With tools enabled, a prompt can look things up or act while it answers instead of relying on what the model already knows. Evaligo runs the loop: the model asks for a tool, Evaligo executes it, the result goes back to the model, and this repeats until the model gives its final answer or the step limit is reached.
Where tools can be enabled
- A prompt in the Playground. The Tools panel sits beside the prompt editor.
- A prompt node in a flow. Same panel, in the node’s settings. The configuration is saved with the flow and travels with it when the flow is exported or deployed.
- A judge in an experiment. A judge can use the same tools, so it can check an output against the live web before scoring it.
Settings
- 1
Enable tools One switch per prompt. Off by default.
- 2
Pick the tools Each tool in the list has its own switch. Only enabled tools are offered to the model.
- 3
Tool choice Auto lets the model decide whether to call a tool. Required forces at least one call. None keeps the tools listed but never called.
- 4
Max iterations How many tool rounds a single run may take. Default 10, maximum 20. When the limit is reached the model is asked for its final answer.
Available tools
| Tool | What it does |
|---|---|
| Web Search (AI-native) | The model’s own web search, run by the provider: OpenAI web search, Anthropic web search, or Gemini Google-Search grounding. Cited source URLs come back on the node’s citations output. Needs a model that supports its provider’s web search. |
| Google Search | Runs a Google search from Evaligo and returns the results to the model. |
| Crawler | Fetches a page and, optionally, the pages it links to. Parameters: url, max_depth (0–2, default 0), max_pages (1–50, default 10), same_site_only. |
| Fetch HTML | Fetches a URL and returns the page content. |
| Get Page HTML | Fetches a URL and returns its raw HTML source. |
| HTTP Request | Calls any HTTP endpoint with the method, headers and body the model supplies. |
| SMTP Email | Sends an email through an SMTP connection you have set up under Integrations. Only offered when an SMTP connection exists. |
| Search Nodes | Looks up Evaligo’s own node types and their documentation. Used by the flow-building assistants. |
| Search Templates, Get Template Flows, Get Template JSON | Look up deployed templates and read their flow structure or full export. Also used by the assistants. |
Every tool except the native web search is executed by Evaligo, so it works with any model that supports function calling. The native web search is executed by the model provider, which is why it needs a model with that capability.
What gets recorded
- Each round of the loop is stored with the run: which tool was called, with what arguments, what came back, and how long it took. In a flow, the node’s step trace shows the rounds one by one.
- Cost and latency for the whole loop are recorded on the run, so a prompt with tools can be compared against one without in an experiment.
- For the native web search, the source URLs the model cited are available as the node’s
citationsoutput and can be mapped into the next node.
On Anthropic models, runs with more than one tool round use prompt caching on the earlier rounds. In a measured run of the Reddit-post agent this removed all cache-write tokens from the generate step and cut its cost by 16%.