Skip to main content

SaaS Edition Mapping Reference

Project in the Marketplace edition works on the same ideas as the SaaS edition. This page lists how to read each page and section of the SaaS edition in the Marketplace edition.

  • As is — the same in the Marketplace edition. Read the SaaS edition's page
  • Read differently — part of it differs. Shows what differs and where to read in the Marketplace edition
  • Does not apply — the Marketplace edition does not have it

Overview​

Where in the SaaS editionStatusIn the Marketplace edition
Project — How a project is builtRead differentlyWhat you place on steps are registered agents that an administrator registered. No goal (instructions), guardrails, playbook or skills are attached to the agent. Claw does not appear (Agent Guide)
Same — The two typesRead differentlyThe default type is workflow. A workflow is serial only and does not divide into lanes. It cannot be started by email
Same — Where things are in the UIRead differentlyThe editor's palette is the list of registered agents. Each workflow run has a case screen. The external API is the Marketplace edition's API

How it works​

Where in the SaaS editionStatusIn the Marketplace edition
Projects and steps — Steps and review steps; Agents and steps are not one to oneRead differentlyThe output of the previous step does not pass to the next step automatically. Bind the outputs you want passed under “Input bindings” in the step settings (Agent Guide)
Same — What happens when several agents share a stepRead differentlyIn a workflow, there is no “Run at the same time”; agents run one at a time, in placement order. In a cycle, placements run without waiting for one another, on their own schedules or by webhook
Same — Playbooks and skillsDoes not applyHow the agent behaves is settled by the container image and the contract
Same — Project memoryDoes not applyThere is neither project memory nor company memory. What is handed over between steps is files (Case Guide)
Cycle and workflowRead differentlyA cycle's placement starts a registered agent used in chat on a schedule or by webhook. The only text the agent receives as instructions is a fixed introductory sentence and the webhook's contents (Agent Guide)
Who decides whatRead differentlyStarting a case, editing its settings and artifacts, cancelling it, and stopping a cycle execution are for owners only. Being named as a reviewer does not let you change the case settings. There is no external API called with a person's token
Human review — Automatic processing at the deadlineRead differentlyThe deadline does not decide. Instead there is the “external wait” (Review Guide)
How much to delegateRead differentlyThe execution rules are used to check, before starting, whether the agent supports them. The platform does not stop operations during the run (Execution Rules and Traffic Guide)
Records and metricsRead differentlyCycle executions have “Run ended (completion not measured)”, which is not part of the denominator of the metrics (Record Guide)

Quickstart​

Where in the SaaS editionStatusIn the Marketplace edition
WorkflowDoes not applyThere are no steps for creating agents. Read the Marketplace edition's Workflow
CycleDoes not applyRead the Marketplace edition's Cycle

Guides​

Where in the SaaS editionStatusIn the Marketplace edition
Design Guide — What to settle for the work, how to divide projectsAs is
Same — What to give an agentDoes not applyAgent Guide
Same — Dividing in a workflowRead differentlyThere are no lanes
Same — Placing agents on stepsRead differentlyIn a workflow, there is no “Run at the same time”; agents run one at a time, in placement order. In a cycle, placements run without waiting for one another, on their own schedules or by webhook. In a workflow, the output of the previous step does not pass to the next step automatically. Bind the outputs you want passed under “Input bindings” in the step settings (Agent Guide)
Same — What to do when it does not go to planDoes not applyThere is no “Handling of results”; it follows fixed rules (Trigger Guide)
Same — Letting the AI draft it for youDoes not apply
Same — Checking a draft and applying itRead differentlyWhen you place a registered agent on a step for the first time, it is saved on the spot (it does not go into the draft)
Execution Rules GuideRead differentlyExecution Rules and Traffic Guide
Review Guide — Who decidesAs is
Same — Waiting, or proceeding at the deadline / What happens once the deadline passesDoes not applyReview Guide
Same — Destination optionsDoes not apply
Trigger Guide — Starting a workflow, a time per step, giving the timeRead differentlyFrom the screen, start with “Start case”. Parameters cannot be passed from the external API or time triggers (Trigger Guide)
Same — Running a cycle, starting it from outsideRead differentlyOverlapping and late sweeps, and how webhooks are handled, differ (Trigger Guide)
Same — Starting by emailDoes not apply
Same — Stopping and resumingRead differentlyThere is cancelling a case and stopping a cycle execution
Connection GuideRead differentlyConnectors are assigned to the contract's connection slots in the step settings. In a cycle, the tokens of the project's connectors (enabled ones) pass to the agent (Agent Guide). The external-wait webhook secret is added
Record Guide — Where to look, reading a single runRead differentlyRecord Guide
Same — Revisiting it from the track record (AI improvement suggestions)Does not applyThere is no consulting on improvements or carrying them out

References​

Where in the SaaS editionStatusIn the Marketplace edition
Settings ReferenceRead differentlySettings Reference
Run Results ReferenceRead differentlyThe seven words for results are the same. Cycles add “Run ended (completion not measured)” (Record Guide)
Records ReferenceRead differentlyThere are no records for improvement suggestions, email or lanes. “Case Config Changed” is added
GlossaryRead differentlyGlossary
Roles ReferenceRead differentlySee the “Who decides what” row above