Skip to main content

Agents

A Pinksheep agent is a configured automation that connects your business tools, executes a series of steps, and produces a result. Every agent has:
  • A name describing what it does
  • An execution plan with ordered steps
  • Connected apps (OAuth-authenticated)
  • An autonomy level controlling how much it can do without your approval
  • A schedule (optional) for automatic execution

Stacks

Your stack is the set of business tools you use. When you select your stack during onboarding (or update it in Settings), Pinksheep:
  1. Scopes the builder to only discover tools relevant to your apps
  2. Filters the template library to show agents compatible with your stack
  3. Recommends templates during onboarding based on what you’ve connected
Stacks currently support 20+ popular integrations including HubSpot, Salesforce, Gmail, Slack, Google Sheets, Notion, Xero, Stripe, and more.

Autonomy levels

Every agent operates at one of three autonomy levels:
LevelBehaviorWhen to use
ObserveReads data and reports. Never writes.Monitoring, reporting, alerting
ProposePrepares write actions and shows them to you for approval before executing.CRM updates, email sends, accounting writes
ActExecutes read and write actions without requiring approval.Low-risk automations like posting Slack summaries
The builder automatically selects the appropriate level based on what your agent does. Agents that write to external systems default to propose.
Even in “act” mode, irreversible actions (deletes, sends, publishes) are flagged in the agent plan so you’re aware before deploying.

Credits

Agents consume credits per execution. The credit cost depends on the number of steps, the tools invoked, and the AI processing required.
  • Every agent plan shows an estimated credits range (min-max) per run
  • Scheduled agents show a projected monthly cost based on frequency
  • Credits are purchased in packs via Stripe
  • Your current balance is visible in the dashboard header
Credit pricing: each credit costs $0.25. A typical agent run costs 1-5 credits depending on complexity.

Execution

When an agent runs (manually or on schedule), the execution flow is:
  1. Pre-flight checks: Verify all OAuth connections are alive. If any are expired, the run is skipped and you’re notified.
  2. Step execution: Each step runs in order. Read steps execute immediately. Write steps in “propose” mode wait for approval.
  3. Result capture: The output of each step is logged. Success/failure status, credit cost, and duration are recorded.
  4. Notification: You’re notified based on your preferences (immediate for failures, digest for successes).

Failure handling

If a step fails:
  • The run halts at that step (remaining steps are skipped)
  • The failure reason and step log are recorded
  • You receive an immediate notification
  • The agent detail page shows a “Fix this agent” button that opens the builder with the failure context pre-loaded

Scheduling

Agents can run on a CRON schedule. The builder suggests an appropriate frequency based on your use case:
  • Daily for reports and digests
  • Hourly for monitors and alerts
  • Weekly for summaries and reconciliation
  • Custom CRON for specific timing needs
The builder will push back if you request an unnecessarily frequent schedule (e.g. checking every minute for data that changes daily).

Teams and Organisations

Pinksheep supports both solo users and organisations:
  • Solo: Everything is tied to your personal account
  • Organisation: Shared agent library, team members, and role-based access
Organisations are set up during onboarding or from Settings. Team members can view shared agents and their run history.

Templates

Templates are pre-built agent configurations that you can activate in a few clicks. Each template includes:
  • Pre-configured execution steps
  • Recommended apps and connections
  • A system prompt tuned for the use case
  • Eval scenarios for testing
Templates are filtered by your stack, so you only see agents you can actually use with your connected tools.