Agents

Cronosaurus

Cronosaurus was one of my experiments around long-running agentic systems. It was similar in spirit to projects like OpenClaw: agents could run on a schedule, send heartbeats, monitor websites or email accounts, and take on more complex tasks. It was technically very interesting, but also relatively costly to run, so I stopped the live version and kept the learnings.

Stopped archived Updated May 26, 2026

Back to projects

Screenshots

Cronosaurus running agent chat with streaming research output — A running agent workspace with streaming research notes, evidence links, and operational state visible in the chat surface.

Cronosaurus task queue for an active agent run — A task queue view that breaks an agent run into evidence loading, layer checks, candidate listing, overlap assessment, decision recording, and notification steps.

Cronosaurus skills editor for reusable agent instructions — A skills editor for reusable instruction bundles that agents can discover and load only when they need them.

Cronosaurus tool catalog with configurable agent capabilities — A tool catalog for enabling and configuring capabilities such as search, email, crypto prices, memory, social posting, and image generation.

Cronosaurus agent activity dashboard with run history — An agent activity dashboard showing scheduled agents, run history, active jobs, tool counts, and model assignments.

Cronosaurus per-agent tools panel — A per-agent tools panel where capabilities can be toggled, configured, and scoped to a specific running agent.

Problem

I wanted to understand what happens when agents do not just answer a prompt, but need to run repeatedly, monitor things, report back, and handle tasks over time.

Solution

A practical playground for scheduled agent runs, heartbeat monitoring, website and email monitoring, and more complex task execution.

Architecture

A modular orchestration layer with scheduled agent runs, heartbeat checks, monitoring flows, and task execution paths.

Lessons learned

The project was useful because it made the operational cost of agentic systems visible. Scheduling, monitoring, state, and tool use are interesting, but they also become expensive if the system runs all the time.