v1.2 — Google Gemini + Ollama

The jQuery of
AI Agents

Drop console.agent(...) anywhere in your code to execute agentic workflows — as easy as console.log()

terminal

$ ▊

+ @console-agent/agent@1.0.0

$ ▊

[AGENT] ✓ 🛡️ Security audit Complete

[AGENT] ├─ ✓ SQL injection detected in user input

[AGENT] ├─ Tool: google_search

[AGENT] ├─ fix: Use parameterized queries

[AGENT] └─ confidence: 0.94 | 247ms | 156 tokens | model: gemini-2.5-flash-lite

───── verbose: true ↑ shows full trace │ default ↓ shows just the answer ─────

SQL injection detected in user input

fix: Use parameterized queries

severity: critical

Get Started GitHub Playground

0 Line to install

0 Config required

0 AI Personas

0 Built-in Tools

See the Difference

From guesswork to answers. One line changes everything.

❌ Before — console.log hell

your-app.js

console.log("error here:", err);

console.log("data:", JSON.stringify(data));

console.log("hmm", typeof result);

console.log("WHY IS THIS NULL???");

console.log("trying fix #47...");

console.log(data?.users?.map);

// 45 minutes later...

console.log("FOUND IT maybe?");

⏱️ 45 min of guesswork. No confidence.

❌ Before — print() debugging hell

your_app.py

print("error here:", err)

print("data:", json.dumps(data))

print("hmm", type(result))

print("WHY IS THIS NONE???")

print("trying fix #47...")

print(getattr(data, 'users', None))

# 45 minutes later...

print("FOUND IT maybe?")

⏱️ 45 min of guesswork. No confidence.

✅ After — console.agent

your-app.js

console.agent.debug("why is this null?", err);

[AGENT] 🐛 Analyzing error...

[AGENT] ├─ Tool: code_execution → Tested fix

[AGENT] └─ ✓ ROOT CAUSE IDENTIFIED

Likely cause: Missing await on async fn

Suggested fix: Add 'await' on line 47

Confidence: 0.92 | 312ms | 189 tokens

⚡ 1 line. 312ms. Root cause + fix.

✅ After — agent()

your_app.py

agent.debug("why is this None?", context=err)

[AGENT] 🐛 Analyzing error...

[AGENT] ├─ Tool: code_execution → Tested fix

[AGENT] └─ ✓ ROOT CAUSE IDENTIFIED

Likely cause: Missing 'await' on async fn

Suggested fix: Add 'await' on line 47

Confidence: 0.92 | 312ms | 189 tokens

⚡ 1 line. 312ms. Root cause + fix.

Why console.agent?

😤

Agents are too complicated

Langchain requires 100+ lines of boilerplate. We need 1 line.

🔧

Wrong abstraction layer

Existing tools are for chat apps, not runtime utilities.

⚡

console.agent is the jQuery of agents

Simple API, powerful capabilities. Zero config works.

Quick Start

Get running in under 30 seconds. No config needed.

Install

npm install @console-agent/agent

Set API Key

export GEMINI_API_KEY="your-key-here"

Free key at aistudio.google.com/apikey

Use it

import '@console-agent/agent';

// Fire-and-forget — never blocks your app
console.agent("analyze this error", error);

// Blocking — await for structured results
const result = await console.agent("validate email", email);

// Persona shortcuts
console.agent.security("check for SQL injection", input);
console.agent.debug("why is this slow?", { duration, query });
console.agent.architect("review API design", endpoint);

AI Coding Skill — Teach Cline / Cursor how to use console.agent

How It Works

📝Parse prompt + context

→

🎭Auto-detect persona

→

🔒Anonymize secrets

→

💰Check budget

→

🤖Gemini ToolLoopAgent

→

📊Structured AgentResult

Test Suite Superpowers

AI-powered debugging, security auditing, and root cause analysis — right inside your tests.

console.agent

The jQuery of AI Agents

AI debugging inside your test suite

Before 45 minutes of console.logs

→

After 2.6 seconds to root cause

Real tests. Real output.

CODE

debug.test.js

it("debug failing test", async () => {
  const result =
    await processPayment(input);

  try {
    expect(result.success)
      .to.equal(false);
  } catch (error) {
    const diagnosis =
      await console.agent.debug(
        "why did test fail?",
        { input, result, error }
      );
    console.log(diagnosis);
  }
});

OUTPUT

test results

✔ surfaces likely root cause

on failing payment test (2669ms)

Payment test failed because

expected failure condition

was not met due to

asynchronous token handling.

