G
dashboard
SIMULATION: not live agents. These are candidate genomes scored by a toy fitness function in your browser. No models run and nothing here is a benchmark.
HELLO EVOLVE · LAB

Agents don't get built.
They evolve. (simulated)

A seeded genetic algorithm using HELLO OS's src/lib/evolution.ts. The best candidate is kept each generation (elitism), and the rest are bred by tournament selection and mutation.

GENERATION
0
POPULATION
32
BEST FITNESS
96.85
AVG FITNESS
85.27
BEST Δ SINCE G0
+0.00
seed 20261020 · mutation 0.12 · profile balanced · stops at G200
EVOLUTION CHAMBER · G0size = fitness · click a candidate
elite (top 15%)above averagebelow average
FITNESS OVER TIME (last 1)
— best— average
GENOME VIEWER
g0-c26
fitness 96.85 · parents none (G0) · 0 mutations
reasoning0.972
coding0.667
memory0.477
collaboration0.744
TOURNAMENT ARENA · G0

A sample head-to-head pairing from this generation. Higher fitness wins.

g0-c1489.84
g0-c08 ✓90.92
QUERY CONSOLEMOCK · IN-MEMORY
-- mock query console: reads this page's in-memory simulation, not a database --
GENERATION TIMELINE
G0 / G0
BEST
96.85
AVERAGE
85.27
MUTATIONS
0
MODEL MICROVM FLEETVISION · NOT BUILT

Proposed: run each candidate in an isolated sandbox with checkpoint, fork and rollback. HELLO OS doesn't include Firecracker or any microVM runtime today, so there are no latency figures to show.