoped Permissions

Permissions define the boundaries the agent operates within. For each agent, you select:
  • Which modules it can read — for example, Production, Inventory, Timesheets, Maintenance.
  • Which entities it can create — for example, Orders, Reports.
  • Which entities it can edit — for example, Recipes, Operations.
Grant only the permissions the agent genuinely needs. An agent configured to edit recipes should never have write access to financial records. Scoped permissions protect your data and make agent behaviour predictable.
A useful approach: read through your instructions and list every Bold module they mention. Enable read access for all of those modules, and add write or create access only where the agent needs to take action.
3

Choose Scheduling

Decide when the agent runs:
ModeWhen to Use
On demandAgent runs when you or a colleague start a conversation with it — ideal for interactive tasks like converting an email into an order.
HourlyBest for tasks that need near-real-time monitoring, such as flagging overdue operations.
Daily at a specific timePerfect for morning reports or end-of-day summaries.
WeeklySuited to weekly KPI digests or maintenance reminders.
For recurring agents, pick the earliest time the output is genuinely useful — there is no point running a daily report at 03:00 if your team starts at 07:00.
4

Choose the Intelligence Level

Bold lets you select how much reasoning power the agent applies to its task:
  • Lighter model — fast, cost-efficient, ideal for well-defined tasks with predictable inputs (e.g., pausing open operations at end of shift).
  • Powerful model — deeper reasoning, handles ambiguous or complex inputs (e.g., interpreting a loosely-written customer email and mapping it to the correct product references).
Start with the lighter model. If the agent makes errors on edge cases, switch to the powerful model. You will pay more credits per run, but fewer corrections to fix.

​
Monitoring Agent Runs and Credit Consumption

Once an agent is active, Bold gives you a full view of its activity:

Run History

See every execution: when it ran, what actions it took, and whether it completed successfully or encountered an issue.

Credit Consumption

Track how many credits each agent used per run and in total for the month, so you can see which agents deliver the most value.
Use the consumption view to identify agents that are using significantly more credits than expected — this usually means the instructions can be made more precise, or the intelligence level can be dialled down.
If you are approaching your monthly credit limit, Bold will alert you in advance. You can top up your credit pack at any time by contacting your Account Manager.

​
Getting Help with Your First Agent

Your Account Manager is there to help you define and configure your first agent. Bring them a description of the repetitive task you want to automate — even a rough one — and they will help you translate it into instructions, choose the right permissions, and set the schedule. After your first agent is running, you will find it straightforward to create the next ones yourself.
Review the run history to see exactly what steps the agent took and why. In most cases, the fix is a small addition to the instructions — for example, adding a condition like “only edit records created in the last 24 hours”. You can also tighten permissions so the agent physically cannot reach the records you want to protect.
Yes. Any team member with access to the agent can start a conversation with it. Each conversation is independent, so two colleagues can run the same agent simultaneously without interfering with each other’s results.