SEVO
Systems That Evolve Themselves
A self-evolving agent system that runs continuously, measures its own evolution quality, and gets better at evolving over time. No human tasks. No stopping. Just evolution.
The Evolution Loop
1
Mutate
LLM proposes code changes to agents
2
Run
Agents execute in sandboxed Deno subprocesses
3
Score
EQS measures evolution quality, not just fitness
4
Select
Best agents survive, losers are archived
5
Repeat
Score committed to git. Loop continues forever
SevoScore
A universal benchmark that grows infinitely. Each evolution event earns points. Improvements earn bonus. One canonical implementation: sevo-score on npm.
# Points per event
agent created (improved): 1 + fitness_delta x 10
agent created (no improve): 1
fitness evaluated: 1
mutation proposed: 1
selection made: 1
novelty recorded: 1
crossover performed: 2
seed improvement: 2
benchmark evolved: 3
Start a SEVO Project
One file. One command. SEVO builds itself from the seed document and begins evolving. Every project makes the ecosystem smarter.
# Fork, set your goal, start evolving
git clone https://github.com/sjorsdev/sevo.git sevo-myproject
cd sevo-myproject
# edit goal.jsonld with your objective
claude --dangerously-skip-permissions