Agents

​

When to use an Agent vs a Flow

• Conversation/search analysis (e.g., “find all transcripts with rebill disputes last week”).
• Internal operations (e.g., “tag these 20 calls as ‘Outage’ and summarize top reasons”).
• Back‑office tasks driven by function calls (reporting, data lookups, ticket updates).

• Customer‑facing conversations (voice, SMS, chat, web).
• Stateful logic, events, and channel integrations.
• Publishing and testing across platforms.

Bottom line: Agents are for your team; Flows are for your customers.

​

Creating a New Agent

1. Go to Build → Flows, click Create Agent (top‑right).
2. Or open Build → Agents (if enabled in your workspace) and click New.

​

Fields

• Name – What your team will see on the card in the Flows grid.
• Description – One‑line purpose statement.
• Prompt – The system instructions that define the agent’s role, guardrails, tools, and workflow.
• Tools – Select which functions the agent is allowed to call (see below).

​

Agent Details Page

• Role – What the agent is (e.g., “conversation insights analyst”).
• Goals – What outcomes it’s responsible for.
• Capabilities – What it can do (search, count, retrieve, tag, summarize, reference docs).
• Tool rules – What to call, when, and how to confirm.
• Tone & constraints – Brief, accurate, no speculation, ask before doing risky actions.

​

Prompt template (copy/paste)

Act as a <ROLE> within our AI contact center platform. You will help internal staff quickly find,
filter, and understand customer conversations. You must:
- Prefer facts from tool calls over assumptions.
- Use the tools listed below to search, count, retrieve, tag, and summarize transcripts.
- Ask for missing inputs once, then proceed with best‑effort defaults.
- If a tool is marked “Requires Approval”, ask for confirmation before running it.
- Be concise and structured: short bullets, callouts, and one clear next step.

Workflow:
1) Understand the request and list the tools you’ll likely need.
2) Collect key filters (date range, keywords, tags, agent, flow).
3) Call tools step‑by‑step; cite what you used and why.
4) Return a short answer and what you can do next on request.

Tip: Use direct language like “must” and “will” instead of “should/can,” and avoid over‑detailed examples that the agent might copy literally.

​

Tools & Approvals

• Enable/Disable – Check the box to allow a tool.
• Approval – Set No Approval Needed or Requires Approval. When approval is required, the agent will ask you to confirm before executing the function.

• listTranscripts – Search using filters (keywords, tags, dates, agent, flow).
• countTranscripts – Return counts for filters (trend checks, QA sampling).
• getTranscriptReport – Stats like length, turns, and message counts.
• getReadableTranscript – Customer‑friendly transcript view on request.
• getTagsForTranscript – Retrieve tags for review or QA.
• Domain‑specific tools (e.g., getAllCampaigns) as needed.

Keep the tool list tight. The more tools you permit, the more the agent must reason about which one to use. Fewer tools = cleaner behavior.

​

Using an Agent (Chat)

• Start with prompts like “What can you help me with?” or “Today’s conversations.”
• Ask targeted questions (“Find outage calls this week tagged ‘billing’, summarize top 3 themes.”).
• The agent will call the allowed tools, then return a concise answer and recommended next step.

​

Chat History

​

Explore Agents

​

Example: Conversation Analyst Agent

• Search transcripts by keywords, tags (key–value pairs), agents, or date ranges.
• Count transcripts that match filters.
• Retrieve readable transcripts when specifically asked.
• Analyze call metadata (duration, turns, messages).
• Read and apply tags to transcripts where appropriate.
• Reference internal documents or playbooks when provided.

1. listTranscripts – Gather candidates.
2. countTranscripts – Size the set.
3. getTranscriptReport – Inspect stats.
4. getReadableTranscript – Provide the transcript only if requested.
5. getTagsForTranscript – Show tags or suggest additions.

​

FAQs