What Moji Router is
Moji Router cuts the API cost of multi-turn AI products. It owns the session, accounts for how each provider caches, and tunes the routing to each customer's own workflow. Your application keeps talking to the providers it already uses. The router decides where each turn lands.
Where the bill comes from
In the last year AI moved to long, multi-turn work: agents that take many steps, coding tools that loop, assistants that hold long conversations. Every step resends the accumulated context, the system prompt, the history, the tools, the retrieved documents, so the tokens per session climb fast.
That turns the model API into a cost of goods: large, variable, and growing faster than revenue. Most of those tokens are the same context, paid for again and again. A cache is the way to stop paying full price for it twice, if something routes to the cache.
How it works
Routing a conversation well is a question about the whole session, not each message on its own. A router that sees one message cannot reason about what is already cached or what a switch would cost. Moji Router holds the session state, so it can.
- Account for caching: the router knows how each provider's cache behaves, so it can place each turn where the next token is cheapest.
- Segment, then commit: the cheapest path through a session groups turns and switches models rarely, often once. The expensive model earns its place early, where it shapes the rest of the session.
- Hold a cost budget: the router keeps a target spend mix you set, for example most tokens to a cheaper model and the rest to a frontier one, and stays inside that envelope while protecting quality.
- Tune to your traffic: a sample of your own sessions fits the policy to how your work behaves, and shows the cost and quality you would get before you commit.
How we start
We begin by measuring your saving on a sample of your own traffic, and we show you the figure before any commitment. Pricing is scoped with you from there, tied to the value the routing delivers.
Who it is for
Moji Router fits companies whose cost of goods is dominated by frontier-API spend, running multi-turn or agentic products across more than one provider, with someone already trying to cut the bill. Coding agents, agentic SaaS, long-session assistants, and retrieval-heavy products with large reused context all fit. It does little for self-hosters running open models for free, where there is no API bill to cut, or for single-provider shops, where routing across providers has nothing to choose between.
Where it came from
We built a multi-agent workspace and ran it ourselves. Watching the hosted-agent API fees climb out of proportion to the work, we built the router to bring them back down. The cost problem is one we hit first-hand, which is why we started here.
The name
Moji Router is the working name while the routing company stands up under the Moji brand. Treat it as provisional. The four shapes carry over from the wider Moji work, a diamond, square, triangle, and circle in gold, blue, green, and orange. Here they stand for the providers the router routes across.
The company
Moji Router is built by Frequency AI Ltd, a company registered in England and Wales (company number 16990328).
Frequency AI Ltd
71-75 Shelton Street, Covent Garden
London WC2H 9JQ, United Kingdom