When we started Botpress, the idea of autonomous agents operating real business processes was still mostly hypothetical. The technology wasn’t ready. The tools weren’t sufficient. And most agent frameworks, even today, still break down once you move beyond scripted flows or simple tool-calling wrappers.
Since then, things have changed. The underlying models have improved. LLMs have unlocked a new layer of reasoning and abstraction. But models alone aren’t sufficient to run reliable agents in production. What’s been missing (and what we’ve spent the last several years building) is the infrastructure layer that sits between raw models and real-world business systems.
Botpress just closed our Series B, raising $25 million to continue scaling that work.
The problem we’ve spent years solving
Agents that operate in production need more than a prompt window into an LLM. They need memory, tool orchestration, safe execution environments, reliable reasoning over multi-step workflows, consistent runtime behaviour, and the ability to return structured outputs that integrate into actual systems. They also need to run without fragile orchestration layers bolted on after the fact.
This is what we’ve built at Botpress. The platform includes:
- A fully isolated runtime that ships with every deployed agent, ensuring stability across platform updates.
- A custom inference engine, which handles reasoning, tool usage, code execution, and multi-turn orchestration.
- A safe code execution layer that allows agents to dynamically write and execute code without compromising system safety.
- Structured primitives for files, tables, workflows, conversations, and users that allow developers to build agents that go far beyond simple Q&A interactions.
- Deployment pipelines that let developers ship agents confidently, knowing that each version is isolated and reproducible.
This is not a theoretical roadmap. Botpress agents built on top of this custom inference engine have been live for over a year, running millions of executions across industries. In the time it might have taken users to build an agent on a local development server, they’ve instead built and shipped an agent on Botpress in a very small fraction of the time.
The efficiency gain isn’t the main story here, though: it’s the advantage that compounds over that time. As Botpress agents live, talk, and operate out in the real world, users monitor their behaviour and provide simple, natural-language feedback to adjust any undesired decisions or outcomes. In other words, they improve. The longer an agent is deployed, it doesn’t just benefit from better, more cost-efficient language models. It also benefits from the bespoke, industry- and company-specific feedback you provide to it.
The market’s catching up
The majority of software companies today are still experimenting with AI at the interface level. Very few are focused on solving the infrastructure problems that allow agents to operate reliably at scale. This gap is where Botpress sits. We’ve deliberately avoided chasing quick demos or surface-level product releases in favour of investing deeply in the foundational work that supports real deployments.
Over the past year, we’ve seen customers move from experimentation into full-scale deployments. Many of the initial early adopters were smaller, more risk-tolerant companies. But as the infrastructure has stabilized and safeguards have improved, we’re seeing growing adoption across more conservative industries and larger enterprises.
As of today, there are Botpress users in more than 190 countries. The volume of agents moving into production grows every quarter. This fundraise is not a pivot point. It simply allows us to keep scaling what is already working.
What this round will let us do
The capital allows us to expand on several fronts that directly reflect our product-led approach:
- Continue building deeper platform primitives that make agents more capable, controllable, and extensible.
- Expand developer-facing SDKs and tooling for teams integrating Botpress into existing systems.
- Support both technical and non-technical teams by improving our visual studio while preserving full programmatic control.
- Extend global coverage and infrastructure to meet increasing demand across North America, Europe, Latin America, and Asia.
Botpress supports both no-code and pro-code development because real-world deployments require both flexibility and control. Some of our customers start with pre-built templates and simple workflows; others integrate deeply with backend systems, building highly customized agents for complex processes.
You’ll likely have noticed a clear divide between no-code tools and tools that rely on skilled developers in this increasingly crowded space. In supporting hundreds of production deployments across industries, what we’ve found is that the most successful deployments have a clear way for all stakeholders to participate meaningfully in an agent’s lifecycle. This doesn’t mean paying lip service to participation: without a clear path to influencing an agent’s behaviour or the outcomes it produces, in a way that aligns with how a team already works, agent projects are doomed to fail.
Where this is all going
Over the next decade, AI agents will replace entire categories of software. Many of the tasks that currently require custom code or human operators will be automated by systems that can reason, act, and adapt across business domains. The market for this shift is large because the problem space itself touches nearly every operational system inside companies.
Our job at Botpress is to continue building the agent platform that makes this transition possible — not as a prototype, but as stable infrastructure that teams can depend on.
Thank you to all of our users, builders, customers, and partners who have helped bring the platform to where it is today. The work ahead remains substantial. But we now have the resources, team, and foundation to scale the product into the opportunity we see in front of us.
— Sylvain