Every AI request, routed to the right model. Up to 40% lower costs.
One OpenAI-compatible API for every leading model. Clara Router picks the most efficient model that clears your quality bar and trims wasted context. Pay with Clara and take at least another 5% off what's left.
- Billed in MXN, BRL or COP
- CFDI, nota fiscal and DIAN invoices
- Switch by changing one line
Request
model: "clara/auto"
prompt: "Summarize this supplier contract in 5 bullet points"
Candidates
- Frontier reasoning model
- Quality
- 97
- Cost
- 1×
- Latency
- 2.4 s
- Balanced modelSelected
- Quality
- 94
- Cost
- 0.6×
- Latency
- 0.9 s
- Fast small modelBelow quality bar
- Quality
- 71
- Cost
- 0.05×
- Latency
- 0.4 s
Saved vs. frontier-only
40%
Clara payment discount
+5%
One API key for models from
- OpenAI
- Anthropic
- Meta
- Mistral
- DeepSeek
- xAI
- Cohere
- Qwen
and more as they launch
The problem
Buying AI from Latin America shouldn't be this hard
AI providers were built for companies that pay in dollars from the United States. Everyone else pays a hidden tax in fees, paperwork and downtime.
Dollar invoices, local budgets
Providers bill in USD. FX spreads and taxes on international payments, such as IOF, inflate every invoice, and no forecast survives a currency swing.
Declined cards, stalled products
Banks flag recurring charges from abroad. When the card is declined, API access stops, and your product stops with it.
No valid tax invoice
A foreign receipt is not a CFDI or a nota fiscal. Finance chases paperwork every month and deductibility is at risk.
Keys and costs everywhere
Every team opens its own account with every provider. Nobody can say how much the company spends on AI, or on what.
Frontier prices for simple tasks
Most requests don't need the most expensive model. Without routing, classification, extraction and short answers all go to it anyway.
Benchmarks that only speak English
Public leaderboards rarely measure Spanish or Portuguese. Your customers notice the difference, even when the benchmark doesn't.
How it works
Live in an afternoon, not a quarter
Explore the demo
Compare sample workloads, set a quality bar and see which model wins. No sign-up or API key needed.
Change one line of code
Point the OpenAI SDK you already use at Clara Router. Streaming, tool calling and structured outputs keep working.
Optimize and save
Track cost, latency and quality per request, then pay with Clara for at least 5% off the optimized bill.
Product
Lower costs, with guardrails on quality
The router sits between your code and the providers, and makes the call your team doesn't have time to make on every request.
Spend less
Smart routing
Use clara/auto and each request goes to a model chosen by task complexity, language, latency target and your cost ceiling.
Context optimization
Redundant context is trimmed and provider-side prompt caching is applied for you, so long system prompts stop costing full price.
Response caching
Identical and near-identical requests are answered from cache, so you don't pay twice for the same answer.
Language on/off-ramps
For eligible Spanish and Portuguese workloads, compact the prompt on the way in and restore the original language on the way out.
Stay in control
Quality guardrails
Set a minimum quality bar, pin models for critical flows and compare candidates on your own traffic before switching.
Automatic fallbacks
When a provider is down or rate-limited, the request is retried on an equivalent model. Your users never see the outage.
Budgets and hard limits
Monthly budgets per key, team or project, with alerts at thresholds and a hard stop at the cap: before the invoice, not after.
Cost analysis
Cost, latency and tokens for every request, by team, feature and model. See why each route was chosen and where the savings came from.
Savings that stack
Optimize first. Then pay at least 5% less.
Routing, context and language optimizations cut spend by up to 40%, depending on your workload. Pay with Clara and take at least another 5% off what remains.
Optimization comes first
The discount applies to optimized spend, so the two stack: 40% and then 5% is 43% combined.
5% is the floor
Committed volume unlocks higher discount tiers. Your rate only goes up.
One AI bill, in Clara
Routed AI spend lands next to the rest of your company's spend, ready to reconcile.
Estimate your savings
Total monthly savings: $4,800/mo. Yearly savings: $57,600/yr.
- Optimization savings
- $4,000/mo
- Clara discount (5% minimum)
- $800/mo
- Total monthly savings
- $4,800/mo
Yearly savings
$57,600/yr
Estimates, not a guarantee. The Clara discount applies to spend after optimization.
Built for Latin America
Made here, for companies that operate here
Clara already runs corporate spend for thousands of companies in the region. The router brings AI into the same system, under the same rules.
| Country | Currency | Invoice |
|---|---|---|
| Mexico | MXN | CFDI 4.0 |
| Brazil | BRL | Nota fiscal (NFS-e) |
| Colombia | COP | Factura electrónica DIAN |
| Rest of the region | USD | Commercial invoice |
Local-currency billing
Pay in MXN, BRL or COP, or in USD if you prefer. One statement in Clara, with no international card charges.
Tax-valid invoices
CFDI in Mexico, nota fiscal in Brazil and factura electrónica in Colombia, issued automatically and attached to each charge.
Spanish- and Portuguese-aware routing
Models are evaluated continuously in Spanish and Portuguese, including regional variants, and the router uses the results.
Lower latency in the region
When providers offer capacity close to your users, such as in São Paulo, the router prefers it.
Privacy under local law
Data processing terms designed around LGPD, LFPDPPP and Ley 1581, with controls over where and how prompts are handled.
Support in your language
Help in Spanish, Portuguese and English from teams in Mexico City, São Paulo and Bogotá.
