I'm a solo founder. No team, no VA, no agency.
But I have Atlas doing research at 11pm. Koko writing blog posts at 2am. Caesar building my marketing plan while I'm debugging something else entirely. Turing fixing bugs I haven't even looked at yet.
None of this costs me anything. Not a dollar.
Here's exactly how it works.
The problem with "AI productivity" advice
Most articles about using AI to get more done are variations of the same thing: use ChatGPT for your emails, use Copilot for your code, use Midjourney for your thumbnails.
That's not a system. That's just outsourcing individual tasks to different tabs.
What I wanted was something closer to an actual team — agents that know their role, remember context, and hand work off to each other without me babysitting every step.
It took me a while to build it. Now it runs.
What APES OS actually is
It's a local multi-agent system I built on top of OpenClaw. Seven agents, each with a defined identity, a dedicated Discord channel, and their own workspace folder.
The lineup:
Main — orchestrates everything, runs the pipeline
APES — the autonomous one, triggers full automated runs
Atlas — deep research, sources, trend analysis
Koko — SEO content, blog posts, newsletters
Caesar — marketing strategy, 30/60/90-day plans
Turing — code, React, bug fixes
Zuri — social posts for every platform
They all talk to a web dashboard running locally at http://127.0.0.1:3000. There's a kanban board. Charts. Real-time agent status. It looks like a proper ops center.
How I actually use it day to day
I open Discord.
I type something in #atlas-research:
Research the current landscape of AI-powered SaaS tools
for indie developers. Focus on pricing models and what's
working vs. not.
Atlas comes back with a structured report — sources, statistics, emerging trends, what to do with the findings.
I forward that to Koko with one message:
Turn this Atlas research into a 1000-word SEO blog post.
Target keyword: AI tools for solo founders. Include a CTA.
Koko writes the post. I review it. Maybe I change two sentences.
Then I run the whole thing as a pipeline:
Run full pipeline on "AI tools for solo founders 2025".
Atlas researches. Koko writes the blog. Zuri makes social posts.
Caesar builds a 30-day promotion plan. Bundle it all at the end.
Twenty minutes later I have a research report, a blog post, five platform-specific social posts, and a marketing plan. I did not write any of it.
The technical side (simplified)
Three services run in the background:
OpenClaw Gateway — the brain, port 18789, handles all agent requests
Dashboard server — serves the UI at port 3000
Discord router — bridges Discord messages to the right agent via the CLI
One PowerShell script starts all three. I right-click, run it, and the dashboard opens automatically.
Every agent has a SOUL.md file in their workspace — a plain markdown file defining who they are, what they do, what they hand off to others, and a mandatory rule: log everything to Supabase after every task.
That logging rule is what makes the dashboard charts useful. It's also what keeps agents from just doing random things without a paper trail.
The cost breakdown
OpenRouter has free model tiers. My primary model is stepfun/step-3.5-flash:free. I have six fallbacks configured — all free or near-free.
Supabase is free up to 500MB. My agent logs and kanban tasks take up maybe 2MB.
Discord is free.
OpenClaw is open source.
I pay nothing per month to run this. If I switched to Claude Sonnet for everything, I'd probably spend $10–$20 depending on usage. I keep that as an option for when I need higher quality output on something important.
What's actually broken (being honest)
The 66% success rate on my dashboard is real. Some runs fail — usually because a free model hits a rate limit mid-task or returns something malformed. The fallback chain catches most of it, but not all.
Long conversations also cause problems. Context overflow after a few thousand tokens means I start fresh threads more than I'd like.
And Zuri's social posts are... fine. Not great. I still edit them before posting anywhere. The research and blog output from Atlas and Koko is genuinely good. The social content is 70% there.
What surprised me most
I expected to save time on writing. That happened.
What I didn't expect was that it changed how I think about my work. When I know Atlas can research something properly in 15 minutes, I research more things. When I know Koko can produce a draft blog in 20 minutes, I write more. I stopped bottlenecking on "I don't have time for this."
The volume of output I ship now versus six months ago is not close.
If you want to build something like this
The full setup takes a few hours — Node.js, OpenClaw, Discord bot, Supabase tables, some JSON config. It's not a one-click install.
I documented the whole thing in detail — every step, every config field, every error I hit and how I fixed it — because I wanted something I could hand to someone and have them actually finish the setup.
If you want the full guide (Notion template + PDF + all 7 agent SOUL files), I put it here: https://nexflowai.gumroad.com/l/npzufj
If you just want to poke around the open source side first, OpenClaw's docs are at docs.openclaw.ai.
Either way — the agents are waiting.
Running APES OS on Windows 11 · OpenClaw 2026.3.x · March 2026
Top comments (0)