Botpress

Botpress is an open-source platform for building, deploying, and extending AI agents powered by large language models. Unlike traditional chatbot builders, Botpress treats agents as code — providing a first-class TypeScript SDK, a CLI for scaffolding and deployment, and a publicly maintained library of integrations covering everything from Slack and Gmail to GitHub and Google Drive. The platform runs on Botpress Cloud, but the SDK, CLI, integrations, and plugins are all open-source and available for community contribution.

At the heart of modern Botpress is LLMz, an agent framework that replaces JSON-based tool calling with real TypeScript code generation. Instead of orchestrating tools through multiple round-trips to the LLM, LLMz agents write and execute TypeScript in a secure sandbox — enabling loops, conditionals, and multi-tool coordination in a single inference pass. This approach dramatically reduces costs and latency compared to conventional agentic frameworks like LangChain or CrewAI.

The repository is organized as a pnpm + Turbo monorepo with clearly separated packages: the @botpress/sdk for building integrations and bots, @botpress/cli for project scaffolding and deployment, @botpress/client for type-safe API access, llmz for the code-execution agent runtime, and zai for LLM-powered data operations. Alongside these devtools, the repository hosts 40+ production integrations maintained by the Botpress team, plus community-contributed plugins and example bots.

Botpress is designed for developers who want to build production-grade AI agents without black-box abstractions. Integrations can be deployed publicly to the Botpress Hub, bots can be written entirely in code using the SDK and CLI, and the plugin system extends agent behavior across conversations. The platform powers millions of production agents and is actively developed with frequent releases.

What You Get

• LLMz Agent Runtime - A TypeScript code-generation agent framework that executes real TypeScript in a secure sandbox instead of JSON tool calls, enabling complex logic, loops, and multi-tool coordination in a single LLM inference.
• @botpress/sdk - A full-featured TypeScript SDK for defining and implementing integrations, bots, plugins, and interfaces, with typed client wrappers, Zod-powered schema validation, and lifecycle hook support.
• @botpress/cli - A CLI tool (bp) for scaffolding new integrations and bots, building, deploying to Botpress Cloud, and publishing to the public Botpress Hub with a single command.
• 40+ Open-Source Integrations - Production-maintained integrations covering communication (Slack, Gmail), productivity (Google Calendar, Notion, Confluence), AI providers (OpenAI, Anthropic, Cerebras, Fireworks AI), and developer tools (GitHub, Linear, Jira).
• Zai Data Operations Library - A composable LLM-powered data transformation library supporting operations like summarize, extract, filter, label, classify, sort, and patch over structured data.
• Plugin System - Reusable agent behaviors (HITL, knowledge, analytics, personality) that plug into bots via the SDK without modifying core agent logic, available from Botpress Studio.
• Type-Safe API Client - @botpress/client provides fully typed HTTP access to Botpress Cloud APIs for managing bots, tables, conversations, messages, and user state programmatically.

Common Use Cases

• Customer support automation - A SaaS company deploys a Botpress agent that handles tier-1 support tickets, queries a knowledge base integration, escalates to a human via the HITL plugin, and logs outcomes to a CRM integration.
• AI-powered internal assistant - An enterprise team builds an agent using the SDK that queries Google Drive for documents, answers HR policy questions via the knowledge plugin, and creates Linear issues from Slack commands.
• Multi-step workflow agent - A developer uses LLMz to write an agent that fetches data from Airtable, applies conditional logic across multiple records, and sends a formatted summary to a Slack channel — all in one TypeScript execution.
• Custom integration publishing - A developer uses bp init and the integration SDK to build a connector for a proprietary CRM, deploys it privately to their workspace, then publishes it to the Botpress Hub for community use.
• Code-driven bot development - An experienced developer bypasses Botpress Studio entirely, writing bots programmatically using @botpress/sdk and @botpress/client for full version-control and CI/CD integration.
• LLM data processing pipeline - A data engineer uses zai operations to extract structured fields from unstructured text, classify records by category, and patch a database — without building custom prompt chains.

