S Selfint
PT EN
AI

Selfint already supports AI and it can be used in several ways.

AI in Selfint is not just a future idea. There are concrete components to help design workflows, run AI steps inside a process and track cost, latency and output.

AI Builder

Suggests next steps and can compose an initial workflow plan from a written objective.

AI inside workflows

AI can read input, interpret content, generate structured JSON and map the result into the next workflow steps.

Controlled operation

Teams can manage AI connections, version prompts, run preview tests and track logs, token usage, cost and latency.

Simple examples

  • Extract key fields from emails, free text or complex payloads.
  • Classify requests by type, priority or intent.
  • Summarise content before sending it to another team or system.
  • Generate structured JSON so the workflow can continue automatically.
  • Route to human review when confidence is too low.

What already exists in the AI editor

  • Provider and connection selection.
  • Model, system prompt, user prompt and variables.
  • Input and output mapping.
  • Preview before saving.
  • Guard rails such as minimum confidence and fallback actions.
  • Privacy options controlling what is stored.

What the technical team can configure

These capabilities are already available on the platform for ongoing configuration and operation.

AI connections

Reusable provider references per team, with base URL, default model and API credential binding.

Versioned prompts

Prompts can be saved with name, version, state and variable schema for consistent reuse.

Preview and testing

Teams can send sample input before activating a step to validate behaviour and output.

Logs and statistics

Filter AI executions by client, workflow, step, provider, connection or model — with visibility into tokens, cost and latency.

MCP Gateway

JSON-RPC 2.0 endpoint compatible with Claude Desktop, Cursor and any MCP client. Each organisation gets a dedicated API key (mcp_…), and available tools can be listed via GET /api/mcp/tools/list. Requires a plan with MCP enabled.

Slack notifications

Each client can configure a Slack webhook to receive automatic alerts for errors, pending approvals and file submissions. The webhook is masked after saving and can be tested directly from the interface. Requires a plan with Slack enabled.