⏱️ 2.6 seconds

🎯 Root cause identified

Before: 45 min of debugging

After: One line of code

CODE

security.test.js

it("detect NoSQL injection", () => {
  const payload = {
    raw: '{"$ne": null}'
  };

  const audit =
    await console.agent.security(
      "audit this input",
      { payload, returnedCount }
    );

  expect(audit.summary)
    .to.match(/injection/);
});

OUTPUT

test results

✔ flags risky payload in

unsafe search path (3544ms)

⚠️ CRITICAL: NoSQL Injection

vulnerability detected

📍 Location: payload.raw

('$ne': null)

🔧 Fix: Sanitize inputs,

use parameterized queries

✅ Confidence: 95%

Real CVE detection in tests

CODE

mongo.test.js

it("debug mongo query", async () => {
  const users =
    await findUsers({ role: "admin" });

  try {
    expect(users.length)
      .to.equal(2);
  } catch (error) {
    const diagnosis =
      await console.agent.debug(
        "why is assertion failing?",
        { filter, users, error }
      );
    console.log(diagnosis);
  }
});

OUTPUT

test results

✔ debugs a failing mongo

expectation with context (2403ms)

MongoDB assertion failed due

to a mismatch between

expected and actual document

counts.

The query returned 1 document,

but 2 were expected.

The filter criteria matched

one document based on

role: 'admin'.

⏱️ 2.4 seconds to diagnosis

🎯 Context-aware analysis

Real Test Results

✅ 7 passing (12s)

✅ 3 AI-powered debugging assertions

✅ 2 security vulnerability detections

✅ Zero manual investigation time

Works with:

Mocha Vitest Jest Pytest Security auditing MongoDB queries Failed assertions

npm install @console-agent/agent

📖 console-agent.github.io ⭐ github.com/console-agent/agent

API Reference

`console.agent(prompt, context?, options?)`

The main API. Call it like console.log().

// Simple — fire-and-forget
console.agent("explain this error", error);

// Await structured results
const result = await console.agent("analyze", data, {
  persona: 'security',
  model: 'gemini-3-flash-preview',
  thinking: { level: 'high', includeThoughts: true },
});

console.log(result.success);    // boolean
console.log(result.summary);    // "SQL injection detected..."
console.log(result.confidence); // 0.94
console.log(result.data);       // { vulnerability: "...", fix: "..." }

Return Type: `AgentResult`

interface AgentResult {
  success: boolean;           // Overall task success
  summary: string;            // Human-readable conclusion
  reasoning?: string;         // Agent's thought process
  data: Record<string, any>;  // Structured findings
  actions: string[];          // Tools used / steps taken
  confidence: number;         // 0-1 confidence score
  metadata: {
    model: string;
    tokensUsed: number;
    latencyMs: number;
    toolCalls: ToolCall[];
    cached: boolean;
  };
}

AI Personas

Auto-detected from your prompt keywords, or set explicitly.

🛡️

Security

OWASP security expert

Detects: injection, xss, vulnerability, csrf, exploit

console.agent.security("audit this query", sql);

🐛

Debugger

Senior debugging expert

Detects: slow, error, optimize, crash, memory leak

console.agent.debug("why is this slow?", metrics);

🏗️

Architect

Principal engineer

Detects: design, architecture, schema, scalability

console.agent.architect("review API", endpoint);

🔍

General

Full-stack senior engineer

Default fallback — handles everything else

console.agent("validate this data", records);

Built-in Tools

Opt-in tools that give your agent real-world capabilities. Only activated when you explicitly pass tools: [...].

🔍

Google Search

Real-time web grounding

Search the web for up-to-date information, facts, CVEs, docs

const result = await console.agent(
  "What is the current population of Tokyo?",
  null,
  { tools: ['google_search'] }
);

💻

Code Execution

Python sandbox (Gemini-hosted)

Run calculations, data processing, algorithm verification

const result = await console.agent(
  "Calculate the 20th Fibonacci number",
  null,
  { tools: ['code_execution'] }
);
// result.data.result → 6765

🌐

URL Context

Fetch & analyze web pages

Read documentation, analyze APIs, extract page content

const result = await console.agent(
  "Summarize this page",
  null,
  { tools: ['url_context'] }
);

Combining Multiple Tools

Tools can be combined — the agent decides which to use based on the prompt.

