Tools

Tools let an inference endpoint do more than generate text. When a model decides to call a tool, Infersec executes it and hands the result back to the model so it can continue reasoning. A tool might read a file from disk, query a database, call out to an MCP server, or even run a second inference pass against another endpoint.

Every tool in Infersec is one of two kinds:

Conduit-run (local) - the tool executes on your own hardware, through a Conduit agent you control. Credentials and the underlying resource (files, the database) are only ever accessed from your host; the tool-call arguments and results are relayed through the Infersec API so the model can use them.
Server-side - the tool is executed entirely by the Infersec API. No Conduit connection is required.

Tools vs tool services

A tool is a single capability (one filesystem, one database connection, one MCP server). A tool service is a named bundle of one or more tools that you attach to an inference endpoint. When the endpoint receives a request, the tools from every attached (and enabled) tool service are offered to the model.

   Inference endpoint
        │
        ├── Tool service "Code tools" (enabled)
        │     ├── Local Filesystem tool
        │     └── Local MySQL tool
        │
        └── Tool service "Vision" (enabled)
              └── Model Endpoint Call tool

This separation lets you reuse the same tool across endpoints, and toggle whole groups of tools on or off per endpoint without redefining them.

Conduit-run tools

Local tools run inside a Conduit tool connect process on a machine you control. The model's tool calls are streamed from the API to your Conduit agent over a secure channel, executed locally, and the results relayed back through the API. Because the tool's credentials and its underlying resource (files, the database) stay on your host - only the tool-call arguments and results transit the API - local tools are the right choice for anything that touches private files or databases.

Available conduit-run tools:

Tool	What it exposes
Local Filesystem	Read and write files within a base directory
Local MySQL	Query MySQL/MariaDB databases (read-only by default)

Server-side tools

Server-side tools need no Conduit agent. The Infersec API resolves the tool's definition, executes it, and returns the result inline. Use these when the capability already lives at a URL, or when the "tool" is really another inference call.

Available server-side tools:

Tool	What it does
MCP Proxy	Forwards tool calls to any external MCP server
Model Endpoint Call	Runs a structured inference pass against an inference endpoint
Model Source Call	Runs a structured inference pass against an inference source

Creating a tool

Tools are created in the console under Inferencing - Tools. Pick a type, give it a name, and (for server-side tools) fill in the type-specific configuration. Once a tool exists, add it to a tool service, then attach that tool service to one or more inference endpoints.

For conduit-run tools, creation only registers the tool - you then connect it from your machine with npx @infersec/conduit tool connect. Each tool's sub-page documents the exact connection command and configuration it accepts.