Vyom
Multi-model AI orchestration. Route smarter, pay less, own the routing.
Useful when replacing: hard-coded OpenAI calls scattered through your codebase
Pricing · USD
Open source
Free · MIT
MIT licensed on GitHub. Clone it and it's yours — there's nothing to buy.
What it costs to run
Bring your own keys
Your provider keys, your bill, your infrastructure. We never see a token.
There's no price, because there's nothing to sell you. Vyom is MIT licensed on GitHub and runs standalone on your own provider keys — OpenRouter, Anthropic, OpenAI, whoever. The models bill you directly and we never see a token. It's open-core: the hosted version adds routing weights tuned on real usage plus managed metering and billing, but the public repo is genuinely useful on its own, and that's the point — a routing layer you can't inspect is a routing layer you shouldn't trust.
// route by capability, latency, or cost
export const routing = {
support: { model: "claude-haiku", maxCost: 0.02 },
analysis: { model: "gpt-4.1", fallback: "gemini-2.5" },
private: { provider: "vllm" }
};Live routing view
support-summarize
healthyClaude Haiku
182ms
contract-review
healthyGPT-4.1
1.4s
private-analysis
self-hostedvLLM / Llama
344ms
One endpoint, your rules, and a clean path to move models without rewriting your product.
The problem
Your product calls one model vendor directly from a dozen places in the codebase. Switching models means a refactor. Costs are opaque. And when the vendor has an outage, so do you.
How Vyom solves it
One gateway, every model
Route requests across Anthropic, OpenAI, Google, and open-weight models by cost, latency, or capability — with automatic failover.
Own the routing layer
Self-hosted and Apache licensed. Your prompts, usage data, and routing logic stay in your infrastructure.
Spend visibility
Per-feature, per-team cost attribution. Know which product surface burns tokens before the invoice tells you.
Delivery terms — stated precisely
How Vyom is delivered, what you control, and where the boundaries sit.
Open source · MIT
Clone, run, and modify it on your own provider keys.
Data and hosting
Deploy in your cloud or use the stated optional managed-hosting plan. Data location and operator access are agreed before production use.
Compliance boundary
Residency, controls, and regulatory requirements are implementation decisions documented in the selected plan or delivery scope.
Rented vs. owned
The rented way
hard-coded OpenAI calls scattered through your codebase
- Metered pricing that scales with your success, not the value
- Your data sits in their database, exportable only through their API
- Prices rise once you're integrated and can't leave
- Switching is a migration project nobody will approve
The Vyom way
Free, MIT, and yours the moment you clone it
- Free and MIT licensed — clone it, run it, fork it, we don't get a say
- Bring your own provider keys; the models bill you and we never see a token
- Runs entirely in your infrastructure — your prompts stay yours
- Open-core: the hosted version adds tuned routing, but the repo stands alone
Built on
Vyom — the questions buyers ask
Who owns the code and IP?
For a Build Sprint, the repository, deployment assets, and work-product IP described in the SOW are handed over to you. Pre-existing products retain the licence shown on their product page; hosted services remain services. We state that boundary before work starts.
What happens if a sprint misses its scope?
We finish it. Fixed price means the risk of estimation is ours, not yours — a missed scope item is completed without a new invoice. Changes you request are re-scoped explicitly, so the line between 'our miss' and 'new ask' stays honest.
How do time zones work with a pod in India?
Every engagement includes 4+ hours of overlap with US or EU working hours. Async by default — written updates, weekly clickable demos — and meetings only when they earn the slot. You'll never chase a status report.
How is our data and code kept secure?
Work happens in your repos and your infrastructure wherever possible — we hold the minimum access needed, revoked at handover. NDAs are standard on request, and everything we self-host for you runs behind your keys, not ours.
How fast can you actually start?
Scoping takes a day; most sprints start within a week of sign-off. Pods are deliberately small and senior, so your start date is confirmed in the sprint plan — not left floating.
Do we need a retainer afterwards?
No — that's the point. A sprint ends with handover docs and training. If you want us to keep maintaining what we built, that's an optional flat monthly add-on you can cancel any month, because the software keeps working without us.
Own Vyom, don't rent it.
See the price, deploy on your infrastructure, keep the code. Or scope a sprint and we'll customise it and hand it over.