Building and managing Agents

Last updated: April 19, 2026

Once you know what kind of analysis you want to run with Glyphic Agents, the next step is building a saved agent — one with a name, a prompt, a way of being triggered, and a delivery destination. This article covers the builder.

Where to find it

Open Agents in the left navigation. You'll land on the agents home (app.glyphic.ai/agents/home), which has:

  • A free-text box to run a one-time prompt (no save, no schedule — for ad-hoc questions).
  • A template library — pre-built agents grouped by tag (CRO, AE, CS, RevOps, etc.) that you can customize and save.
  • A clock icon at the top that opens a history popover of your recent agent runs.
  • A My Agents link that takes you to the full list of saved agents.

Two ways to start a saved agent

From a template

Click any template card on the home page. You'll be dropped into the builder form with the title and prompt pre-filled. Tweak as needed and save.

From scratch

Click Create an agent on the My Agents page. Pick:

  • Conversational builder — Glyphic asks you a few questions about what you want the agent to do, then generates a draft configuration. Best when you're not sure where to start.
  • Form builder — go straight into the three-step form (name, trigger, prompt, output). Best when you know what you want.

Both routes end up at the same form, just with different starting points.

The builder form

Step 1: Trigger

Choose how the agent runs:

  • Manually — the agent only runs when someone clicks "Run" on it. Good for prompts you want to save and reuse but not automate.
  • Scheduled — runs on a recurring cadence:
    • Daily — every day at the chosen hour.
    • Weekdays — Monday through Friday at the chosen hour.
    • Weekly — pick a day of the week and an hour.
    • Monthly — pick a day of the month and an hour.
    Times are at the top of the hour and run in your timezone (the timezone on your Glyphic user account).
  • Deal Stage Change — fires when a deal in your CRM moves to a particular stage. You can optionally narrow further by requiring the deal to have come from a specific previous stage. Useful for things like "when a deal moves to Closed Won, generate a CS handoff note." HubSpot and Salesforce only.

Step 2: Prompt

The prompt is the instruction the agent runs. The same prompt-design tips apply as for one-time runs — see How to Write a Prompt and the Common Glyphic Agent Prompts reference for working starting points.

Two extras here:

  • Variables — for triggered agents, you can reference fields from the trigger (e.g. for a deal-stage change, the deal name, owner, value, stage). Click Variables to see what's available for the trigger you've configured.
  • Web Search — toggle on if the agent should be able to look things up on the open web (competitor news, public company info, etc.).

Step 3: Output

Decide who gets the result and how:

  • Generalised — the agent runs once, and the same output goes to everyone you've selected. Outputs can go to a list of users (each as email and/or Slack DM) and/or to one or more Slack channels.
  • Individual — the agent runs once per selected user, scoped to that user's data (their deals, their calls). Outputs go to that user only.

Slack delivery requires the Slack integration to be connected at the org level — see Slack Integration.

Test before saving

Click Test Agent at the bottom of the form to run a preview against your real data. For deal-stage triggers, the test runs against a sample deal so you can see what the output would look like in production. Adjust your prompt and re-test until you're happy, then click Save Agent.

My Agents

The list page (Agents → My Agents, or app.glyphic.ai/agents/list) groups your saved agents into:

  • Live Agents — agents with an active trigger (scheduled or deal-stage). Each row shows the agent's name, creator, total runs to date, and the next scheduled run.
  • On-Demand Agents — agents set to run manually.

For each agent, the kebab menu offers:

  • Edit — open it back in the builder form.
  • Duplicate — make a copy as a starting point for a new agent.
  • Delete — remove it (with confirmation).

Click an agent's name (or the play button on a manual agent) to see its history.

Agent history and continuing the conversation

The agent history page lists every run of a saved agent, with the date, the prompt that ran, and the assistant's response. Click any past run to open it in the chat surface — from there you can keep asking follow-up questions about that result without re-running the original schedule.

Outputs from scheduled or triggered runs are also delivered to whoever was set as the recipient (email, Slack DM, Slack channel). The history view is the in-app record.

Permissions

Anyone in your workspace can create, edit, delete, and run their own agents. An agent runs against the data the agent's owner has access to, so private calls or restricted accounts are respected per user.

Common patterns

  • Friday weekly digest — Scheduled, weekly on Friday afternoon, prompt = pipeline summary, output = email to your team.
  • Closed-won handoff — Deal Stage Change → Closed Won, prompt = CS handoff note using deal context, output = post to your CS Slack channel.
  • Daily at-risk deal alert — Scheduled, weekdays at 8am, individual mode, prompt = "Flag any of my open deals that haven't had activity in 14 days", output = Slack DM to each rep.
  • Quarterly win/loss — Manual, prompt from the win/loss template, run by RevOps when prepping the QBR.

Related articles