Skip to content
AgenFleet Docs

Agents Overview

An agent on AgenFleet is a persistent AI process — not a stateless API call. It lives in its own container, holds memory across conversations, and can be reached at any time by a human, a cron job, or another agent.

What makes an agent different from a chatbot

Section titled “What makes an agent different from a chatbot”

A traditional chatbot sends your message to an API and returns a response. Each call is independent — no memory, no persistence, no ability to take action.

An AgenFleet agent is different in three ways:

1. Persistence The agent runs continuously. It doesn’t spin up on demand — it’s always there, already loaded with context, ready to respond or execute a scheduled task within seconds.

2. Memory The agent maintains a private knowledge store that grows over time. It remembers facts, summaries, decisions, and context from previous sessions. When a new session starts, it can query its memory to pick up where it left off.

3. Tools The agent can take action — search the web, read and write files, call APIs, send messages, run scripts. It doesn’t just generate text; it can do work.

Agent anatomy

Section titled “Agent anatomy”

Every agent on AgenFleet is composed of:

ComponentWhat it is
ModelThe LLM doing the reasoning (Claude Haiku, Sonnet, or Opus)
SOUL fileThe agent’s identity, instructions, and personality
Memory storeA private, searchable database of accumulated knowledge
ToolsThe set of capabilities the agent is allowed to invoke
SessionsActive conversation threads the agent is participating in
Cron jobsScheduled tasks the agent runs automatically

Agent lifecycle

Section titled “Agent lifecycle”
Create → Configure → Deploy → Active
                                 │
                    ┌────────────┼────────────┐
                    ▼            ▼            ▼
               Chat session   Cron job    Tool call
               (human-led)  (scheduled)  (autonomous)

Once deployed, an agent is always in the Active state. It responds to:

  • Messages in a chat session (initiated by a human)
  • Cron jobs (fired automatically on a schedule)
  • Incoming webhooks or handoffs from other agents

Agent isolation

Section titled “Agent isolation”

Each agent runs in a fully isolated container:

  • Separate memory — agents never read each other’s memory stores
  • Separate tool context — tool credentials are scoped per agent
  • Separate sessions — no cross-agent session bleed
  • Separate cost tracking — every token is attributed to the specific agent that consumed it

This isolation is especially important in multi-tenant deployments where different clients share the same platform but must never see each other’s data.

Agent roles in practice

Section titled “Agent roles in practice”

Organizations typically deploy a fleet of specialized agents rather than one general-purpose agent. Common patterns:

RoleWhat it does
ConciergeFirst point of contact — routes requests, answers FAQs, escalates to humans
Research AnalystMonitors sources, surfaces intelligence, drafts summaries
Ops MonitorWatches system health, alerts on anomalies, triggers remediation
Compliance OfficerTracks regulatory changes, checks policy adherence, generates reports
Content WriterDrafts, edits, and schedules content on a set cadence
Data AnalystProcesses metrics, identifies trends, produces dashboards

Next steps

Section titled “Next steps”