-
-
A clock that doesn't track time—it deploys an autonomous AI workforce that learns and works continuously.
-
Every hour learns, adapts, and builds new capabilities—turning time into an autonomous workforce.
-
From on-demand learning to enterprise AI workforces, every hour compounds capability and value.
-
Every organization has hours to automate. clocks.works turns those hours into productive output.
-
clocks.works transforms autonomous AI labor into a scalable enterprise business model.
Inspiration
Most AI assistants wait. You ask a question, they answer, and then they stop.
But work doesn't stop. Every hour, teams lose time to repetitive research, reporting, monitoring, and administrative tasks that never seem important enough to automate. When we received the domain clocks.works, we started with a simple question:
What if a clock didn't measure time? What if it deployed work? That idea became Evo-Ninja — an autonomous AI workforce where every hour activates intelligent agents that learn skills, generate tools, and complete tasks continuously.
What it does
Evo-Ninja transforms time into productive output. Instead of waiting for instructions, the platform uses scheduled execution to activate specialized AI agents throughout the day.
Agents can : Discover skills required for a task Search existing capabilities Generate new scripts when capabilities do not exist Execute tasks locally Save successful solutions for future use Continuously improve over time
The result is a system where every hour contributes value. At clocks.works, time literally works.
Evo-Ninja's execution model is a four-step loop that runs continuously until a goal is resolved:
- Predict Next Step — Evo analyzes the current state and makes an informed prediction about the optimal next action
- Select Best Agent — Based on the prediction, the orchestrator selects the agent persona best suited to the task domain
- Contextualize Chat History — Only the most relevant prior messages are surfaced, keeping the model focused and token-efficient
- Evaluate and Execute — The agent executes a function, advances the task, and the loop begins again
Agent Personas
| Agent | Domain |
|---|---|
| Synthesizer | Reading, analyzing, and summarizing text files and documents |
| Csv Analyst | Parsing CSV data, extracting key figures, and deriving insights |
| Researcher | Searching the internet, comprehending sources, and reporting findings |
| Developer | Architecting and building software, specialized in Python |
How we built it
The system combines :
- Multi-agent orchestration
- Runtime skill discovery
- Dynamic script generation
- Local-first execution
- Persistent skill libraries
- Automated task completion loops
A Clock Scheduler triggers execution cycles, an Agent Orchestrator assigns specialized agents, and a growing script library allows the platform to become more capable with every completed task.
Unlike traditional assistants, Evo-Ninja can create new functionality when it encounters unfamiliar problems.
Technical Innovation
Evo-Ninja advances the state of autonomous AI in three concrete ways:
1. Adaptive persona selection at runtime Most agent frameworks require the developer to pre-specify which agent handles which task. Evo evaluates the task at each step and re-selects the best persona. The workforce is fluid, not fixed.
2. Self-writing and self-expanding skill library When an agent encounters a task for which no script exists, it writes one. That script is immediately available to all future runs. The system accretes capability automatically — no developer intervention required.
3. Local-first, sandboxed execution
All execution happens within a scoped ./sessions/ workspace. Scripts cannot read or write outside this boundary. This makes the system safe to deploy in sensitive environments where cloud-based code execution is not acceptable.
Tech Stack
| Layer | Technology |
|---|---|
| Agent Runtime | TypeScript / Node.js |
| Monorepo Management | Turborepo + Yarn Workspaces |
| UI | Next.js (browser app) |
| Database (UI) | Supabase (PostgreSQL) |
| Language Model | OpenAI API (GPT-4-class) |
| Web Search | SerpAPI |
| Benchmarking | Agent Protocol / agbenchmark |
| Containerization | Docker (Supabase local) |
Why clocks.works
clocks.works is not a clock application. The domain is the thesis.
| What people expect | What clocks.works delivers |
|---|---|
| A clock that displays time | A clock that deploys AI agents |
| Passive time tracking | Active time execution |
| Reminders and alerts | Completed work |
Time is the most universal resource. Every organization has hours. Most waste them on tasks that could be automated if someone had the time to automate them. clocks.works closes that loop: the hours themselves do the automation.
The domain name makes the value proposition memorable in one word. It is domain-first product design.
Challenges we ran into
The biggest challenge was autonomy. Most AI systems assume a developer has already built every required tool. We wanted agents that could continue working even when those tools did not exist.
Another challenge was balancing flexibility with safety. We solved this through sandboxed local execution, allowing generated scripts to operate within controlled boundaries.
Finally, we needed a domain interpretation that felt natural rather than forced. The breakthrough came when we stopped thinking about clocks as displays and started thinking about clocks as schedulers.
Accomplishments that we're proud of
- Built a working autonomous agent platform
- Implemented runtime skill discovery
- Enabled agents to generate new scripts automatically
- Created a self-expanding capability library
- Designed a local-first execution model
- Reimagined the domain clocks.works as an AI workforce platform Most importantly, we transformed a simple domain name into a product vision with real commercial potential.
What we learned
We learned that the future of AI is not better chat interfaces. The future is autonomous systems that perform work without requiring constant human supervision. We also learned that domain-first product design can unlock ideas that would never emerge from a traditional brainstorming process. The constraint of clocks.works ultimately became the source of the product's strongest idea.
What's next for Evo-Ninja
Our vision is to evolve Evo-Ninja into a fully autonomous AI workforce platform.
Future capabilities include :
- Long-running business workflows
- Team-based agent collaboration
- Enterprise deployment
- Custom workforce templates
- Multi-agent coordination across departments
- Marketplace for reusable skills and scripts
Today, clocks.works automates tasks. Tomorrow, it becomes an operating system for autonomous digital labor.
Built With
- docker
- next.js
- node.js
- openai-gpt-4
- postgresql
- serpapi
- supabase
- turborepo
- typescript
Log in or sign up for Devpost to join the conversation.