How threefish-ai/negentropy works
Negentropy is a self-evolving agent system: a scheduling core dispatches work to orthogonal faculties, with documented pipelines for knowledge acquisition, problem solving, and value delivery. A typical knowledge workflow accepts a document, extracts and indexes it, then supports retrieval and knowledge management; agents can also use persistent memory and sandboxed execution. The repository includes separate Cognizes and Perceives applications alongside the main backend and UI. Major capabilities are represented from the README and source index; unsampled implementation details and inter-service wiring are not assumed.
Main entry points: agents Five faculties · engine Scheduling core · pipelines.py Knowledge API · corpus_engine.py Corpus lifecycle
Components 19
Agent orchestration
- Scheduling core
apps/negentropy/src/negentropy/engine - Five faculties
apps/negentropy/src/negentropy/agents - Standard pipelines
apps/negentropy/src/negentropy/engine
Knowledge lifecycle
- Content extraction
apps/negentropy-perceives/src/negentropy/perceives - Corpus lifecycle
apps/negentropy/src/negentropy/knowledge/lifecycle/corpus_engine.py - Knowledge API
apps/negentropy/src/negentropy/knowledge/routes/pipelines.py - Search and retrieval
apps/negentropy/src/negentropy/knowledge/retrieval/repository.py - Knowledge graph
apps/negentropy/src/negentropy/knowledge/graph/service.py
Memory and reasoning
- Dynamic memory
apps/cognizes/src/cognizes/engine/hippocampus/retention_manager.py - RAG pipeline
apps/cognizes/src/cognizes/engine/perception/rag_pipeline.py - Code sandbox
apps/cognizes/src/cognizes/adapters/postgres/sandbox - Prompt evolution
apps/negentropy/src/negentropy/engine/evolution/handlers/memory_pipeline.py
User interfaces
- Agent UI
apps/negentropy-ui/features/session/hooks/useSessionService.ts - Knowledge UI
apps/negentropy-ui/features/knowledge/utils/knowledge-api.ts - Knowledge wiki
apps/negentropy-wiki/src/app
Runtime services
- PostgreSQL
apps/negentropy/src/negentropy/models/storage.py - Observability
apps/negentropy/src/negentropy/config/logging.py
Other components
- User
- LLM provider
Connections 20
- User → Agent UI: starts task
README.md - Agent UI → Scheduling core: submits intent
- Scheduling core → Five faculties: dispatches work
README.md - Scheduling core → Standard pipelines: offers workflows
README.md - Standard pipelines → Content extraction: acquires knowledge
README.md - Content extraction → Corpus lifecycle: extracts content
- Knowledge API → Corpus lifecycle: manages runs
- Knowledge UI → Knowledge API: calls API
- Knowledge API → PostgreSQL: reads and writes
apps/negentropy/src/negentropy/knowledge/routes/pipelines.py - Corpus lifecycle → PostgreSQL: assesses corpus
apps/negentropy/src/negentropy/knowledge/lifecycle/corpus_engine.py - Search and retrieval → Knowledge graph: supports search
- Five faculties → Dynamic memory: uses memory
- Five faculties → Code sandbox: executes code
README.md - Five faculties → LLM provider: requests inference
- RAG pipeline → PostgreSQL: stores vectors
apps/cognizes/src/cognizes/engine/perception/rag_pipeline.py - RAG pipeline → LLM provider: generates answers
apps/cognizes/src/cognizes/engine/perception/rag_pipeline.py - Prompt evolution → LLM provider: proposes prompts
apps/negentropy/src/negentropy/engine/evolution/handlers/memory_pipeline.py - Five faculties → Prompt evolution: evolves prompts
- Knowledge wiki → Knowledge API: presents knowledge
- Scheduling core → Observability: records activity