Skip to main content
Building a Piper Agent takes just a few minutes. You work through four configuration steps — Schedule, Scope, Brain, and Capabilities — and the agent starts running automatically on the cadence you define. The most impactful step is writing clear, specific instructions in the Brain section, which determines the quality of everything the agent produces.
1

Schedule — When Does It Work?

Define how often your agent runs. You can set it to trigger on a recurring cadence or after specific events.Common scheduling options include:
  • After every call — ideal for real-time coaching or immediate post-call summaries
  • Daily — good for pipeline monitoring or overnight digests
  • Weekly — best for rollup reports, sentiment snapshots, or team-level reviews
Choose a frequency that matches how often you need fresh insights. A weekly cadence works well for strategic reviews; a post-call trigger is better for rep-level coaching.
2

Scope — Who Does It Watch?

Set which team members’ calls the agent analyzes. You can target:
  • All users in your workspace
  • Specific teams or groups
  • Individual reps or account managers
Keeping the scope focused improves result quality. A coaching agent scoped to one rep’s calls will give tighter, more actionable feedback than one trying to cover an entire organization at once.
A single run analyzes up to 100 meetings, and a scheduled run covers up to 15 targeted users at a time. If the scoped users generate more activity than that, narrow the scope or increase the run frequency so no meetings are missed.
3

Brain — Instructions & Principles

Write the instructions that tell your agent what to look for, how to analyze it, and what to output. This is the most important part of your agent — the quality of your instructions directly determines the quality of your results.Your instructions should answer:
  • What signals or patterns should the agent look for?
  • What should it flag, score, or summarize?
  • What format should the output take (bullet list, table, narrative, etc.)?
  • Are there specific meeting types or time frames to include?
Tips for writing great agent instructions:
  • Be specific about what the agent should look for — vague instructions produce vague outputs. Instead of “summarize the call,” write “identify whether the rep asked a discovery question about budget, timeline, and decision-maker in the first 10 minutes.”
  • Define the output format clearly — specify whether you want a bullet list, a scored table, a Slack-style message, or a narrative paragraph.
  • Include examples of what “good” and “bad” looks like if relevant — this anchors the agent’s judgment to your standards.
  • Mention meeting types or time frames — for example, “analyze demos and discovery calls from the past 7 days” is far more precise than “look at recent calls.”
  • One clear purpose per agent — agents with a single, focused goal outperform multi-purpose agents every time. Build separate agents for coaching, risk detection, and competitive tracking rather than combining them.
If you’re unsure where to start, use one of the pre-built templates as a foundation and customize from there.
4

Capabilities — Tools

Select which tools and data sources your agent can access during its runs. Granting the right capabilities ensures the agent has the context it needs without unnecessary access.Available capabilities include:
  • Transcriptions — read the full text of recorded calls
  • CRM Data — access deal, contact, and opportunity records from your connected CRM (HubSpot, Salesforce, or Pipedrive)
  • Notes — read meeting notes created by you or your team
  • Tasks — view and create follow-up tasks
  • Email — read threads from your own connected Gmail/Outlook mailbox that relate to a contact or deal. Fetched on demand and not stored — this doesn’t pull emails logged in your CRM (e.g. Salesforce EmailMessage records), and it’s scoped to your own mailbox only.
If no CRM is connected, the CRM Data capability has nothing to read — agents fall back to the transcripts, notes, and tasks stored in Piper.
Enable only the capabilities your agent needs. A coaching agent focused on call quality only needs Transcriptions. A deal risk detector benefits from combining Transcriptions with CRM Data for full context.

​
After You Save

Once you save your agent, it begins running automatically on the schedule you defined. You can return to any agent at any time to update its instructions, change its scope, or adjust its schedule.
There is no set limit on the number of agents you can create in Piper. Build as many as your team needs — one per use case is the recommended approach to keep each agent focused and its outputs easy to act on.
Yes — after saving your agent, you can trigger a manual run to see how it performs before the scheduled cadence kicks in. Review the output and refine your instructions in the Brain step if needed.
Chat is interactive — you ask a question and receive an answer immediately. Agents are automated — they run on a schedule and surface insights proactively, without you needing to prompt them each time. Use Chat for on-demand questions and Agents for recurring workflows.