Agent Concepts

New to AI agents? This page explains the core ideas behind Autonomis in plain language. No jargon, no prerequisites — just the concepts you need to get the most out of your agents.

What Is an AI Agent?

An AI agent is like a smart employee that lives in the cloud. It can think, remember things, use tools, and work around the clock. Unlike a simple chatbot that just answers questions, an agent can take actions, make decisions, and collaborate with other agents.

Think of the difference between asking someone a question (chatbot) and hiring someone to manage a project (agent). An agent doesn't just respond — it does things. It can search the web, write documents, send emails, review code, and much more.

The Manager + Specialist Pattern

The most effective way to use Autonomis is with a team of agents, just like a small company:

  • One "manager" agent — This is your go-to agent. It coordinates everything, delegates tasks, and keeps track of the big picture. It's the first agent you talk to.
  • Specialist agents for specific tasks — These are experts at one thing. A "researcher" agent finds information. A "writer" agent creates content. An "analyst" agent crunches numbers.

When you give the manager a complex task, it automatically figures out which specialist to call and delegates the work. You don't need to manage the hand-off — the manager handles it.

Example

You tell your manager: "Research our top 3 competitors and write a comparison report." The manager sends the research task to the researcher agent, then passes the findings to the writer agent to create the report. You get a polished deliverable without managing the workflow yourself.

How Agents Think

Under the hood, your agents use large language models (LLMs) like Claude to understand what you say and figure out what to do. But they're more than just a language model — they have superpowers:

  • They can use tools — Read files, search the web, send emails, write code, generate images, and more
  • They remember context — Your agent knows what you talked about earlier in the conversation
  • They follow your rules — Whatever instructions you put in their templates, they follow consistently
  • They can work independently — You can give them tasks and they'll work on them without constant supervision

Agent Memory

One of the biggest challenges with AI is memory. Regular chatbots forget everything once you close the window. Autonomis agents are different — they have multiple layers of memory:

Short-Term Memory

This is the current conversation. Your agent remembers everything said in the current chat session. When conversations get very long, older messages are automatically summarized to make room for new ones.

Long-Term Memory (Vector Memory)

This is your agent's superpower. Memory files are automatically scanned every hour and indexed using AI. This means your agent can search its memory by meaning — if you discussed "reducing customer churn" last week, the agent can find it even if you ask about "keeping users happy."

Vector memory persists across restarts, so your agent never truly forgets. Memories are retained for 90 days.

Daily Notes

Agents can write notes about what they did each day. This creates a running log that helps them (and you) keep track of progress over time.

MEMORY.md

This is a special file your agent maintains on its own. Think of it as a curated knowledge base where the agent writes down important decisions, lessons learned, and key facts. Unlike vector memory (which is automatic), MEMORY.md is carefully organized by the agent itself.

Your agent is pre-configured to maintain MEMORY.md automatically. At the start of every conversation, the agent reads this file as part of its Session Boot sequence, ensuring it never forgets key context across sessions.

Agent Lifecycle

Here's the journey of an agent from creation to retirement:

  1. Create — You create an agent from the dashboard. It gets deployed in its own isolated environment within seconds.
  2. Configure — Set up its personality with templates, add secrets for external services, and install skills to give it abilities.
  3. Use — Chat with your agent, assign tasks, schedule automated routines, or connect it to external platforms via integrations.
  4. Monitor — Check how your agent is doing from the dashboard. Review its conversations, files, and activity.
  5. Delete — When you no longer need an agent, you can permanently remove it from the dashboard. This frees up your agent slot for a new one.

BYOK (Bring Your Own Key)

Autonomis uses a Bring Your Own Key model. This means you provide your own API key from an AI provider like Anthropic (Claude), Google (Gemini), or OpenAI (GPT).

Why does this matter? Two big reasons:

  • Full control over your AI usage — You use your own provider account, so you have full visibility into usage and billing with the AI provider directly.
  • You choose your model — Want to use Claude Opus for important tasks and Haiku for simple ones? You can mix and match models per agent to optimize for quality and speed.

Your API keys are stored securely with encryption and are never visible to anyone, not even Autonomis administrators.

Choosing the Right Model

Different tasks need different levels of AI power. Here's a simple guide to choosing the right model for each agent:

TaskRecommended ModelWhy
Initial setup and personality tuningOpus (most capable)Best results for getting your agent's personality right
Everyday conversations and tasksSonnet (balanced)Handles most tasks well with good speed
Background checks and monitoringHaiku (fast and lightweight)Perfect for simple, repetitive tasks
Writing and content creationSonnet (balanced)Great balance of quality and speed for creative work

You can set the model for each agent individually. Using the right model for each task helps optimize performance.

Getting Started

Ready to build your agent team? Here's the quickest path to a great setup:

  1. Create your first agent (the "manager") from the dashboard
  2. Fill in USER.md with your name, email, timezone, and business context — this is the most important step
  3. Customize SOUL.md for personality using templates
  4. Add your AI provider API key in secrets
  5. Start chatting — your agent automatically maintains memory across sessions
  6. When ready, add specialist agents and let the manager delegate to them

Pro Tip

Start with one agent and get comfortable. Once you see how templates, skills, and memory work together, adding more agents and building a team becomes second nature.

Next Steps