Service · AI platform
AI platform and LLM cost engineering
A pilot that cost little can become a significant monthly bill once it reaches production volumes. We put a gateway, routing, evaluation and budgets in place so each team knows what its AI costs and whether a cheaper model would do.
Comonad builds the AI platform layer that controls model spend: a gateway that attributes every call to a team and use case, routing each use case to the lowest-cost model that passes your evaluations, caching, and budgets with alerts.
You cannot safely move a use case to a cheaper model without an evaluation that tells you whether quality held. So evaluation comes first, and routing decisions follow from it rather than from a price list.
We are vendor-neutral and take no resale or referral fees, so the recommendation is whichever model and hosting option passes your tests at the lowest cost. We can work within your own environment.
Tell us roughly what you spend on models, and which use cases drive it.
- CTOs and platform leads
- Finance and FinOps
- Engineering leads
On this page
Where the money goes
Spend tends to grow in a few recognisable ways: requests go to the largest model by default, long prompts and context are resent on each call, agents loop more than they need to, and nobody can say which team or use case a given invoice line belongs to.
The usual remedy is a thin platform layer that every application calls through, with the numbers to decide where a smaller model, a cache or a shorter context is good enough.
What an engagement covers
Spend review
Where the spend goes today, by use case, model and team, and which changes are likely to matter. Output: a written review with the options ranked.
Evaluation
Test sets for each use case that say whether a cheaper model, a shorter prompt or a cache keeps quality where you need it.
Gateway and routing
One entry point for model calls, with attribution, routing rules, caching, rate limits and fallbacks, built in your estate.
Budgets and reporting
Budgets per team or use case, alerts before they are breached, and a report finance and engineering can both read.
What you keep
Spend review
Where the money goes, and what to change first.
Evaluation suites
The tests that decide routing, kept in your pipeline.
Gateway
Routing, caching, attribution and limits, in your estate.
Runbooks
How to change a route, add a model and handle an outage.
How an engagement runs
Review
Spend, use cases and the bills behind them.
Evaluate
Test sets for the use cases that cost most.
Build
Gateway, routing and budgets, through your release process.
Hand over or run
Your platform team takes it over, or we run it as a managed service you can take back.
What we do not do
- We do not give legal advice. Legal sign-off stays with your counsel.
- We do not resell a model, an agent platform or a coding assistant, and we take no referral fees.
- We do not promise a percentage saving before we have seen your usage and run the evaluations.
- We do not move a use case to a cheaper model without an evaluation that shows quality held.
Who it is for
- CTOs and platform leads with several teams calling models directly.
- Finance and FinOps teams that cannot yet attribute AI spend.
- Engineering leads asked to reduce model costs without losing quality.
Alex Vakhitov
Founder, Comonad · AI and software architect, London
Alex Vakhitov founded Comonad in 2024. An AI and software architect, he was co-founder and CTO of Cloud KB, UK energy-market software, from August 2015 to March 2022.
Questions about LLM cost engineering
How much will we save?
We do not quote a number before the spend review and the evaluations. Some use cases move to a smaller model with no visible change; others need the larger one. The review tells you which is which.
Do we have to buy a gateway product?
No. We can build on open-source components, on your cloud provider's services or on a product you already license. We take no fees from any supplier.
Does this fit with governance?
Yes. A single gateway is also where logging, data boundaries and model approval are easiest to enforce, which supports our AI governance work.
Related
Discuss AI spend
Tell us roughly what you spend on models, and which use cases drive it.
Discuss AI spend