Skip to content
OPQAI.
Sourced advanced / 🏪 SME Operations Free tools

Automate Your Business with Gemini AI Agents

Job to be done: Automate content creation, sales lead generation, security scanning, and operational tasks for a one-person tech agency using AI agents.

🇳🇬 Ways to use this in Nigeria

Ideas to get you started, adapt to your situation.

  • Entrepreneur

    Deploy an AI agent to scan for security vulnerabilities in public websites, generating a list of potential clients for your web development or cybersecurity consultancy.

  • Student

    As a computer science student with a freelance web development side hustle, automate generating social media posts for your brand, with an AI agent drafting and reviewing content before you publish.

  • 9-5 employee

    As a technical marketing specialist in a startup, set up an AI agent to gather market intelligence from tech blogs and Hacker News, informing your content strategy and product insights.

What this is, in plain English

A developer ran a one-person tech agency mostly on autopilot by building four “AI agents”, small programs that do a job on their own and call an AI when they need to think. They handle content, community replies, research, finding sales leads, and keeping an eye on operations, all on the free tier of Gemini. The agents run through OpenClaw (a free framework for building AI agents) on WSL2 (a way to run Linux inside Windows).

Be honest about the level: this is an advanced, custom build, and the author did not publish every line of their setup. You cannot copy it exactly. What is genuinely valuable, and very copyable, is the set of ideas: which jobs to hand to an agent, how to keep quality up, and clever tricks to stay inside a free AI quota. This page lays those out and walks the realistic shape of building your own version.

What you can use it for

These are the agent patterns from the original, each useful on its own:

  • A self-checking content agent. It writes a post, scores its own quality from 1 to 10, and rewrites anything under 7 before you ever see it.
  • A capped community-reply agent. It answers comments in context but stops after two rounds, so conversations never spiral into wasted cost.
  • A near-free research pipeline. It gathers intelligence from RSS feeds, Hacker News, and a page-reader, using plain web requests so it spends almost no AI quota.
  • A lead-finding agent. It scans public sites for security weaknesses and turns the findings into a list of potential clients to pitch.
  • An operations monitor. It watches key services, flags stale leads, and keeps customer data in sync.
  • A publish-to-Discord agent. Triggered by a code push, it posts blog updates using the commit message, spending zero AI quota.

Tools you need

  • Gemini 2.5 Flash (freemium): the AI brain of the agents. The free tier allows around 1,500 requests a day, which is the whole point.
  • OpenClaw (free): the open-source framework you build and run the agents in.
  • WSL2 (free): runs a Linux environment on Windows, where the agents live. (Mac and Linux users can skip this.)
  • Supporting free services: RSS feeds, the Hacker News API, and a page-reader like Jina Reader for the research pipeline; Discord, Vercel, or Firebase (free tiers) for notifications and hosting if you want them.

How it actually works

You build agents one at a time inside the framework. The realistic shape:

  1. Set up OpenClaw on WSL2. Install WSL2 (if on Windows), then install OpenClaw following its README. This is the home for your agents.

  2. Connect Gemini’s free tier. Add your Gemini API key so the agents can call the model.

  3. Build the first agent with a focused prompt. Start with the content agent. Give it one clear, self-contained instruction, including the self-scoring rule, for example:

    Generate a social media post about [topic].
Then review it for clarity, engagement, and relevance and give it a score from 1 to 10.
If the score is below 7, rewrite the post to improve it, and return the better version.

  4. Keep research off the AI quota. Pull from RSS, the Hacker News API, and a page-reader with ordinary web requests, and save the results to local files. The agents read those files instead of asking the AI to browse.

  5. Add the other agents as needed. Community replies (capped at two rounds), lead scanning, ops monitoring, and the push-to-Discord poster.

  6. Optimize tokens. Feed each agent a single, fully-loaded prompt (all the context it needs at once) rather than long back-and-forth chats. That is how the whole thing fits in the free quota.

Words you’ll see, explained

  • AI agent: a small program that performs a task on its own and calls an AI when it needs to reason or write.
  • OpenClaw: an open-source framework for building and running such agents.
  • WSL2: Windows Subsystem for Linux, which runs Linux inside Windows.
  • Token: the unit AI usage is measured in; fewer tokens per task means more tasks fit in a free quota.
  • Quality gating: having the AI score its own output and redo it if it falls short.
  • Endpoint: a single service or address the ops agent checks to confirm things are working.

Original source

Based on a Show HN post by ppcvote describing four AI agents built to run a one-person tech company on the free tier of Gemini 2.5 Flash, with heavy attention to keeping token usage low.

Notes & variations

  • Do you even need four agents? No. Start with the one that saves you the most time (usually content), get it solid, then add others. The ideas stack; you do not need all of them at once.
  • Free-tier viability: 1,500 Gemini requests a day is plenty if you follow the token discipline, local files for data, single focused prompts, short conversations.
  • Common pitfall: long, chatty exchanges with an agent burn through your quota fast. Keep each interaction concise and self-contained.
  • Tip for better results: give the content agent a few of your best past posts as examples in the prompt, so its writing matches your voice.

Keep going

More SME Operations workflows