Skip to content

Key Concepts

Understanding a handful of concepts will make everything else in AgenFleet click. This page is worth reading before you go deep on configuration.


An agent is the core unit of AgenFleet. It’s a persistent AI process with:

  • A model (the underlying LLM doing the reasoning)
  • A SOUL file (its identity, instructions, and personality)
  • A memory store (what it remembers across conversations)
  • Tools (capabilities: web search, file access, API calls, etc.)
  • Sessions (the active conversations it’s participating in)

Agents are always-on. Unlike a simple API call, a deployed agent retains context, has persistent memory, and can be reached at any time — by a human or by a scheduled cron job.


A session is a conversation thread with an agent. Every interaction happens inside a session.

  • Sessions are append-only — nothing is ever deleted from a session’s history
  • Sessions accumulate context over time, which improves continuity but increases cost
  • You can open multiple sessions with the same agent (e.g., one per project or client)
  • Cron jobs run in their own isolated sessions by default — they don’t pollute your chat sessions

A SOUL file is a Markdown document that defines who an agent is. It contains:

  • The agent’s name, role, and purpose
  • Its communication style and tone
  • Standing instructions and priorities
  • What it should and shouldn’t do
  • Any domain knowledge you want embedded in its context

Think of the SOUL file as a permanent system prompt that loads every time the agent starts. It’s the difference between a generic AI assistant and an agent that feels purpose-built for your organization.

See SOUL Files for a full guide with examples.


Tools are the capabilities an agent can invoke during a task. Without tools, an agent can only reason and generate text. With tools, it can take action.

Built-in tools available on AgenFleet:

ToolWhat it does
web_searchSearch the web and retrieve current information
memory_searchQuery the agent’s long-term memory store
memory_storeSave facts, summaries, or context to memory
send_messageSend a Telegram or Slack message
read_fileRead files from the agent’s workspace
write_fileWrite or update files in the agent’s workspace
run_scriptExecute scripts in the agent’s container
http_requestMake HTTP API calls

You control exactly which tools each agent has access to. An agent with no tools enabled cannot take external actions — useful for agents you want to keep strictly conversational.


A cron job is a scheduled task that runs an agent automatically at a set time or interval. Each cron job:

  • Sends a prompt to the agent (what you want it to do)
  • Runs in an isolated session (so it doesn’t bleed into your chats)
  • Delivers the output to a channel you specify (Telegram, Slack, email)

Example: a daily briefing agent that runs every morning at 8 AM, searches for industry news, and sends a summary to your Telegram.

See Cron & Scheduling for full setup instructions.


The fleet is all your agents, viewed as a system. The Fleet dashboard gives you:

  • Real-time health status for every agent
  • Token usage and cost per agent
  • Session counts and activity trends
  • Alerts for agents that are offline or over budget

Most organizations run multiple specialized agents rather than one general-purpose agent. The fleet view is how you manage them without losing track of any.


Each agent has a private memory store — a searchable database of facts, summaries, and context it has accumulated over time.

Memory works in two directions:

  • Storage — the agent writes important information to memory during a task
  • Retrieval — the agent queries memory at the start of a task to recall relevant context

Memory is scoped to the individual agent. Two agents don’t share memory unless you build an explicit hand-off.


ConceptOne-line definition
AgentA persistent AI process with memory, tools, and identity
SessionA conversation thread with an agent
SOUL FileThe agent’s identity, instructions, and personality
ToolsCapabilities the agent can invoke (search, files, APIs)
Cron JobA scheduled task that runs an agent automatically
FleetAll your agents, monitored as a unified system
MemoryThe agent’s private long-term knowledge store