AI
Assistants and search built into your product, with OpenAI or self-hosted models
Useful features, not a demo
Most AI projects fail on the unglamorous parts: getting your content into a form a model can retrieve from, keeping that index current, and deciding what happens when the answer is wrong. We treat those as the actual project. The model is one component in a system that also needs indexing, caching, queues, permissions and a fallback.
We will also tell you when AI is the wrong answer. Plenty of problems described as needing AI are better solved with a sensible content model, a properly configured search index, or a form that asks a clearer question.
What we build
Assistants and chatbots
Assistants that answer from your own documentation and content rather than guessing, with a handover to a real person when the question needs one. The chat on this site works exactly that way.
Semantic and predictive search
Search that matches meaning rather than exact keywords, using embeddings. We built this for a membership knowledge platform where members could never find documents that were definitely there.
Self-hosted models with Ollama
Where content cannot leave your infrastructure, we run open models locally with Ollama. No per-token cost, no third party holding your data, and it still works when an API has an outage.
Content and editorial tooling
Summaries, suggested tags and metadata, and drafting help inside the CMS your editors already use, so it fits the workflow instead of sitting in a separate tab.
Integrated into your stack
AI features wired into Drupal, Laravel or Next.js with the retrieval, indexing and queues around them. That plumbing is usually the bulk of the work, not the model call.
Cost, privacy and guardrails
Which model runs where, what gets sent to a third party, what it is allowed to answer, and what it costs per month. Decided deliberately and written down.
OpenAI or Ollama, chosen on the requirements
OpenAI is the pragmatic default: strong models, no infrastructure to run, and predictable quality. It suits public content and general assistants where sending text to a hosted API is not a problem.
Ollama lets us run open models such as Llama and Mistral on your own server or ours. That matters when you are handling member data, commercially sensitive documents or anything with a compliance requirement attached, and it keeps a high-volume feature affordable because there is no per-token bill. Both are open tooling, so you are not locked into a single vendor if the economics or the models change.
Thinking about adding AI to something?
Tell us the problem rather than the technology. We will tell you whether AI helps, which model belongs where, and what it would cost to run.