// Search + calculate in one call
const result = await console.agent(
  "Search for the current world population, then calculate 1% of it",
  null,
  { tools: ['google_search', 'code_execution'] }
);

// The agent will:
// 1. Use google_search to find current population
// 2. Use code_execution to calculate 1%
// 3. Return combined structured result

Configuration

Zero config works. Customize when you need to.

Full Config

import { init } from '@console-agent/agent';
import { z } from 'zod';

init({
  apiKey: process.env.GEMINI_API_KEY,
  model: 'gemini-2.5-flash-lite',
  persona: 'general',
  mode: 'fire-and-forget',
  timeout: 10000,

  budget: {
    maxCallsPerDay: 100,
    maxTokensPerCall: 8000,
    costCapDaily: 1.00,  // USD
  },

  anonymize: true,             // Auto-strip secrets/PII
  localOnly: false,            // Disable cloud tools
  dryRun: false,               // Log without calling API
  logLevel: 'info',
  includeCallerSource: true,   // Auto-read source file
});

// ── Structured Output ────────────────────────

// Option A: Zod schema → typed, validated
const result = await console.agent(
  "analyze sentiment", review,
  {
    schema: z.object({
      sentiment: z.enum(["positive","negative","neutral"]),
      score: z.number(),
      keywords: z.array(z.string()),
    }),
  }
);
result.data.sentiment; // "positive" ✅ typed

// ── File Attachments ─────────────────────────

import { readFileSync } from 'fs';

const doc = await console.agent(
  "What does this document say?", null,
  {
    files: [{
      data: readFileSync('./data/report.pdf'),
      mediaType: 'application/pdf',
      fileName: 'report.pdf',
    }],
  }
);

🔒 Privacy & Anonymization

Auto-strips API keys, emails, IPs, AWS keys, private keys, and connection strings before sending to AI.

💰 Budget Controls

Hard daily limits on calls (100/day), tokens (8K/call), and cost ($1.00/day). Prevents cost explosion.

⏱️ Rate Limiting

Token bucket algorithm spreads calls evenly across 24 hours. Graceful degradation when limits hit.

🧠 Thinking Mode

Enable extended reasoning with thinking: { level: 'high' } for complex architecture and debugging tasks.

🛠️ Built-in Tools

Code execution (Python sandbox), Google Search (real-time grounding), and file analysis (PDF, images, video).

📊 Structured Output

Every call returns typed AgentResult. Pass a Zod schema or responseFormat (JSON Schema) for custom-shaped output.

Models

Model	Best For	Speed	Cost	Default
`gemini-2.5-flash-lite`	General purpose, fast	~200ms	Very low	✅ Default
`gemini-3-flash-preview`	Complex reasoning, thinking	~400ms	Low

🦙 Ollama — Local Models (Free, Private)

Run models locally with Ollama. No API key needed. 100% private.

Model	Best For	Size	Cost
`llama3.2`	General purpose	~2GB	Free	🦙 Default
`mistral`	Strong reasoning	~4.1GB	Free
`codellama`	Code-focused tasks	~3.8GB	Free

Use Cases

🛡️

Security Auditing

console.agent.security("check for SQL injection", userInput)

🐛

Debugging

console.agent.debug("why is this slow?", { duration, query })

📊

Data Validation

await console.agent("validate batch meets schema", records)

🏗️

Architecture Review

console.agent.architect("review API design", { endpoint })

🔢

Math & Reasoning

await console.agent("calculate optimal batch size", metrics)

🔍

Research

console.agent("research CVEs for lodash@4.17.20")

The jQuery ofAI Agents

See the Difference

Why console.agent?

Agents are too complicated

Wrong abstraction layer

console.agent is the jQuery of agents

Quick Start

Install

Set API Key

Use it

How It Works

Test Suite Superpowers

The jQuery of AI Agents

Real Test Results

Works with:

API Reference

console.agent(prompt, context?, options?)

Return Type: AgentResult

AI Personas

Security

Debugger

Architect

General

Built-in Tools

Google Search

Code Execution

URL Context

Combining Multiple Tools

Configuration

Full Config

🔒 Privacy & Anonymization

💰 Budget Controls

⏱️ Rate Limiting

🧠 Thinking Mode

🛠️ Built-in Tools

📊 Structured Output

Models

🦙 Ollama — Local Models (Free, Private)

Use Cases

Security Auditing

Debugging

Data Validation

Architecture Review

Math & Reasoning

Research

The jQuery of
AI Agents

`console.agent(prompt, context?, options?)`

Return Type: `AgentResult`