What we can build
We design the agent, build the product it runs inside, and turn rule-heavy domains into calculation engines. You can hand us only the part you need.
AI agent development
We design the flow that takes a question, finds grounds, and answers.
We build a graph, not a single prompt. When to call a tool, what to remember, and where to fix an answer that came out wrong — all of it is visible in the code. Document retrieval is wired in so the model reports what it found instead of inventing it.
- Conversation graph, branching, and retry design (LangGraph)
- Long-term memory stored apart from chat history, so the user is remembered across sessions
- Vector search for grounded answers, connected to your own documents
- LangGraph
- OpenRouter
- Qdrant
- MongoDB
AI product build and operation
We build the whole service the agent runs inside.
An agent alone is not a product. One team builds the screen people use, the streaming responses, the conversation storage, and the admin tools your operators work in. After launch the same team keeps tuning answer quality.
- Web and app front ends with streaming response UI
- API server, conversation storage, and admin tools
- Post-launch operation of prompts and retrieval quality
- Next.js
- NestJS
- Flutter
- Cloudflare
Domain engines
We move rule-heavy domains into verifiable calculation.
Ask a model to calculate and it gets it wrong. Domains with fixed rules, such as Saju and astronomy, are implemented as separate calculation engines, and the agent calls them as tools. The engine owns the arithmetic; the model owns the explanation.
- Accurate chart derivation from astronomical and calendrical rules
- Engines exposed as tools the agent can call
- Separated per engine so other services can reuse them
- astronomy-engine
- lunar-typescript
- tyme4ts
How we work
What happens after you get in touch, and in what order.
Inquiry
Tell us what you want automated. An early idea is fine. We reply within one or two business days.
Scope
We decide what the agent will answer and what it will not. This is also where we tell you what today's technology cannot do.
Agent design
We settle the conversation graph, what to remember, which tools to call, and which documents to retrieve. What we agree here becomes the build spec.
Build
The agent and the product are built together. Progress is shared as a screen you can actually talk to.
Launch
We ship to production and wire up conversation logs, response latency, and error visibility.
Operate
We read real conversations and tune the prompts and retrieval. Answer quality rises here, not at launch.
Frequently asked
- How is an AI agent different from a chatbot?
- Instead of picking a scripted reply, it works through a procedure. It reads the question, picks the tools it needs, pulls facts from a calculation engine or your data, and answers grounded in those facts. That is why, when an answer is wrong, we can point to the step that caused it.
- Do you train a model on our company data?
- No. We do not train on it. We retrieve the documents an answer needs and use them as grounds (RAG). We can also leave the data in your own storage and connect only the search. Which data lives where is written down during the design phase.
- Which model do you use? Can it be changed?
- We use several models through OpenRouter. The model is a configuration value, so you can use different models for different jobs, or switch to a better one when it arrives without touching the code.
- How long does it usually take?
- It depends on scope, but adding an agent to an existing service is typically one to two months, and building the product alongside it three to four. The tuning period after launch is counted separately.
- How do you estimate cost?
- We fix the scope first, then estimate by feature. Model calls cost money continuously on top of the build, so we also work out an expected monthly running cost from your projected usage.
- Can we operate it ourselves later?
- Yes. Prompts are managed as data rather than code, so they can be changed without a deploy. At handover we document the graph structure and how to operate it.
Take only the part you need
The agent only, the product only, or all of it. We set the scope together.