Skip to main content

​
Overview

Dependencies let you inject runtime variables into your agent’s instructions and user messages using {template} syntax. Values can be static, synchronous functions, or async functions — all resolved at the start of each run.

​
Quick Start

import { Agent, openai } from "@radaros/core";

const agent = new Agent({
  name: "support-agent",
  model: openai("gpt-4o"),
  instructions: "You are a support agent for {company}. Today is {date}. The user's plan is {plan}.",
  dependencies: {
    company: "Acme Corp",
    date: () => new Date().toLocaleDateString(),
    plan: async () => {
      // fetch from database
      return "Enterprise";
    },
  },
});

const output = await agent.run("What features do I have access to?");

​
Per-Run Overrides

const output = await agent.run("Help me with billing", {
  dependencies: {
    plan: "Pro",       // override the default
    locale: "en-US",   // add a new dependency
  },
});

​
Accessing in Tools

Resolved dependencies are available in RunContext: 
import { defineTool } from "@radaros/core";

const lookupTool = defineTool({
  name: "lookup_account",
  description: "Look up the user's account",
  parameters: z.object({}),
  execute: async (args, ctx) => {
    const company = ctx.dependencies.company;
    const plan = ctx.dependencies.plan;
    return `Account: ${company}, Plan: ${plan}`;
  },
});

​
Template Syntax

  • {key} — replaced with the resolved value
  • Unknown keys are left as-is (e.g. {unknown} stays {unknown})
  • Templates work in both instructions and string input

​
Resolution Order

  1. Agent-level dependencies from config
  2. Per-run dependencies from RunOpts (overrides agent-level)
  3. All callable values are invoked and awaited
  4. Results are flattened to Record<string, string>