54 lines (44 loc) · 2.27 KB

Project Sentinel Documentation

Personal AI agents swarm - self-evolving assistant system.

Index

Document	Description
roadmap	Development phases, milestones, current status
architecture	System design, components, data flow
agents	Agent types, roles, orchestration
memory	Memory hierarchy, persistence, retrieval
interfaces	Human-AI communication channels
task_system	Task scheduling and execution
tool_calling	LLM-driven function execution
safety	Boundaries, sandboxing, self-modification rules

Package Structure

Path	Purpose
`src/sentinel/core/`	Orchestrator, config, shared types, logging
`src/sentinel/agents/`	Agent implementations (dialog, sleep, awareness, code)
`src/sentinel/memory/`	Memory hierarchy and persistence
`src/sentinel/interfaces/`	Telegram, CLI adapters
`src/sentinel/llm/`	LLM provider abstraction and intelligent routing
`src/sentinel/tasks/`	Task scheduling, parsing, execution
`src/sentinel/tools/`	Tool calling framework, registry, built-in tools
`src/sentinel/workspace/`	Sandboxed code execution environment
`tests/`	Unit tests
`tests/integration/`	Integration tests with live APIs
`data/`	SQLite DBs, workspace, file storage

Tech Stack

Component	Technology
Runtime	Python 3.12+, uv
LLM Providers	Anthropic (Claude), OpenRouter, Local (Ollama/LM Studio)
Tool Calling	Native Anthropic tool use, OpenAI function calling
Storage	SQLite (aiosqlite) with FTS5 full-text search
Task Scheduling	SQLite-backed with async execution
Code Execution	Isolated Python venvs with sandbox validation
Interface	Telegram Bot API, CLI
Testing	pytest with integration tests

Agent Persona Files

File	Purpose	Edited By
`data/identity.md`	Agent personality, capabilities, working style	User
`data/agenda.md`	Ongoing tasks, plans, preferences, notes	Agent (self-edited)

These files are loaded at agent initialization and used to build the system prompt context.