How dainius1234/kai-system works
Kai is a multi-service sovereign AI system. The principal documented workflow starts with operator input, gathers context and persistent memories, then runs a staged reasoning swarm; when action is proposed, policy and approval gate capability dispatch, execution, and verification. Memory, perception, world-state, autonomy, and vault synchronization are substantial parallel capabilities. The README documents these relationships, but most implementation excerpts are absent, so the graph uses only documented links and does not imply that every service is connected or that the documented autonomy path is fully cut over.
Main entry points: agentic/app.py Agentic Orchestrator · swarm.py Reasoning Swarm · dashboard/app.py Dashboard · memu-core/app.py Memory Engine
Components 23
User Experience
- Dashboard
dashboard/app.py - Telegram Bot
telegram-bot/app.py
Cognition and Memory
- Agentic Orchestrator
agentic/app.py - Reasoning Swarm
agentic/swarm.py - Swarm Stages
agentic/swarm_stages.py - Memory-Driven Planner
agentic/planner.py - Plan Adversary
agentic/adversary.py - Conviction Scoring
agentic/conviction.py - Memory Engine
memu-core/app.py - Knowledge Graph
memu-graph/app.py - Obsidian Vault Sync
vault-sync/app.py
Perception and World
- Perception Ingress
common/perception_spine/ingress.py - World State
common/world_state/reducers.py
Governance and Autonomy
- Verifier
verifier/app.py - Policy Engine
common/policy_bridge/policy_engine.py - Approval and Capability
common/policy_bridge/capability.py - Tool Gate
tool-gate/app.py - Bounded Autonomy
common/autonomy/authority.py
Action and Operations
- Sandboxed Executor
executor/app.py - Actuator Registry
common/actuator_registry/handlers.py - Fleet Supervisor
supervisor/app.py
Other components
- Operator
- LLM Provider
Connections 25
- Operator → Dashboard: submits requests
README.md - Operator → Telegram Bot: sends messages
README.md - Dashboard → Agentic Orchestrator: routes requests
- Telegram Bot → Agentic Orchestrator: routes requests
- Agentic Orchestrator → Memory Engine: gathers context
README.md - Agentic Orchestrator → Reasoning Swarm: runs pipeline
README.md - Reasoning Swarm → Swarm Stages: executes stages
README.md - Reasoning Swarm → Memory-Driven Planner: uses planning
README.md - Reasoning Swarm → Plan Adversary: challenges plans
README.md - Reasoning Swarm → Conviction Scoring: scores conviction
README.md - Swarm Stages → LLM Provider: reasons with
- Agentic Orchestrator → Verifier: requests critique
README.md - Verifier → Memory Engine: gates promotion
README.md - Memory Engine → Knowledge Graph: uses graph memory
README.md - Obsidian Vault Sync → Knowledge Graph: syncs notes
README.md - Perception Ingress → World State: updates state
README.md - World State → Agentic Orchestrator: enriches context
README.md - Agentic Orchestrator → Tool Gate: submits actions
- Tool Gate → Policy Engine: evaluates proposals
- Policy Engine → Approval and Capability: returns decision
- Approval and Capability → Actuator Registry: authorizes capability
README.md - Actuator Registry → Sandboxed Executor: dispatches action
README.md - Sandboxed Executor → Verifier: submits for verification
README.md - Bounded Autonomy → Actuator Registry: bounds authority
README.md - Fleet Supervisor → Dashboard: reports fleet health
README.md