Botpress
Type: low-code · Vendor: Botpress · Language: TypeScript · License: MIT (legacy v12 OSS); proprietary cloud platform · Status: active · Status in practice: mature
Visual + code AI agent platform (Studio drag-and-drop, ADK TypeScript library, Desk for human handoff, Webchat, Hub integrations) for building LLM-powered chatbots with autonomous nodes, knowledge bases, and event triggers.
Description. Botpress is a complete AI agent platform combining a visual Studio (drag-and-drop), an ADK (TypeScript library for code-first agents), Desk (customer support workspace for human-AI teams), Webchat, and a Hub of integrations. Agent behaviour is built from Workflows containing Nodes, including Autonomous Nodes that delegate decisions to an LLM with tools. Eight built-in Agents (Summary, Personality, Policy, Translator, Knowledge, Vision, Analytics, HITL) run in a fixed sequence each turn. The legacy v12 OSS engine is MIT-licensed; the current cloud Studio/Desk product is proprietary.
Agent loop shape. Per-turn execution runs eight built-in Agents in a fixed order (Summary, Personality, Policy, Translator, Knowledge, Vision, Router, Analytics, HITL). The bot's logic itself is a graph of Workflows of Nodes; a standard Node executes its Cards top to bottom, while an Autonomous Node hands control to an LLM that decides which tools to call and which transition to take. Triggers (e.g. Conversation Started, Custom Triggers from Webchat or integrations) enter Workflows at well-defined entry points.
Primary use cases
- customer-support chatbots with human handoff to live agents via Desk or HITL plugin
- visual no-code agent design with LLM-backed Autonomous Nodes
- code-first agents built with the Botpress ADK (TypeScript)
- multi-channel conversational agents across Webchat, Slack, WhatsApp, Teams, and other Hub integrations
- knowledge-base-grounded Q&A agents over uploaded documents and websites
Key concepts
- Autonomous Node → tool-use (docs) — LLM-driven node that picks tools, writes replies, and chooses transitions from context.
- Agents (Summary/Personality/Policy/Translator/Knowledge/Vision/Analytics/HITL) (docs) — Eight built-in Agents that extend a bot's capabilities; they execute in a fixed sequence each turn.
- Knowledge Base → agentic-rag (docs) — Documents, websites, and files the bot can query for grounded answers.
- Triggers → event-driven-agent (docs) — Nodes that only execute when a named event fires (conversation start, custom Webchat events, integration events).
- Desk / HITL plugin → conversation-handoff (docs) — Escalates conversations from the bot to a human agent in Botpress Desk or a third-party platform.
- Conversation variables → session-isolation (docs) — Per-conversation scoped state cleared when the conversation ends.
Patterns this low-code implements —
- ★★Agentic RAG
Knowledge Bases are a documented first-class resource for grounding bot answers in documents/websites/files; the Knowledge Agent retrieves from them.
- ★★Conversation Handoff to Human
Desk + HITL plugin escalate to human agents; Start HITL Card placed in a Workflow triggers the handoff and the bot resumes if the user cancels or queue times out.
- ★★Event-Driven Agent
Triggers are first-class nodes that only execute when a named event (Conversation Started, Custom Trigger from Webchat, integration event) fires.
- ★★Session Isolation
Conversation-scoped variables auto-isolate per conversation and are deleted at end; bot/user/conversation scopes are explicit.
- ★★Tool Use
Autonomous Node calls user-defined tools; ADK 'Define tools' API exposes them to the LLM.
- ★★Visual Workflow Graph
Studio is a visual, drag-and-drop interface for building agents; Workflows are graphs of Nodes connected by transitions.
- ★★Approval Queue
HITL integration explicitly supports approval of AI-generated outputs before delivery; the Start HITL Card is placed in a Workflow to hand off conversations, and the HITL Agent's Queue Timeout govern…
- ★★Code Execution
Tools in the ADK are functions the AI model can call during execute(); the model decides which tools to call based on the conversation, and TypeScript handler functions hold the custom logic. Workflo…
- ★Agent Skills
Built-in Agents extend a bot's capabilities, but they are a fixed catalogue (Summary, Personality, Policy, Translator, Knowledge, Vision, Analytics, HITL) running in a fixed order, not user-authored…
- ★★Cross-Session Memory
Conversation variables persist within one conversation and are cleared at end; ADK 'Manage state' exposes user-scoped and bot-scoped state, but the documented default scope is the single conversation.
- ★Multilingual Voice Agent Stack
Botpress ships a Translator Agent in its 8-agent pipeline; voice-specific multilingual handling is not a documented headline feature.
- ★★Scheduled Agent
Honest 'none' for scheduled-agent — the docs index page (llms.txt) describes Workflows as 'long-running background processes' in its index entry, but the actual Workflows concept page (fetched 2026-0…
References
Provenance
- Last analyzed:
- Last updated:
- Verification status: partial