The Six-Step Build Sequence
1
Configure Connectors
Register the external systems you’ll need — from the catalogue or as custom MCP connectors. This is done once and reused broadly, so start here.See Connectors.
2
Build Skills
Compose connectors into single-purpose skills, each with a clear instruction and a defined input/output contract. Keep each skill focused on one job.See Skills.
3
Add Pipelines to skills that need data
Within a skill, define its pipeline — the source connectors, the transformation, the schedule, and the workstream binding. The data lands in the Data Store, ready for agents to read.See Pipelines.
4
Build Agents
Chain skills together with glue logic, wire in the Data Store and File Explorer reads through those skills, place human-in-the-loop gates at the review moments, and bind the agent to its workstreams.See Agents & Human-in-the-Loop.
5
Set up Workstreams
Create a workstream per client and add the people who should have access. Point pipelines and files at the right client systems.See Workstreams.
6
Run it in the App
End-users open the agent, pick the workstream, converse, and resolve any human-in-the-loop steps. Every run is logged under the workstream.See The End-User App.