Skip to main content
The Agent Foundry is where you create and run AI agents. An agent is a tireless digital worker you configure once — you tell it what job to do, which data it can touch, which model powers its thinking, when it should wake up, and what guardrails it must respect.

The agent listing

When you open Agent Foundry you land on a grid of every agent your departments can see, grouped into sections by what each one is currently doing.
Control / ButtonWhat it does
Search agentsFilters the grid by agent name
Departments filterNarrows the grid to selected departments
Statuses filterNarrows by state: Running, Evolving, Paused, Idle, Error
New Agent buttonOpens the agent creation wizard (or press “n”)
Running sectionAgents actively working or evolving
Paused sectionAgents you have stopped
Idle sectionAgents set up but not currently doing anything
Errored sectionAgents that hit a problem on their last run
Agent cardClick to open Overview; shows status, description, quick stats
Card ”…” menu — OpenGoes to the agent’s Overview
Card ”…” menu — Pause / ResumeStops or restarts automatic activity
Card ”…” menu — RestartRestarts the agent
Card ”…” menu — EditOpens the agent for editing
Card ”…” menu — DeleteMoves to Trash; confirms first

The agent creation wizard

Eight steps. Nothing is created until you press Create agent on the final step.

Step 1 — Identity

Control / FieldWhat it does
Name (required)The agent’s display name
DescriptionA short summary
Departments (required)The teams that can see and use this agent
OwnerWho controls sharing, deletion, transfer; defaults to you

Step 2 — Model & modes

Control / FieldWhat it does
ReAct toggleChat with tools (on by default)
Quick toggleSub-2-second answers with no tools
Deep toggleMulti-agent teamwork for research
Graph toggleBranching, parallel, human-in-the-loop
Persistent toggleLong-running background agent
Model selector (per enabled mode)Picks the AI model; only org-approved models appear
Initial system promptOptional standing instructions

Step 3 — Tools

Control / FieldWhat it does
Tool treeTick the individual tools the agent may use
Selected countShows how many tools you picked

Step 4 — Data sources

Control / FieldWhat it does
Data source checkboxGrants the agent access
Access levelRead only, or Read & write
Daily query limitCaps queries per day (0 = unlimited)

Step 5 — Triggers

Control / FieldWhat it does
Add trigger menuAdds a new trigger
Schedule (cron)Recurring timetable in a timezone
Data change (CDC)Runs when chosen sources/tables change
WebhookOutside system triggers the agent; IP restrictions allowed
ManualA person fires the agent on demand
Enabled switchTurns a trigger on or off
RemoveDeletes a trigger

Step 6 — Security

Control / FieldWhat it does
Role nameLabel for the access role
Permission levelRead-only SELECT, Read & write, or Admin
PII accessMasked, Hashed, or Clear
PII mask fieldsSpecific field names to mask
Row limit per queryCaps rows per query
Rate limitCaps queries per hour
Cost guardrailDaily spend ceiling
Approval thresholdActions costing more require an approver
Flag-confidence thresholdOutputs below this confidence are flagged
Data retentionHow long generated artefacts are kept
Query validationPre-checks every generated query

Step 7 — Notifications

Control / FieldWhat it does
Add bindingNew notification connection
ChannelIn-app, Email, Slack, or PagerDuty
DestinationTarget for that channel
Credential referenceRequired for Slack/PagerDuty
EventsWhich events trigger a message
Enabled switchTurns binding on or off
RemoveDeletes the binding

Step 8 — Review

Read-only summary. Press Create agent to save everything in one atomic commit.

Who can use it

  • Organization members can create agents.
  • The departments you select determine who can see and use the finished agent.
  • Approved tools and models are decided upstream by your Org Admin in Settings → AI Capabilities.