Our Conversational Agent Is Now Open Source: What Two Years in Production Taught Us 

Jaime is an open-source AI agent built and operated by First Line Software since 2023. It runs live on firstlinesoftware.com — handling visitor conversations, detecting intent, surfacing content, and routing qualified leads to HubSpot automatically. In May 2025, FLS released Jaime 2.0 and open-sourced the full codebase at github.com/firstlinesoftware/fls-jaime.

Most AI services companies explain what they build with decks and case studies. We’ve decided to publish our code, because it’s the only claim that can be verified. 

When we say that our conversational agent Jaime uses intent recognition, a CTO can open the repository and confirm that. When we say we’ve run it in production for two years across two major versions with zero downtime, the commit history is the evidence. No press release replaces that. Enterprise AI investments fail most often not at launch — but at month 12, when the underlying model changes and the system that looked good in the pilot becomes a maintenance liability. Jaime has been through that. Twice. The commit history is the record.

This is what transparency as trust actually means for an engineering company: not a principle stated in an About page, but a practice visible in a public repository.

“Publishing production code is a signal of confidence, not risk. We are certain enough in how we build to show it publicly.” 

Vladimir Litoshenko
SVP, Global Business, First Line Software

We also believe strong engineering should be reusable. Open sourcing selected parts of our work means clients benefit from production-tested foundations rather than starting from scratch on every engagement. It means the community gets building blocks that have been exercised in real traffic, not toy examples. And it means FLS builds on a codebase that improves continuously — not one that quietly ages inside a single client contract.

Jaime is the first repository. More will follow. This is the beginning of FLS building in public — not a one-off release.

What Jaime is — and why we built it on our own site first

Jaime is an AI-powered website agent built and operated by First Line Software since 2023. It runs live on firstlinesoftware.com, handling visitor conversations 24/7: answering questions, detecting user intent, surfacing relevant content, and routing qualified leads directly into HubSpot.

We built it on our own site deliberately. Before recommending AI agents to clients, we needed to operate one ourselves — through version upgrades, model changes, regression events, and cost fluctuations. Two years of continuous operation gave us something no proof-of-concept can: an honest operations record. That record includes a model deprecation we didn’t plan for, a regression we caught before it reached users, and a version upgrade that took weeks instead of months — because the architecture was built for it.

In May 2025, we released Jaime 2.0 and simultaneously open-sourced the full codebase. The upgrade incorporated intent recognition, Smart Action Buttons, conversation history, and deeper CRM integrations. The open-source release is the operations manual for how we manage AI in production.

What Jaime can do for a website 

Jaime is not a chatbot. The distinction matters technically: it operates with awareness of page context, visitor history, and detected intent — and it acts on that awareness rather than waiting to be prompted.

Intent-aware navigation Detects what the visitor is trying to accomplish and proactively surfaces relevant paths — without waiting to be asked. A visitor on a services page gets prompted with the right next step because Jaime knows where they are and why they’re likely there.

Automated lead routing Routes qualified visitors to HubSpot, triggers Slack and email notifications — automatically, based on conversation content and detected intent signals. 200+ conversations logged in 60 days with zero SDR hours spent on initial qualification.

Rich content responses Summarizes the current page, surfaces case studies, and delivers structured answers in context — reducing the time between “I’m interested” and “I want a conversation,” without requiring a human to be available.

Multilingual by default Responds in the visitor’s language automatically. Voice interaction is on the roadmap. Built for enterprise scale, not US-only deployments.

Session memory Visitors can save answers and return to previous sessions. Designed for B2B buying cycles where evaluation spans multiple visits over weeks — Jaime picks up where the conversation left off.

Modular architecture Supports new LLMs, input modalities, and integrations without a platform rewrite. The v1-to-v2 transition is the working proof: a major capability upgrade that didn’t require starting over.

Why open-source matters for AI evaluation 

Enterprise AI investments fail most often not at launch, but at month 12 — when the underlying model changes, dependencies shift, and the system that looked good in the pilot becomes a maintenance liability. The question a CTO should ask of any AI vendor isn’t “does it work in a demo?” It’s: what happens when the next model ships?

Jaime’s architecture was designed from the start for continuous evolution. There are no hidden dependencies or brittle integrations. The v1-to-v2 transition is the working proof — a major capability upgrade that didn’t require a platform rewrite. The open source release removes the need to take our word for it. Read the repository. Inspect the abstractions. Form your own view.

For companies evaluating FLS for Managed AI Services, this is also the reference implementation. We manage AI for clients the same way we managed Jaime: continuous evaluation, regression detection, cost monitoring, version upgrades. The difference is now you can inspect the evidence before you decide.

“We didn’t rebuild Jaime. We evolved it. That’s the difference between AI that ages and AI that compounds.”

Vladimir Litoshenko
SVP, Global Business, First Line Software

How companies can deploy Jaime on their own site

The open-source release is a deployable framework, not a demo. If your organization operates a website serving enterprise buyers, technical evaluators, or multi-geography visitors. The setup time depends on your stack, but the codebase is structured to be readable and deployable without a dedicated AI team. If you can run a Node.js application and connect a HubSpot account, you have what you need to get started. The Jaime codebase gives you a production-grade starting point for a site agent with the following properties:

Integrations included: HubSpot, Slack, and email automation are wired in. Lead routing, conversation logging, and CRM triggers work from day one.

Architecture you can extend: The modular design supports swapping LLMs, adding new input types, and integrating with your existing toolchain without a rebuild. Customize the intent model, the action buttons, and the conversation logic to match your use case.

No black box: Every component is readable, auditable, and modifiable. If your security or legal team needs to review what the agent does with visitor data, they can read the code.

FLS does not sell Jaime as a standalone product. We offer it as a framework — and as the foundation for Managed AI Services engagements where we build, deploy, and operate AI agents for clients the same way we operate Jaime ourselves.

Read the code

The full codebase is on GitHub. For technical audiences, it is the most complete answer to the question “how does FLS build AI?”

→ Explore the repository
→ ⭐ Star the repo

Frequently Asked Questions

What is Jaime? Jaime is an open-source AI agent for enterprise websites, built by First Line Software. It detects visitor intent, routes qualified leads to HubSpot, and responds in multiple languages — 24/7, without SDR involvement.

Is Jaime free to use? The codebase is open source and available at github.com/firstlinesoftware/fls-jaime. First Line Software also offers deployment and Managed AI Services engagements for companies that want Jaime operated for them.

How long has Jaime been running in production? Since 2023 — over two years, across two major versions. Jaime 2.0 was released in May 2025.

What integrations does Jaime support? HubSpot, Slack, and email are included out of the box. The modular architecture supports additional integrations without a platform rewrite.

What is the difference between Jaime and a chatbot? A chatbot responds to questions. Jaime detects intent based on page context and conversation history, and proactively guides the visitor toward the right next step — without waiting to be asked.

Is Jaime related to First Line Software’s Managed AI Services? Yes. Jaime is FLS’s own MAIS reference implementation. The open-source release documents how FLS builds, operates, and evolves AI in production — the same approach applied to client engagements.

Last updated May 2026 · First Line Software