How e-ai-model/shadowseed-pro works
Shadowseed Pro is a model-independent memory and validation engine, with a local Workbench as its first complete client. The principal workflow prepares authorized context, lets the host or Workbench call a model, then observes the answer to detect and intake candidate seeds, handle recurrence, and route evidence or contradictions through the Validation Gate. Workbench services add chat management, inspection, comparison, feedback, scenarios, and export. The graph follows the README and sampled orchestration/API excerpts; links into unsampled runtime subsystems are limited to documented responsibilities.
Main entry points: chat.py Turn orchestration · controller.py Workbench controller · engine.py Engine API · sessions.py Session service
Components 25
Engine integration
- Engine API
src/shadowseed/engine.py - Turn orchestration
src/shadowseed/chat.py
SSL pipeline
- Seed registry
src/shadowseed/manager.py - Answer detection
src/shadowseed/detection/model_detector.py - Seed intake
src/shadowseed/intake.py - Recurrence handling
src/shadowseed/recurrence.py - Validation Gate
src/shadowseed/gate/runtime_adapter.py - Seed lifecycle
src/shadowseed/lifecycle.py - Influence authorization
src/shadowseed_agent/agent_contract.py
Local Workbench
- Workbench UI
src/shadowseed/workbench/app.py - Workbench controller
src/shadowseed/workbench/controller.py - Production controls
src/shadowseed/workbench/production_controller.py - Session service
src/shadowseed/application/sessions.py - Inspection
src/shadowseed/application/inspection.py - Comparison
src/shadowseed/application/comparison.py - Reports and export
src/shadowseed/application/exports.py - Tester feedback
src/shadowseed/application/feedback.py - Scenarios
src/shadowseed/application/scenarios.py
State and providers
- SQLite workspace
src/shadowseed/storage/sqlite.py - Ollama adapter
src/shadowseed/adapters/ollama_client.py - OpenAI adapter
src/shadowseed/adapters/openai_client.py
Retrieval bridge
- Seed retrieval probe
src/shadowseed/retrieval_probe.py - Vector store
src/shadowseed/vectorstore/base.py
Other components
- Host application
- Language model
Connections 26
- Host application → Engine API: prepares turn
- Engine API → Turn orchestration: delegates turns
- Engine API → Host application: returns context
- Host application → Language model: generates answer
- Host application → Engine API: submits answer
- Turn orchestration → Seed registry: orchestrates state
- Turn orchestration → Answer detection: detects candidates
- Turn orchestration → Seed intake: intakes candidates
- Turn orchestration → Recurrence handling: handles recurrence
- Turn orchestration → Validation Gate: routes validation
- Turn orchestration → Seed lifecycle: advances lifecycle
- Turn orchestration → Influence authorization: checks influence
- Workbench UI → Workbench controller: dispatches actions
- Production controls → Workbench controller: wraps controls
- Workbench controller → Session service: manages sessions
- Workbench controller → Comparison: requests comparison
- Workbench controller → Inspection: requests views
- Workbench controller → Reports and export: requests exports
- Workbench controller → Tester feedback: records feedback
- Workbench controller → Scenarios: runs scenarios
- Session service → Turn orchestration: runs turns
- Session service → SQLite workspace: persists state
- Ollama adapter → Language model: requests generation
- OpenAI adapter → Language model: requests generation
- Seed retrieval probe → Vector store: searches vectors
- Influence authorization → Validation Gate: checks Gate authority