Build, deploy, and monitor AI agents across channels, tools, and data. The complete AI agent platform, powered by the latest LLMs.
In 2017, two builders in Quebec started a company on a hypothesis that sounded slightly absurd at the time: that computers would one day speak and operate software the way people do. They called it Botpress and shipped it as an open-source chatbot framework - the kind of thing developers starred on GitHub and quietly wired into support widgets. Then large language models arrived, the ground shifted, and the unglamorous plumbing Botpress had been laying turned out to be exactly what everyone suddenly needed.
Today Botpress calls itself "the complete AI agent platform." Strip the marketing and the claim is more specific than it sounds. The company builds the layer between a language model and a working product - the part that decides what an agent is allowed to do, remembers what happened three messages ago, calls the right tool, hands off to a human when it is stuck, and does all of that without falling over in front of a real customer.
Most people meet AI agents as a chat bubble in the corner of a website. Botpress is the machinery behind that bubble. Its cloud platform, launched in 2023, bundles a visual Agent Studio for designing conversation logic, an Autonomous Engine that lets an agent reason with an LLM, Knowledge Bases that ingest a company's documents and websites, Human Handoff for the moments a bot should get out of the way, and Tables for structured data. You can build with no code, or drop down to developer-level control when you need it.
The idea Botpress keeps returning to is the Autonomous Node. Older chatbots were flowcharts - rigid trees of "if the user says X, reply Y." Pure LLM bots are the opposite: fluent, but hard to trust. The Autonomous Node lets a single agent choose which mode to use. You prompt it in plain language, hand it a set of tools, and it decides in the moment what to say and which tool to call. Structure where you want guarantees, reasoning where you want flexibility.
Anyone can build an agent that dazzles in a five-minute demo. The trouble starts when it has to work against a real CRM, a real payments system, and a customer who phrases things in ways no one anticipated. That gap - between a prototype and something a business can actually depend on - is where Botpress has planted its flag.
One of Botpress' own investors put the same point more bluntly. "Many companies can build agent prototypes," said Adam Coccari of HubSpot Ventures, "but getting them to work reliably with existing business systems is where most efforts stall." The pitch, then, is not a smarter chatbot. It is the safe execution environment, the memory, the tool integration, and the runtime isolation that keep an agent honest once it is live.
The customer list is unusually wide. On one end, solo builders and startups spin up agents on the free plan - one memorable example is a French-language agent that recommends cheese. On the other, large enterprises and Fortune 500 teams run production agents for customer support, fintech, IT services, and consumer tech, backed by SOC 2 and GDPR compliance and dedicated customer success. The fact that the same platform bends to both without changing shape is part of the point.
That range shows up in the numbers: roughly 3,000 paying customers, nearly a million free users, and around 2,000 developers signing up every day. Revenue, the company has said, has doubled every quarter since it began charging, and Botpress expected to pass $10 million in annual revenue in 2025.
Botpress' sharpest strategic decision is a refusal. It will not pick a side in the LLM wars. Builders can mix and match models across a workflow - OpenAI here, Anthropic there - optimizing each step for cost, speed, or capability. Perron describes the ambition as becoming the "Switzerland of AI agents": neutral ground that works with everyone and is beholden to no single model provider.
That neutrality is also a hedge. If the frontier model landscape reshuffles every six months, a platform that is model-agnostic gets stronger as the field churns, not weaker. It is the same instinct that made the company keep its name through the shift from "chatbot" to "agent" - stay in the lane, and let the lane widen into a highway.
In June 2025, Botpress closed a $25 million USD Series B (about $34 million CAD) led by Framework Venture Partners, at a $120 million post-money valuation - more than double the $50 million it was worth after its 2021 Series A. That brought total funding to roughly $45 million.
The more interesting detail is who joined. HubSpot Ventures and Deloitte Ventures both wrote checks - two firms whose parent companies are exactly the kind of organizations that would buy agent infrastructure. When your potential customers become your investors, it is a quieter, more credible signal than any valuation headline.
Botpress competes in a crowded and fast-moving market. On the builder side sit Voiceflow and Rasa; in enterprise conversational AI, Yellow.ai, Haptik, and Ada; and from the platform giants, Microsoft's Copilot Studio and Google's Dialogflow. What Botpress leans on to stand apart is its developer-first heritage, its model-neutral stance, and a decade of unglamorous groundwork on the reliability problem that most competitors are only now confronting.
Whether Botpress becomes the default infrastructure layer for agents or one strong option among many is still an open question. But its bet is coherent and old: the value is not in the flashy demo, it is in the plumbing nobody wants to build. From a 2017 open-source project to a company shipping agents for thousands of businesses, Botpress has mostly been doing the same thing the whole time. The world just caught up to it.