Workflow fundamentals
Understand how gospace runs agentic workflows in a workspace.
An agentic workflow is a repeatable run that turns inputs into outputs. In gospace, workflows live in a workspace and can be run manually, scheduled, or triggered by events.
Core concepts
- Workspace — The boundary for data, agents, integrations, permissions, and audit logs.
- Agent — The unit of work you run. Agents can be single-step or multi-step.
- Blueprint — A reusable template for creating agents and standardising settings.
- Run — One execution of an agent with its inputs, outputs, and timestamps.
- Inputs — Data sources, parameters, and workspace variables provided to a run.
- Outputs — Results written back to the Platform, delivered to integrations, or pushed via webhooks.
- Guardrails — Permissions, audit trails, and (optionally) approval gates for sensitive actions.
Execution lifecycle
- Validate — Confirm credentials, workspace scope, and payload shape.
- Resolve — Load data sources and workspace variables referenced by the agent.
- Run — Execute the workflow and capture structured logs.
- Record — Persist outputs and audit entries for traceability.
- Notify — Emit events to the Platform, notifications, or webhooks.
Common patterns
| Pattern | When to use | Typical outputs |
|---|---|---|
| Forecast → Recommend | Planning and decision support | Recommendations, dashboards |
| Forecast → Allocate | Resource allocation and seat planning | Allocations, change sets |
| Integrate → Act → Audit | Automations that touch production systems | Signed payloads, audit logs |
Autonomy levels
- Advisory — Runs produce recommendations for human review.
- Supervised — Automations run automatically but require approvals for sensitive actions.
- Automatic — Fully hands-off execution with audit-first guardrails.
Design workflows incrementally: start with advisory, add guardrails, then graduate to supervised or automatic execution as confidence grows.