Browse the docs
AI opt-in per step
How AI steps work, when to use one, and why they stay contained.
In Destaris, AI is a node type you opt into for a specific step — not the substrate the whole workflow runs on. You add an agent exactly where you need a decision or a draft, and the rest of the workflow stays deterministic.
When to reach for an AI step
Use one when a task needs judgment that's hard to express as plain rules:
- Classify or route something ("is this urgent?", "which team owns this?").
- Summarize or draft text from messy input.
- Decide which of a fixed set of actions to take.
Keep everything else — fetching, filtering, branching, formatting, sending — deterministic. That keeps your workflow fast, free, and predictable, with the model involved only where it earns its place.
How an AI step runs
An AI step runs as your own already-authenticated CLI — Claude Code is the default first pick — on your own plan. Destaris orchestrates the step; it never holds your keys and never resells tokens. The work happens locally, as you.
You can keep a model's output structured: ask it to return a typed decision (for
example { action: "reply" | "escalate" | "skip" }) and then branch on that with a
deterministic step. The agent makes the call; deterministic nodes carry it out.
Treat any content a workflow pulls in (comments, emails, web pages) as untrusted data, never as instructions to the model. The deterministic graph around the agent is what decides what actually happens.
Models and cost
Model choice isn't gated — pick the vendor you're signed into. Because the tokens are yours, on your own subscription, AI steps don't add a Destaris charge; only the steps you mark as AI spend anything at all.