Choosing conversational AI open source software means you own the build: hosting, models, integrations, security and every fix after launch. For a company with an engineering team and an unusual requirement, that can be the right trade. For a 60-person firm that wants enquiries answered and meetings booked, it usually is not. This article explains what an open source route really involves, how to decide, and how to set up the configured alternative on the levelpartner.ai platform if you choose not to build.
What open source gives you and what it asks for
Open source conversational AI frameworks give you the code for understanding messages, managing a dialogue and connecting to channels. You can read it, change it and run it on your own servers. Nobody can raise your licence fee or switch off a feature you depend on.
In return, you take on the parts a vendor would otherwise carry. Someone must host the system and keep it patched. Someone must connect it to your CRM, calendar and phone line, and keep those connections working when either side changes. Someone must train or configure the models, monitor the conversations and answer the call when it breaks on a Saturday.
The questions to answer before choosing
- Do you have a developer who can own this for years, not just build it once?
- Is there a requirement no configured product meets, such as on-premise hosting or a proprietary system?
- Who will be on call when the agent stops replying?
- What will the integrations to your CRM and calendar cost to build and to maintain?
- How will non-technical staff update answers without a code change?
If most answers are "we would need to hire", the hidden cost of open source is a salary.
Before you start with the configured route
- Your account with contacts and the channels you want to cover.
- The conversations you planned to build, written as a list with the outcome of each.
- Any open source prototype you already have, so you can reuse its question and answer content.
How to set it up
- Go to AI Agents > Conversation AI and click + Create Bot.
- Select Prompt Based Bot.
- Choose Start from Scratch to rebuild your prototype's intents.
- Continue into the bot editor.
- Configure Bot Settings.
- Configure Bot Training with the prototype's question-and-answer content.
- Configure Bot Goals using prompts, model settings and actions.
- Select the bot, then click Deploy.
- Turn on the one channel you will go live on first.
Before step one, list every intent the prototype handles, or would have handled, with a sample message for each. In Bot Goals, the custom code you would have written becomes platform actions such as Appointment Booking, Update Contact Field and Trigger Workflow. Test each intent with its sample message, compare the replies with the prototype, and archive the open source code rather than running both side by side.
Comparing the two routes in practice
Say the developer estimates six weeks to build an open source agent with booking and CRM updates, then a day or two a month on maintenance. The configured route might take a few days of set-up inside the platform, with maintenance done as content review by an operations lead, roughly an hour a week. On those estimates, the developer's six weeks could go to the firm's client portal instead.
The point is not that open source is worse. It is that the cost lands in a different place: engineering time instead of a licence. For a firm trying to grow without adding headcount, engineering time is usually the scarcer resource.
When open source conversational AI is the better choice
Choose the open source route if you must host everything inside your own infrastructure, if you are building conversational AI into your own product for customers, or if your dialogue logic is so specific that no configured agent can follow it. In those cases the control is worth the work. Read conversational AI with Rasa for one example of a developer framework and Huggingface conversational AI for the model side.
Teams often compare the open source licence cost, which is zero, with a product subscription. Compare the full cost: build hours, hosting, maintenance and the person on call.
Weekly running time on the platform
Expect about an hour a week of content review in the first month, falling after that. Nobody needs to deploy code to change an answer. Whoever owns customer service edits the knowledge base and the change takes effect on the next conversation.
Troubleshooting a migration from open source
Your old bot had strict flows the agent does not follow. Describe the order in the instructions, or split the flow so a workflow handles the fixed steps and the agent handles the conversation.
Custom integrations have no direct equivalent. Use a workflow with a webhook step to call your own system, and keep that the only custom piece.
Answers differ from the old bot. Old answers were hard-coded. Copy the approved wording into the knowledge base so the agent uses it.
Frequently asked questions
Is open source conversational AI free?
The software licence usually is. Hosting, development, model usage and maintenance are not. For most businesses those costs are larger than a subscription.
Can we keep our data in-house with the platform?
Your conversations and contacts are held in the platform, not on your own servers. If on-premise hosting is a hard requirement, an open source route may be the only fit.
Can we start on the platform and move to open source later?
Yes. Your answer content and conversation history can be exported. Starting configured and building later, only if you outgrow it, is lower risk than the reverse. The AI agents overview explains what the configured route covers.