Building a Summarisation Agent

Recurring summarisation is one of the most immediately valuable use cases for the Cosine CLI — especially for teams that generate a lot of content to process: customer calls, meeting recordings, support tickets, or feedback surveys.

The Idea

Instead of manually reviewing call recordings or transcripts every week, you create a summarisation agent folder with a persona specifically designed for that task. Each time you need a summary, you jump into that folder and prompt the agent. It already knows:

• What to look for (the trends, themes, and questions you care about)
• How to format the output
• Where to find the data (via MCP or saved files)
• What’s been summarised before

Setting Up the Folder

mkdir ~/call-summaries
cd ~/call-summaries
cos init

Edit agent.md to define the summarisation persona:

You are an expert analyst specialising in extracting customer insights from sales and support call recordings.

Focus areas:
- Recurring objections and blockers
- Features customers ask for most
- Terminology customers use to describe their problems
- Moments of confusion or friction

Output format:
- Always produce a structured Markdown summary
- Group findings by theme, not chronologically
- Include specific quotes where relevant
- Flag any patterns that appear more than twice

Data sources:
- Use the Attio MCP to retrieve recent call recordings
- Filter to calls tagged as "demo" or "sales" in the last 7 days
- Only include calls that I (the current user) was on

Running a Weekly Summary

Each Monday, you can jump into this folder and run:

“Summarise all my call recordings from last week. Look for recurring themes and flag anything that came up more than twice. Save the output to 2026-03-03-summary.md.”

Switch to Auto mode and let it run. By the time you’ve had a coffee, the summary is done.

Because each summary is saved as a dated file in the folder, you can also ask the agent to spot longer-term trends across multiple weeks:

“Compare this week’s summary with the last three weeks. What themes are growing in frequency?”

Connecting to Call Recording Platforms

If your call recordings live in a platform with an MCP (like Attio, Gong, or a custom integration), configure the MCP first (see Setting Up MCPs and MCP Examples). Then your agent can pull the recordings directly without you needing to download or upload anything.

If no MCP is available, you can:

• Export transcripts and drop them into the folder manually.
• Ask the agent to look at a specific file: “Summarise the transcript at transcript-2026-03-03.txt.”

Sharing Summaries with Other Agents

Once the summary is saved as a file, other agents can read it. This is how you create a pipeline between agents:

1. The summarisation agent produces customer-pain-points.md.
2. The content agent reads that file and produces blog posts addressing those pain points.
3. The sales agent reads that file and personalises outreach based on recurring objections.

You coordinate this by pointing each agent at the relevant file:

“Read ~/call-summaries/customer-pain-points.md and use the recurring themes to plan our next 5 blog posts.”

Key Takeaways

• A summarisation agent folder accumulates knowledge over time — each run builds on the last.
• Use an MCP to pull data directly from your call recording platform.
• Save summaries as dated files to enable trend analysis across time periods.
• Summaries can be shared with other agents as input files, creating a multi-agent pipeline.

---

Next: Creating a Sales Outreach Agent