AI Operators

Zenvara treats an LLM call the same way it treats a database query: a typed step with declared inputs and outputs, logged prompt and response, composing with retries, secrets, and audit. These operators are native — not plugins or wrappers.

Operators

Operator	invoke:	Actions	What it does
Claude	claude.*	call	Anthropic Claude — text generation and tool/function calling.
ChatGPT	chatgpt.*	call	OpenAI GPT models.
Gemini	gemini.*	call	Google Gemini models.
Claude CLI	claude-cli.*	invoke, start-session	Run Claude Code autonomously inside a flow, with quota management and human-assist.
Web Search	web-search.*	search	DuckDuckGo, Google, StackOverflow, and Ollama-backed search providers.

Claude — example

using:
  - zenvara/claude

steps:
  - $summary:
      invoke: claude.call
      with:
        SystemPrompt: "You are a concise analyst."
        Prompt: "Summarise these orders in two paragraphs:\n${orders.rows}"

Outputs: content (generated text), stopReason, inputTokens, outputTokens, toolCalls (when function-calling). Reference them as ${summary.content}, ${summary.outputTokens}.

Environment config

The model, key, and limits live on the environment, not in the flow — so the same flow can target a different model per environment:

Claude:
  ApiKey: !secret              # Anthropic API key (referenced as a secret)
  Model: "claude-sonnet-4-20250514"
  MaxTokens: 4096
  TimeoutSeconds: 120
  EnableCaching: true          # emit cache_control breakpoints for cheaper multi-turn loops

ChatGPT and Gemini follow the same shape under their own environment keys.

Tool / function calling

The Claude operator supports structured content blocks (text, tool_use, tool_result) and a Tools list of tool definitions (Name, Description, InputSchema). This lets a flow run a full agentic tool loop where each tool result is fed back as a typed message — the model’s toolCalls output drives the next step.

Treating AI output as data

Because the response is a typed value, the next step consumes it like any other:

  - $tickets:
      invoke: jira.search
      with:
        Jql: "${summary.content}"     # AI output → next operator's input

That composition — AI output flowing into a database query, an HTTP call, or a filter: — is the point of native AI operators. See Authoring Flows for how values flow between steps.