Agents
An agent in DAEMI is more than a model configuration. It's a persistent entity with a name, a role, a personality, a memory, and a set of tools it can use to act in the world.
What Makes an Agent
Every agent is defined by a YAML profile — a single file that describes everything DAEMI needs to bring it to life.
| Field | Purpose |
|---|---|
agent_id | Unique identifier, used for memory collection naming, routing, API calls |
name | Display name in the Console |
role | One-line description of what this agent does |
personality | Tone, traits, and behavioral tendencies |
memory.transfer_policy | How memories flow from session to session |
llm | LLM provider override (defaults to system setting if enabled: false) |
Anatomy of a Profile
agent_id: koda
name: Koda
role: Coding Agent
personality:
tone: focused, pragmatic
traits:
- high conscientiousness
- detail-oriented
- direct communicator
memory:
transfer_policy: carry_all
llm:
enabled: false # use system default provider
This is a complete, valid profile. DAEMI will:
- Load it at startup
- Initialize memory store entries for
koda - Make it available in the Console as a selectable agent
- Apply the personality to every LLM prompt involving this agent
How Agents Run
When you start a session with an agent:
- Profile loads — personality, role, and memory policy are read
- LTM query — DAEMI searches the agent's LTM for context relevant to your first message
- STM initializes — a fresh short-term memory context begins for this session
- Conversation begins — each turn writes to STM; the agent can call tools via the tool graph
- Session ends — STM is distilled into LTM based on
transfer_policy
Agents vs. Teams
A standard agent operates on its own. A team is a special agent type that orchestrates multiple member agents toward a shared goal.
# Standard agent
agent_id: koda
type: agent # default, can be omitted
# Team
agent_id: tema
type: team
members:
- role: Frontend
name: Rona
- role: Backend
name: Baca
From the Console's perspective, teams and agents look the same — you start a session and talk. Internally, DAEMI handles the orchestration.
→ See Teams for multi-agent orchestration details.