# Botpress Autonomous Nodes: When to Use One, When Not

*How-to — 2026-10-23 — by Mahmoud Zalt*

What an Autonomous Node does, its instructions, tools and exit conditions, and where a Standard Node still wins, including human handoff.

**Short answer.** **An Autonomous Node hands the decisions to an AI model. A Standard Node runs its cards in order, top to bottom, every time.** Use the Autonomous Node where the conversation is open-ended and you want the bot to work out what to do next. Use a Standard Node where the sequence must be identical on every run, which is most escalation, payment and compliance work.

**About this guide.** This guide is based on publicly available information and Botpress's own documentation, checked on 17 August 2026. Sistava is not affiliated with Botpress. Interfaces and pricing change often, so confirm the current steps in their documentation before you rely on them.

This is the design decision that shapes everything else about a Botpress bot. Get it right and the flow stays small and the bot handles surprises. Get it wrong and you either spend weeks drawing branches for conversations nobody has, or you watch an AI improvise its way through a refund policy.

The good news is that this is not an either-or choice at the bot level. Botpress lets you mix node types in the same workflow, so the usual answer is an Autonomous Node handling the conversation and Standard Nodes owning the few paths that must never vary.

## What is an Autonomous Node in Botpress?

It is a node that uses AI to make decisions, such as what the bot should say or which of its tools it should run. Rather than executing a fixed list of actions, it reads the conversation, weighs the instructions you wrote, and chooses.

The practical effect is that one node can cover a lot of ground. A support assistant built on an Autonomous Node does not need a separate branch for every phrasing of every question, because deciding what the user meant is part of the node's job rather than yours.

## How is an Autonomous Node different from a Standard Node?

A Standard Node executes each of its cards one by one, then transitions to the next node. An Autonomous Node uses a model to decide which tools to run and in what order, based on context. Everything else follows from that one difference.

|  | Standard Node | Autonomous Node |
|---|---|---|
| What decides the order | You do. Cards run top to bottom. | A model does, from context and your instructions. |
| Moving on | Transitions after its cards finish, or when a card transitions manually. | Loops on each user message until an exit condition is met. |
| Knowledge Bases | Not searched automatically. You add a card that queries them. | Searched by a default Search Knowledge Card covering all Knowledge Bases. |
| Best at | Sequences that must be identical every run. | Open-ended conversation where the next step depends on the answer. |

The Knowledge Base row is the one that surprises people most, and it is worth reading twice. Adding sources does nothing for a Standard Node unless a card explicitly queries them.

## What can you configure inside an Autonomous Node?

Six sections, and they map neatly onto the questions you would ask a new hire on their first day. What is the job, what may you touch, what tools do you have, what should you read, what larger processes can you start, and when are you finished.

- **Instructions: **the node's main prompt. Botpress's guidance is that the more specific you are, the better the node's decision-making.
- **Variables: **read and write access to variables. Disabled by default, which is a sensible default to leave alone until you need it.
- **Tools: **the capabilities the node can call. The model decides when to use each one based on your instructions and the conversation.
- **Search Knowledge: **which Knowledge Bases this node is allowed to reach.
- **Workflows: **sub-workflows the node can execute mid-conversation.
- **Exit Conditions: **the conditions under which the node should leave its loop.

One detail from the documentation is easy to skim past and worth acting on. You can refer to tools in natural language inside your instructions, but Botpress notes the bot may produce better responses when you refer to them using code instead. If a node keeps ignoring a tool you clearly described, that is the first thing to try.

## What tools does an Autonomous Node have by default?

Four, before you add anything of your own. Knowing they exist explains a lot of behaviour that otherwise looks like the node inventing things to do.

- global.think: process a response or pause before continuing.
- global.search: query internal knowledge bases.
- global.Message: send a text message.
- clock.setReminder: schedule a reminder.

Botpress also documents three model-level settings on the node: override the default model, override the default model used for retrieval, and a Vision Agent option that extracts text from images the user sends. That last one is quietly useful for support, where people screenshot the error rather than describe it.

**Do not economise on the model here.** Botpress's own recommendation for Autonomous Nodes is to use a high-performing model, on the level of GPT-4, to avoid prompt truncation and to make sure the tool definitions are understood properly. A node that reads its tools imperfectly does not fail loudly, it just quietly stops using them, which is a much harder thing to debug.

## How does the Autonomous Node loop work?

By default the node responds to every user message in a loop until one of its exit conditions is met. That is what makes it feel like a conversation rather than a single scripted reply, and it is also what makes exit conditions a required piece of design rather than an optional refinement.

You can turn that behaviour off. Botpress documents an advanced setting controlling whether the node loops until an exit condition, which you disable when you want the node to process logic in the background rather than hold a conversation.

There is a cost dimension to this too. A DataCamp write-up on Botpress makes the point directly: pricing scales with model calls, so more elaborate logic costs more to run. A node that loops and thinks on every message is not the same running cost as one that answers and exits.

This is the part of the design where the two philosophies diverge honestly. Botpress gives you the dials so you can tune the trade-off yourself. At Sistava we took the other route and removed the dials: you hire an AI Employee, tell it in plain English what you want handled, and it sets itself up with the tools already connected. Neither is universally right. It depends on whether tuning this is work you want.

