The agency delivery model
Agencies use Pinksheep to build AI agents for clients without writing integration code. The no-code builder connects to the client's existing tools via OAuth, and the MCP Builder handles custom or niche tools outside the standard library.
The typical engagement looks like this: the agency scopes the agent with the client, confirms which tools and approvals are needed, builds and tests it in a staging workspace, then deploys to production and either hands over management or runs it as a managed service.
Scope and build
Identify the agent job, connect the client's stack via OAuth, and configure the instructions plus approval gates.
Test and approve
Run the agent in sandbox mode. Review proposed actions with the client before enabling live writes.
Deploy and hand over
Go live in production. Optionally transfer workspace management to the client or retain as a managed service.
Client use cases by business type
| Client type | Common agents | Client stack |
|---|---|---|
| SMB e-commerce (Shopify) | Order status, returns triage, review requests, support inbox | Shopify + Slack |
| Professional services firm | Invoice follow-up, AP coding, month-end close prep | QuickBooks |
| SaaS startup | Lead routing, CRM hygiene, pipeline updates, meeting summaries | Salesforce or HubSpot |
| Support-heavy business | Ticket triage, SLA breach alerts, reply drafts, escalation routing | Zendesk or Intercom |
| Operations-led company | Sprint planning, issue intake, release notes, standup summaries | Jira + Slack + Notion |
Setup and deployment steps
Step 1: Connect the client's stack
- Authorise via the client's OAuth credentials (they authenticate in the client's account, not yours)
- Review and narrow the requested scopes to minimum required
- Confirm the connected integration is live and returning data
Step 2: Configure the agent
- Describe the agent in plain language so it knows what job to handle
- Set approval gates on all write operations
- Configure who receives approval notifications (usually the client's team lead for that function)
Step 3: Test in sandbox mode
- Run the agent against real data with writes disabled
- Review proposed actions with the client
- Adjust instructions if the agent's proposals don't match expectations
Step 4: Go live
- Enable live writes
- Run once with the client watching the approval queue
- Hand over approval management to the client or retain in your agency account
Pricing and resale
Pinksheep pricing is usage-based, making it straightforward to build an agency margin on top. You pay for runs and seats, and you bill clients for the agent value delivered.
Common agency pricing models for AI agent services include a one-time setup fee per agent, a monthly retainer for ongoing management, or a service package tied to the business job the agent handles.
For agency accounts managing multiple client workspaces, contact the Pinksheep team for volume pricing.
Frequently asked questions
Can I resell Pinksheep agents under my own brand?
Yes. You build the agents, configure the setup, and deploy them in your client's connected stack. The agents can run under your agency account with your branding on the delivery.
Do my clients need their own Pinksheep accounts?
That depends on the deployment model. You can manage everything from your agency account for clients who want a fully managed service. For clients who want to manage their own agents after initial setup, you can transfer workspace ownership.
How long does it take to build an agent for a client?
A standard client agent can usually be configured, tested, and handed over quickly once the stack is connected and the approvals are clear. More complex multi-step agent setups take longer, especially when several tools or teams are involved.
What stacks does Pinksheep support?
Pinksheep supports 500+ integrations. The most common agency-client stacks are Salesforce, HubSpot, Zendesk, Slack, Notion, Jira, QuickBooks, and Shopify. All connect via managed OAuth with no code required.
Can I use Pinksheep's MCP Builder to build custom integrations for clients?
Yes. If a client uses a tool not in the standard integration library, the MCP Builder lets you create a custom server connection without writing integration code. This is particularly useful for industry-specific tools.