Skip to content

Agent Hub

Build a custom agent

Describe what you want in plain language. Muntri wires the data, schedule, outputs and guardrails — then you test it and switch it on.

Browse the Marketplace How autonomy works

Describe it, don't script it

You don't write code or queries. Tell the builder what you want the agent to do in plain English — "watch my enterprise accounts and flag any where usage drops two weeks running" — and it assembles a working agent. From there you refine the details: which data it reads, when it runs, where it reports, and what it may do on its own.

The build flow

Describe the job. Write, in plain language, what the agent should watch for and produce. Give it a name and an avatar so it fits your fleet.

Choose capabilities & data sources. Pick the parts of your record it should analyze — pipeline, accounts, revenue, forecasts, team activity, and more. The agent reads real tenant data with no hand-written queries.

Set a schedule. Decide whether it runs on demand or on a recurring cadence.

Choose outputs. Route results to email and/or a Slack channel, and set the recipients who should see them.

Set guardrails & autonomy. Define the boundaries it operates within, and set each action to run automatically, require approval, or stay advisory.

Test, then activate. Run it once, review the output, adjust, and switch it on when you're happy.

Configuration at a glance

Setting What it controls Examples
Name & avatar The agent's identity across the app "Pipeline Sentinel", a portrait from the avatar library
Brief What the agent does, in plain language "Summarize new deal risks each morning"
Data sources Which parts of your record it reads Pipeline · accounts · revenue · forecasts · team · activity
Schedule When and how often it runs On demand · daily · weekly
Outputs Where results go and to whom Email, a Slack channel, chosen recipients
Guardrails The boundaries it must operate within What it may touch; what it must never do
Autonomy Per-action: auto, require approval, or advisory See running agents

Wiring the data

flowchart LR
  B[Plain-language brief] --> D[Data sources] --> R[Agent reasons<br/>over live data]
  R --> O[Outputs:<br/>email / Slack]
  classDef n fill:#ffffff,stroke:#2563eb,stroke-width:1.5px,color:#0B1024;
  class B,D,R,O n;

Pick the sources that matter

Selecting a data source tells the agent which slice of your record to analyze — for example pipeline, accounts, revenue, forecasts, team or activity. The agent works over your real data automatically; you don't write any queries.

Outputs & delivery

Custom agents can report where your team already works:

  • Email

    Send the agent's summary to chosen recipients on its schedule.

  • Slack

    Post results to a channel so the whole team sees them. (Connect Slack once in settings.)

Test before you trust

Run the agent on demand and read its output before you let it run on a schedule or take actions on its own. Tune the brief, sources and guardrails, then set your autonomy tiers and activate. You can always start every action as advisory and promote it once you've seen a few good results.

Next