## When should you use a Standard Node instead?

Whenever the sequence must be identical on every run and a wrong order has real consequences. Payments, cancellations, anything with a compliance script, and above all handing the conversation to a person. In those paths, predictability is the feature and improvisation is the bug.

Human handoff is the clearest worked example, and Botpress has two routes for it. For new workspaces, Human Handoff has moved to Botpress Desk, reached from Human Handoff and then Open Botpress Desk. Existing workspaces can use the legacy human-in-the-loop integration, which Botpress lists as requiring a Plus plan or higher.

1. **Install the HITL integration** — From Explore Hub in Studio, install the official human-in-the-loop integration.
2. **Install and configure the plugin** — Install the HITL plugin from Botpress Hub, then configure it by selecting the HITL integration from the dropdown.
3. **Add the Start HITL card where the handoff belongs** — Place it at the point in your workflow where the conversation should transfer. Its required fields are a Ticket Title, the User ID and the Conversation ID, which are supplied from the event.
4. **Work the tickets from the dashboard** — Tickets appear in the left sidebar. You can filter by status and take one with Assign to me, then chat with the user directly in the thread.
5. **Give control back deliberately** — When the person is finished, Resume Chatbot returns the conversation to the bot.
6. **Test the escalation before you rely on it** — Botpress suggests using the agent's preview to simulate an escalation and confirm the ticket really appears in the Human Handoff dashboard.

Notice what that sequence is: fixed, ordered, with required fields. It is exactly the kind of path you do not want a model deciding its way through, which is why it belongs in a Standard Node even in a bot that is otherwise autonomous.

## How do you debug what an Autonomous Node decided?

Open the Emulator in Studio with Cmd or Ctrl plus E. Alongside the conversation it prints inline logs covering bot events, Knowledge Base activity and results, code execution, AI Task operations, workflow transitions and variable capture attempts.

That log is the whole game with an autonomous node. When a reply is wrong you are not looking for a broken line of code, you are looking for a decision, and the decision is only visible if you read what the node did between messages rather than what it said at the end.

A closing rule of thumb that has held up well. Start autonomous, then carve out. Let one node handle the conversation, watch real transcripts, and every time you find a path where a specific sequence had to happen and did not, move that path into a Standard Node. You end up with a small flow and a bot that copes.

## FAQ

### What is the difference between an Autonomous Node and a Standard Node in Botpress?

A Standard Node executes its cards one by one, top to bottom, then transitions to the next node. An Autonomous Node uses an AI model to decide what to say and which tools to run, based on the conversation and the instructions you wrote. The Standard Node is predictable by design, the Autonomous Node is flexible by design, and most real bots use both.

### What are exit conditions in a Botpress Autonomous Node?

They define when the node should leave its loop. By default an Autonomous Node responds to every user message in a loop until one of its exit conditions is met, so without them the node never hands over. Botpress also documents an advanced setting for whether the node loops at all, which you disable when you want it to process logic in the background rather than hold a conversation.

### Which tools does a Botpress Autonomous Node have out of the box?

Four built-in tools: one for thinking, which lets it process a response or pause, one for searching internal knowledge bases, one for sending a text message, and one for setting a reminder. You add your own on top of those, and the model decides when to use each based on your instructions and the conversation context.

### Which model should an Autonomous Node use?

Botpress recommends a high-performing model, on the level of GPT-4, to prevent prompt truncation and to ensure the tool definitions are understood properly. The node exposes settings to override the default model and the default retrieval model separately, plus a Vision Agent option that extracts text from images users send.

### Should human handoff be an autonomous or a standard node?

Standard. Escalation is a fixed sequence with required fields, a ticket title, a user ID and a conversation ID, and a specific card that starts it. That is the kind of path where a model choosing its own order adds risk without adding value. Keep the conversation autonomous and put the escalation itself on rails.

### Why does my Autonomous Node ignore a tool I told it to use?

Two common causes. Botpress notes that while you can refer to tools in natural language in your instructions, the bot may generate better responses when you refer to them using code, so the reference itself may be too vague. The other is model choice: a weaker model may truncate the prompt or misread the tool definitions, which shows up as the tool simply never being called rather than as an error.

## Sources

- [Botpress documentation, ](https://botpress.com/docs/studio/concepts/nodes/autonomous-node)Autonomous Node
- [Botpress documentation, ](https://botpress.com/docs/studio/concepts/nodes/introduction)Nodes
- [Botpress documentation, ](https://botpress.com/docs/studio/concepts/knowledge-base/introduction)Knowledge Bases introduction
- [Botpress documentation, ](https://botpress.com/docs/studio/get-started/manage-your-agent/human-handoff/)Configure Human Handoff
- [Botpress documentation, ](https://botpress.com/docs/integrations/integration-guides/hitl/introduction)Human-in-the-Loop (HITL)
- [Botpress documentation, ](https://botpress.com/docs/studio/concepts/emulator)Emulator
- [DataCamp, ](https://www.datacamp.com/tutorial/botpress)Botpress Tutorial: Build and Deploy AI Chatbots

**Tags:** botpress, autonomous-agents, chatbot-design, human-handoff, conversational-ai