Under The Hood

Architecture Botpress is structured as a large TypeScript monorepo with clearly bounded package contexts enforced through pnpm workspace isolation and Turbo task orchestration. Each package exposes a clean public API through explicit index exports, with integrations, plugins, bots, and devtools organized into distinct workspace members that can be built, tested, and deployed independently. The SDK defines well-typed contracts for bot and integration lifecycle events — registration, action dispatch, message handling, webhook processing, and workflow orchestration — while the CLI handles the toolchain for code generation and deployment. The integration pattern follows a declarative definition file coupled with an implementation file, enabling the platform to generate typed client wrappers at build time and maintain compatibility across SDK versions.

Tech Stack The entire codebase is TypeScript targeting ES2022 with strict compiler options, managed through pnpm workspaces and Turbo pipelines for parallel builds and test runs. The SDK uses Zui — a Botpress-internal extension of Zod — for schema definition and validation, while the CLI bundles via esbuild and includes a template generation step for scaffolding new integrations. Testing runs through Vitest with both unit tests and end-to-end test suites for the CLI. Code quality is enforced by a layered toolchain: Oxlint for fast lint passes, ESLint with typescript-eslint for deeper static analysis, Prettier for formatting, and Husky with lint-staged for pre-commit validation. The LLMz package implements a TypeScript sandbox for executing LLM-generated code with abort signals, citation tracking, component rendering, and multi-exit state machines.

Code Quality Test coverage is present across core packages, with Vitest test files found in the SDK, CLI, plugins, and LLMz runtime. The SDK’s bot and integration implementations are thoroughly typed, with handler maps, ordered dispatch queues, and workflow proxy abstractions all expressed through TypeScript generics. Zui schema validation is used pervasively for runtime data integrity across integration configurations, action parameters, and message payloads. The codebase follows consistent naming conventions, uses structured error hierarchies, and applies the reporter/wrapper pattern for cross-cutting concerns like error reporting in integrations. CI tooling is comprehensive with multi-step lint, type-check, format, and test passes orchestrated via Turbo.

What Makes It Unique The defining innovation in Botpress is LLMz — an agent execution model that generates and runs TypeScript code rather than JSON tool calls. This allows LLMs to express loops, conditionals, multi-tool composition, and error handling in a single inference, dramatically reducing both cost and latency compared to traditional agentic frameworks. Rather than prompt-chaining or sequential tool invocation, LLMz agents produce executable programs that are validated, sandboxed, and run with full TypeScript type-checking. Paired with Zui — an extended Zod library that provides schema introspection for code generation — the platform achieves a tight integration between LLM output and typed runtime execution that most agent frameworks treat as separate concerns.

Self-Hosting

Botpress is licensed under the MIT License, which is one of the most permissive open-source licenses available. You can use it commercially, modify it, distribute it, and include it in proprietary products without any copyleft obligations. The repository explicitly states that all packages are open-source software under MIT, and contributors agree to release their code under the same terms. There are no open-core restrictions, source-available clauses, or commercial license requirements for the code published in this repository.

The key operational reality for self-hosting is that this repository contains the developer tools — the SDK, CLI, integrations, and agent frameworks — not a self-hosted server runtime. Botpress Cloud is the managed platform where bots and integrations actually run. Developers building on this codebase deploy to Botpress Cloud using bp deploy; there is no bundled self-hostable server in this repository. If you need an on-premise deployment option, that would require a separate commercial arrangement with Botpress, as the classic Botpress v12 (which did ship a self-hosted server) is maintained in a separate repository at github.com/botpress/v12.

Compared to using Botpress Cloud directly through their visual Studio interface, working from this repository means you get full control over integration code, version history, and CI/CD workflows, but you give up the no-code Studio editor, managed infrastructure, SLAs, and the streamlined onboarding experience. Public integrations deployed to the Botpress Hub must be made immutable once published, which limits hotfix workflows. For teams that need a fully self-hosted conversation AI stack with no dependency on Botpress infrastructure, evaluating alternatives like Rasa, Flowise, or a custom deployment of v12 would be more appropriate.