How tamimlabs/nexusmind-ai works
NexusMind is an autonomous, event-driven task-execution agent: users submit goals, or watchers detect platform events; the system checks standing instructions, plans and executes work, then returns results and learns from the task. The README documents adaptive execution, memory-gated watchers, platform write-back skills, approvals, live task updates, and self-evolving skills. The sampled source confirms the API, orchestrator, memory retrieval, watcher lifecycle, Gemini client, and Firestore roles; some subsystem links reflect README documentation rather than inspected implementation.
Main entry points: orchestrator.py Task Orchestrator · watchers Event Watchers · main.py Task API · agent_loop.py Adaptive Loop
Components 21
Entry and Events
- Task API
api/main.py - Watcher API
api/watcher_routes.py - Watcher Manager
agent/watchers/manager.py - Event Watchers
agent/watchers
Task Runtime
- Task Orchestrator
agent/orchestrator.py - Task Planner
agent/core/planner.py - Adaptive Loop
agent/core/agent_loop.py - Tool Executor
agent/core/executor.py - Gemini Client
agent/core/gemini_client.py
Skills and Safety
- Skill Library
agent/core/skill_library.py - Platform Skills
agent/skills - Approval and Steering
agent/telegram.py
Memory and Learning
- Hybrid Memory
agent/core/memory - Memory Retrieval
agent/core/memory/retrieval.py
Cloud Platform
- Firestore
cloud/firestore/client.py - Pub/Sub Events
cloud/pubsub/events.py - ADK Agent
cloud/vertex_ai/agent.py
Other components
- User
- Platform Events
- Telegram
- Connected Services
Connections 25
- User → Task API: submits goal
- Task API → Task Orchestrator: dispatches task
- Task API → Watcher API: manages watchers
- Watcher API → Watcher Manager: controls lifecycle
- Watcher Manager → Event Watchers: creates and starts
- Connected Services → Event Watchers: provide events
- Platform Events → Event Watchers: deliver webhook
- Event Watchers → Task Orchestrator: trigger task
- Task Orchestrator → Memory Retrieval: recalls context
- Memory Retrieval → Hybrid Memory: searches facts
- Task Orchestrator → Task Planner: plans task
- Task Orchestrator → Adaptive Loop: runs task
- Adaptive Loop → Gemini Client: requests decisions
- Adaptive Loop → Tool Executor: executes actions
- Tool Executor → Platform Skills: invokes tools
- Platform Skills → Connected Services: reads and writes
- Tool Executor → Approval and Steering: requests approval
- Approval and Steering → Telegram: sends approval
- Task Orchestrator → Hybrid Memory: stores reflections
- Task Orchestrator → Skill Library: synthesizes skills
- Skill Library → Platform Skills: makes skills available
- Task Orchestrator → Firestore: persists task state
- Task Orchestrator → Pub/Sub Events: publishes events
- Pub/Sub Events → ADK Agent: triggers agent
- ADK Agent → Task Orchestrator: falls back to