Playground
Watch an agent think.
Two experiments from our build path. The first is an agent loop: give it a goal and watch it plan, call tools with JSON arguments, recover from errors, and stop for your approval before it changes anything. The second is the single neuron that Genesis built by hand.
Experiment 1
The agent loop
Goal → plan → act → observe → review → answer. The tools are real: they read this site's progress data, its docs list and the GitHub repo.
Mission control
Tools
Scratchpad memory
- empty
- Facts the agent learns stay here between goals until you clear them.
Trace
turn 0- Pick a goal and press Run. Every step the agent takes will show up here.
The tools and data are real. The planner is a small rule-based router, not an LLM, so it handles the goals above and close variations. Swapping the router for a Claude or GPT call is exactly what Awaken will do.
Experiment 2
One neuron, by hand
Move the weights and the bias and watch the neuron's line move. Then train it. It learns AND and OR, but no single line can separate XOR. That limit is why networks have layers.
Want to see the fix? The XOR live demo trains a 2-layer network that bends the line.