Core Concepts
Companion (AI Partner)
A Companion is the heart of DesireCore — your personal AI partner. Each Companion has its own persona, memory, and skills, stored locally in a CompanionFS repository:
companions/<agent_id>/
├── companion.json # Configuration
├── persona.md # Personality & character
├── principles.md # Behavioral rules
├── memory/ # Memory system
│ ├── timeline/ # Chronological memory
│ ├── topics/ # Topic-based memory
│ ├── pinned/ # Pinned memories
│ └── lessons/ # Lessons learned
└── skills/ # Skill definitions
Receipts
Every time a Companion executes a task, it generates a Receipt — a complete evidence chain including inputs, outputs, tool calls, and reasoning. You can always see "what it did and why."
Tool Registry
Companions interact with the outside world through tools. DesireCore includes 7 built-in tools:
| Tool | Description |
|---|---|
Read | Read files |
Write | Write files |
Edit | Edit files |
Bash | Execute commands |
Glob | Search for files |
Grep | Search file contents |
Ls | List directories |
Additional tools and services can be integrated via MCP (Model Context Protocol).
Dual Runtime
DesireCore supports two AI runtimes, switchable on demand:
- Claude Agent SDK — Powered by Anthropic Claude
- pi-agent-core — Multi-model LLM powered
Step Control
Each execution step falls into one of three categories:
| Type | Description |
|---|---|
| Frozen | Deterministic execution, strictly follows rules |
| Flexible | Adaptive execution, AI decides autonomously |
| Human Gate | Requires human confirmation before proceeding |