How seratul03/evocode works
EvoCode turns a programming problem into candidate code, evaluates it in a sandbox, and uses population evolution and feedback to improve results. Its autonomous and chat entry points lead into the evolution engine; the evaluation layer coordinates generation, review, testing, and scoring. A separate meta-evolution branch monitors performance and proposes upgrades to the system’s agents. The README documents these capabilities, while excerpts and the source index provide only partial implementation evidence, so unsampled relationships are omitted where they cannot be established.
Main entry points: evoflow.py Evolution engine · evaluator.py Population evaluator · client.py LLM client · clone_builder.py Clone builder
Components 28
Problem solving
- Autonomous runner
run_autonomous.py - Interactive chat
chat.py - Evolution engine
src/evoflow.py
Code agents
- Code generator
src/agents/generator.py - Solution architect
src/agents/architect.py - Code reviewer
src/agents/reviewer.py - Bug fixer
src/agents/bugfixer.py - Code optimizer
src/agents/optimizer.py - Genetic mutator
src/agents/mutator.py - Memory historian
src/agents/memory_agent.py
Evaluation runtime
- Population evaluator
src/orchestration/evaluator.py - Agent genome
src/genome.py - Docker sandbox
src/sandbox.py - Fitness scoring
src/fitness_scorer.py - Property testing
src/property_tester.py - Run event logger
src/event_logger.py - Evolution memory
src/memory_manager.py
Meta-evolution
- Performance watcher
src/meta_evolution/watcher.py - Clone builder
src/meta_evolution/clone_builder.py - Patch judge
src/meta_evolution/source_judge.py - Safety referee
src/meta_evolution/referee.py - Fast duel arena
src/meta_evolution/meta_arena.py
LLM access
- LLM client
src/client.py - Ollama provider
src/llm/ollama_provider.py - Groq provider
src/llm/groq_provider.py - OpenRouter provider
src/llm/openrouter_provider.py
Other components
- User
- LLM services
Connections 31
- User → Autonomous runner: starts run
README.md - User → Interactive chat: starts chat
README.md - Autonomous runner → Evolution engine: runs evolution
run_autonomous.py - Interactive chat → Evolution engine: solves problems
- Evolution engine → Population evaluator: evaluates population
src/orchestration/evaluator.py - Population evaluator → Code generator: generates code
- Population evaluator → Code reviewer: reviews code
src/orchestration/evaluator.py - Population evaluator → Code optimizer: checks optimization
src/orchestration/evaluator.py - Population evaluator → Property testing: generates tests
src/orchestration/evaluator.py - Population evaluator → Docker sandbox: executes candidates
- Population evaluator → Fitness scoring: scores results
- Evolution engine → Solution architect: plans solutions
src/evoflow.py - Evolution engine → Bug fixer: repairs failures
src/evoflow.py - Evolution engine → Genetic mutator: mutates candidates
src/evoflow.py - Evolution engine → Memory historian: uses experience
- Evolution engine → Run event logger: records events
src/evoflow.py - Evolution engine → Agent genome: evolves agents
src/evoflow.py - Evolution engine → Performance watcher: triggers upgrades
run_autonomous.py - Performance watcher → Clone builder: requests patch
src/meta_evolution/__init__.py - Clone builder → Patch judge: proposes changes
- Clone builder → Safety referee: checks safety
- Clone builder → Fast duel arena: tests candidates
- Evolution engine → Evolution memory: records lineage
- Code generator → LLM client: requests completion
- Code reviewer → LLM client: requests review
src/agents/reviewer.py - LLM client → Ollama provider: tries first
src/client.py - LLM client → Groq provider: falls back
src/client.py - LLM client → OpenRouter provider: falls back
src/client.py - Ollama provider → LLM services: calls model
- Groq provider → LLM services: calls model
- OpenRouter provider → LLM services: calls model