Back to the changelog

• Feature

  Create with AI: understand first, then build

  Describe the process and FlowSharp shows you what it understood (current situation and goal), asks you from zero to seven questions with suggested answers and tells you where each piece of information comes from. The Draft is created only after you confirm, then the AI builds its structure and detail. You can close the window and resume later from where you were, even after an error: a second Draft is never created. At the end a clear outcome tells apart what is only structure, what needs configuring and what is ready for review, with direct links to the activities to complete.

• Improvement

  Incomplete requests: questions instead of invented flows

  If the description is not enough, the AI asks for clarification instead of building a flow at random, even when generation is requested by an external assistant. Suggested answers do not count as confirmation, phases that need a person stay that way and no integration is tested during generation.

• Integration

  Your assistant can read the design contract and validate flows

  Five new tools in the MCP connector let an authorised external agent read the FlowSharp design contract, get a flow with its protected fields listed, learn which references can be used and validate a design before saving it, telling what can be saved apart from what can be published. They work without calling any AI model and without writing anything.

• Integration

  Preview and controlled application of designs

  An authorised external agent can first show a plan with a readable comparison of the changes and then apply it: it creates a flow, updates a Draft or prepares a new version of a published flow. Applying is atomic, does not duplicate if the request is repeated, rechecks permissions, policy and revision and always leaves the flow as a Draft: nothing is published and no case is started.

• Integration

  FlowSharp Designer plugin

  A privately distributed package of skills and references guides an MCP-compatible agent to design and edit flows with the FlowSharp design tools, with no second AI generation. Before writing it checks that the plugin's contract matches the server's and, if it differs, it does not write. A new guide in Integrations and API explains installation, connection and limits.

• Security

  You decide whether the policy leaves for external agents

  New "Share the policy with authorised external agents" switch in Settings, Agent Policy: off by default, restricted to administrators and recorded in the audit. While it is off, the external agent only sees that the policy exists, its revision and its fingerprint, not the text.

• Security

  Previews without credentials and policy respected at confirmation

  The comparison shown in a design preview never reveals credentials or other protected values, which are always masked. If your organization's policy changes while an AI design is under review, the confirmation is rejected and you are asked to review your answers with the updated policy.