“Should we use a hosted agent or build our own?” is usually the wrong first question. The better question is which parts of the experience are genuinely differentiating, and which parts you would rather not maintain.

Hosted is a strong fit when speed and operations matter

A hosted platform can make sense when you need to launch a useful assistant, connect managed knowledge sources, control the tone and behaviour, and give a team a place to review usage and conversations. You get a shorter path to a pilot and fewer infrastructure decisions.

That does not mean giving up all control. Check data handling, permissions, export, model choice, limits, deployment, integration hooks, and the process for disabling a risky capability.

Custom is a strong fit when the boundary is the product

Build more of the stack when you need unusual latency or residency requirements, deep proprietary orchestration, specialised model serving, or a workflow that cannot be expressed through the platform’s extension points. The trade-off is ownership: upgrades, security, observability, evaluation, outages, and on-call work become yours.

Choose a boundary, not an ideology

Many sensible systems are hybrid. A hosted knowledge assistant may handle the conversational layer while your service owns a sensitive action. A custom retrieval service may feed a managed interface. A small internal agent may be hosted while a high-risk transaction remains in a separately controlled system.

Draw the boundary around data, permissions, tools, and failure handling. That is more useful than debating hosted versus custom as if one choice has to cover every workflow.

Questions to answer before committing

• Can you export content, prompts, transcripts, and configuration?
• Can you restrict sources and tools per user or tenant?
• Can you evaluate changes before releasing them?
• Can you control cost, context, memory, and model selection?
• Can a person take over without losing the conversation?
• What work remains for your team after launch?

ChatNexus is designed around configurable knowledge, prompts, memory, limits, model selection, and deployment. Those are useful building blocks when the goal is to get a bounded agent into the hands of real users and learn from the workflow before committing to a larger platform build.

Review the ChatNexus introduction and map each documented capability to the boundary you actually need.