Skip to main content

Agents & environments

Agents are project-scoped definitions used by chat (agentDefinitionId), event automations, headless Kubernetes jobs, and scheduled AgentRun targets.

Open Studio → Agents when the agents feature is enabled.

Config agents (chat & automations)​

  1. Create an agent: name, system prompt, temperature, optional provider, MCP server IDs / account names.
  2. Keep it active.
  3. Add tool grants for Pipedream or MCP tools (resource patterns, max calls, approval required).
  4. Reference the agent from an automation or from chat.

Member-facing create/edit does not set repository URLs or setup/build/test commands — those belong to admin development agents.

Environments (Admin)​

Environment profiles control workspace image, CPU/memory, storage, and egress for interactive sessions and related runtimes. Ensure the shared universal profile exists before users start sessions.

Development agents & runs​

Admins can seed a development template agent (repository + setup/build/test) and queue headless runs as Kubernetes Jobs. Monitor runs from the Agents / runs UI: events stream over SSE; cancel, retry, or supply input when the run waits.

Tips​

  • Start with read-only or draft behaviour before granting send/write tools.
  • Prefer grants with requiresApproval for external communication.
  • Automations must point at an existing agent — create the agent first.