Skip to content
Trakli Docs

Overview

The agent is how you operate Trakli by talking instead of filling in forms. You ask a question or give an instruction, and the agent uses a small set of tools to read your data, make changes you approve, and answer in plain language.

How it works

Section titled “How it works”

The agent is a language model wired to tools. A tool is a named, typed action the model can call: read these transactions, create that one, fetch the current date. The model decides which tools to call to satisfy your request, calls them, reads the results, and replies.

This matters because the model never touches your database directly. It can only do what its tools allow, and every tool runs under your identity. The tools are the boundary.

Trakli’s agent is built on eloquent-agents, an open package that defines:

  • Tools, each with a name, a description, typed parameters, and a permission level (read, write, or external).
  • A harness, which is the agent configuration: its system prompt, which tools it may use, and which permission levels are allowed.
  • A tool context, which carries the acting user. Tools scope their work to this user rather than trusting any identifier the model supplies.

The shape of a turn

Section titled “The shape of a turn”
  1. You send a message.
  2. The harness builds the agent with its allowed tools and a context carrying your user.
  3. The model calls tools as needed. Before each call runs, the tool’s authorization check decides whether it is allowed for you.
  4. The model reads the tool results and writes a reply.

If a tool call is not permitted, the agent is told the action is not allowed and continues without it. It does not silently do something you did not authorize.

Section titled “What to read next”