We built Parlant so your team can shape every customer conversation.
Parlant is the open-source way to build a customer-facing AI agent with a coding agent. Parlant provides the agent: the engine that keeps it on your policies, the chat widget, handoff to people, analytics and hosting. Your coding agent fits it to your business.
Our story
Parlant started as a research lab and open-source project, funded by NFX, to understand what it takes to make customer-facing AI agents effective and reliable. We spent our first year working closely with large global banks, learning the happy paths, the edge cases and the real-world patterns of AI conversations with customers.
What we learned went into an open-source engine. Parlant 3 became the most trending open-source project on GitHub worldwide, and teams everywhere picked it up to build their customer-facing agents.
Then we noticed something. Companies large and small were creating their agents with Parlant using coding agents like Claude Code, rather than writing them by hand. So many, in fact, that we realized we needed to adapt the framework so that coding agents could better understand and use it. And in many cases, the users who were building enterprise agents with Parlant were not even developers, and this made us pause.
That set our new direction with Parlant 4: turning it into a complete way for the people who know the customer experience to build their own agent with a coding agent, change it directly, check each change, and trust it to hold.
What we believe
Brands should shape their own conversations.
Brands now get millions of conversations, from both people and from their personal agents. Each one is part of how customers experience the brand. You should have control over those conversations, and be able to see how customers actually experience your brand.
Behavior should be written down.
Every rule your agent follows should live somewhere your team can read, test and change it. When a policy changes, you should know exactly what changed, which conversations it affects, and how.
You shouldn't pay per resolution.
Per-resolution fees charge you for every conversation a platform counts as resolved, which rewards the wrong kind of containment. Moreover, we believe that per-resolution fees ultimately favor the CX vendor's revenue, rather than your value as a client, because of their vague success criteria. With Parlant, you pay your model provider directly, at market prices, for what each conversation actually costs—and you control how those conversations go.
Open source, for good
Parlant's engine is open source under the Apache 2 license, and it stays free. There's no open-core catch: we don't hold features back for a paid edition.
Our commercial offering is the Parlant Dashboard, where teams build, test, host and monitor their agents. It's free while you build, then a reasonable, flat monthly price per workspace with no usage meters. We also support enterprise teams directly.
Who we are
Parlant is built by our team of researchers, engineers and designers. Research is still where our work starts: our research team develops the methods behind the engine and publishes them, including Attentive Reasoning Queries for following instructions accurately and Signals and Queries for finding which rules apply to a conversation.
Get in touch
Whether you're exploring or already building, we'd like to hear what you're working on.