Turgai, an agent harness built on objectsEarly stage
Turgai extends your product.
Plug AI agents into your product today. Add tools, rules and whole features as you need them.
Plug agents into what you already run.
You or your developer connect it once. Turgai runs next to your service: your app talks to it through its API, your customers write to it in Telegram, and it reaches your systems, whatever they’re built in.
For example
Imagine Dopline, a small travel agency run by Anna. Her developer plugs two agents into its app, and now travellers just write.
Dopline · Telegram
10 days in Japan in April. Two kids, my mother can’t walk much, we love food.
Here’s a first plan: step-free hotels, short walks, a food tour in Osaka. It comes to €9,200. Shall I hold it for you?
Every new idea becomes a new part.
From then on, it grows with you: new rules, new tools, new features. What already works keeps working.
At Dopline
Anna said “Trips over €8,000 need my approval.” She checked the new rule and switched it on, and from then on it holds every time. Later she added group tours the same way.
behavior: trip desk
fields:
approveBelow: 8000
receive:
- on: tripPlanned(trip, total)
when: total < approveBelow
do:
- send: bookings.confirm(trip)
- on: tripPlanned(trip, total)
do:
- send: owner.approve(trip)Your product holds, whatever fails.
A supplier that stops answering, a model that’s down or wrong: no part loses what it knew, and none breaks your rules. Your rules are kept by Turgai, not left to the model.
At Dopline
When a flight is cancelled, Dopline rebooks everyone on it, and no trip over €8,000 goes through without Anna.
dopline.travel · tripslive
Kyoto group · 12 travellers
Flight cancelledRebooked · 12 travellers told
New flight JL 48, same day, two hours later
The product is yours.
Turgai runs on your machines or a cloud you choose and keeps everything in your own storage. Each agent can move to any model, local or cloud, and keeps its name, its history and its rules; try a new model before you rely on it.
You set the rules, they hold below the model, and the journal shows everything it did.
At Dopline
The concierge runs on a local model on Dopline’s own server, so travellers’ messages are read there, not by a cloud provider. The planner works from the trip plan and uses a bigger cloud model.
concierge
Local · Ollama
Answer travellers. Pass trip plans to the planner.
planner
Cloud · OpenAI
Plan trips. Send anything over €8,000 to Anna.
Every part is an object.
Turgai is built so that adding something new never means rebuilding what works, and a part that fails comes back with what it knew.
Small bird.
Big possibilities.
Being independent should never mean settling for less.
Turgai is in early alpha. Tell us what you’re building.