@any-model/agent

A small, provider-independent model/tool loop for @any-model/core.

bun add @any-model/agent @any-model/core

import { runAgent } from "@any-model/agent";
import { tool } from "@any-model/core";

const weather = tool({
    name: "weather",
    inputSchema: {
        type: "object",
        properties: { city: { type: "string" } },
        required: ["city"],
    },
    execute: (args) => ({ city: (args as { city: string }).city, conditions: "sunny" }),
});

const result = await runAgent({
    model,
    messages: [{ role: "user", content: "What is the weather in Paris?" }],
    tools: [weather],
    maxSteps: 8,
});

console.log(result.text);

To render output as it is generated, handle normalized stream parts. The callback runs for every model step, including steps that produce tool calls, and runAgent() still resolves to the complete folded result:

const result = await runAgent({
    model,
    messages: [{ role: "user", content: "Research this and summarize it." }],
    tools,
    onStreamPart({ stepNumber, part }) {
        if (part.type === "text-delta") process.stdout.write(part.text);
    },
});

runAgent() owns only one invocation. It does not retain conversation state between calls. It returns the complete transcript, each model/tool step, accumulated usage, and a normalized stop reason.

Extensions may provide stateless hooks directly. Stateful extensions should create isolated hooks for every run:

const extension = {
    createRun() {
        let toolCalls = 0;
        return {
            beforeToolCall() {
                toolCalls++;
                return toolCalls > 10
                    ? { action: "deny" as const, reason: "Tool-call limit reached" }
                    : { action: "allow" as const };
            },
        };
    },
};