One platform, two teams
Engineering ships faster. Finance finally sees the bill.
Developers
If you use the OpenAI SDK, you're already integrated
Change the base URL and the key. Keep your code, your framework and your prompts.
- Works with the OpenAI SDKs, LangChain, LlamaIndex and the Vercel AI SDK
- Use clara/auto for smart routing, or pin any model by name
- Tag requests with team and cost center for reporting in Clara
# Same SDK you already use
import os
from openai import OpenAI
client = OpenAI(
base_url="https://clararouter.com/v1",
api_key=os.environ["CLARA_ROUTER_API_KEY"],
)
response = client.chat.completions.create(
model="clara/auto", # or pin any model by name
messages=[{"role": "user", "content": "Classify this support ticket by urgency"}],
extra_body={
# optimize for cost, latency or quality
"routing": {"optimize": "cost", "fallbacks": True},
# shows up in Clara reports
"metadata": {"team": "support", "cost_center": "CX-104"},
},
)
Comparison
Why not just go direct?
| Feature | Clara Router | Direct provider accounts | Global routers |
|---|---|---|---|
| Clara payment discount | 5% minimum | None | None |
| Billing in MXN, BRL and COP | Yes | Mostly USD | USD credits |
| Local tax invoices | CFDI, NFS-e, DIAN | Varies by provider | Foreign receipt |
| One API for all models | Yes | One per provider | Yes |
| Cost-optimized routing and fallbacks | Yes | Build it yourself | Varies |
| Routing informed by Spanish and Portuguese evals | Yes | No | No |
| Budgets by cost center, with approvals | Native in Clara | No | Key limits only |
| Accounting and ERP sync | Through Clara | Manual | Manual |
| Support in Spanish and Portuguese | Local teams | Limited | Mostly English |
Based on typical publicly available offerings. Check each provider for current details.
Security and privacy
Your prompts are your business
Built with the same care Clara applies to your company's money.
Never used for training
We don't train on your prompts or outputs, and we only route to provider endpoints whose terms say the same.
Zero-retention mode
Turn it on and prompt content is never stored, only the metadata needed for billing and reports.
PII redaction
Optionally mask identifiers such as CPF, CURP, RFC and card numbers before a request leaves for a provider.
Access control and audit
Single sign-on with Clara, roles per team, key rotation and an audit log of every administrative action.
Encryption everywhere
Traffic encrypted in transit and data encrypted at rest, with keys managed and rotated for you.
Data processing agreements
Contract terms designed around LGPD in Brazil, LFPDPPP in Mexico and Ley 1581 in Colombia.
Pricing
Pay for tokens, not subscriptions
Transparent per-token prices for every model, shown before you send a request. No minimums, and at least 5% off when you pay with Clara.
Clara customers
5% minimum discount
Everything in the router, billed through the Clara account you already have.
- All models and smart routing
- Billing in MXN, BRL, COP or USD
- Local tax invoices
- Budgets, approvals and cost centers
- At least 5% off optimized spend
High volume
More than 5% off
For teams with committed AI spend that want better terms and closer support.
- Everything for Clara customers
- Higher discount tiers for committed volume
- Dedicated capacity and custom rate limits
- Custom data processing terms
- Named solutions engineer
New to Clara
Open your account first
Clara Router is available to Clara customers. Opening an account is free and takes minutes.
- Corporate cards and payments in one place
- Spend management across the company
- Access to the router as soon as you're approved
FAQ
Questions, answered
What is Clara Router?
Clara Router is one OpenAI-compatible API for every leading model. It routes each request to the most efficient model that meets your quality bar, optimizes context and language handling, and shows the cost of every decision. Pay with Clara for at least 5% off the optimized spend. The public demo uses illustrative data.
How does the 5% Clara discount work?
Pay for your AI usage with Clara and get a minimum 5% discount on spend after optimization. Committed volume can qualify for better terms. The demo shows the estimate with and without paying with Clara.
Do optimization and the Clara discount stack?
Yes, one after the other. 40% optimization on $10,000 leaves $6,000, and a 5% Clara discount takes another $300 off, for a final cost of $5,700: 43% combined. Actual optimization depends on your workload.
Do I need to be a Clara customer?
Not to explore: the landing page and demo are open to everyone. Paying with Clara is what unlocks the discount on billed usage.
Which models are supported?
Models from OpenAI, Anthropic, Google, Meta, Mistral, DeepSeek, xAI, Cohere, Qwen and others, including text, vision and embedding models. New models are added as they launch, under the same key.
How hard is it to migrate?
If you use an OpenAI-compatible SDK, you change the base URL and the API key. Most teams send their first request in minutes and move production traffic gradually, one service at a time.
Will smart routing lower the quality of our answers?
Routing only considers models that meet the quality bar for the detected task, and you control that bar. For critical flows you can pin a specific model, and you can compare models on your own traffic before changing anything.
Which currencies and invoices are supported?
You can be billed in MXN, BRL, COP or USD. We issue CFDI in Mexico, nota fiscal in Brazil and factura electrónica in Colombia, attached automatically to each charge in Clara.
What happens to our data?
Prompts and outputs are never used for training. With zero-retention mode, content is not stored at all. You can also enable redaction of personal identifiers before requests are sent to providers.
Can we use our own provider contracts?
Bring-your-own-key routing is part of the product direction. The Clara discount applies to usage paid through Clara, not to spend billed directly by a provider. The demo does not connect provider accounts.
What happens if a provider goes down?
The router detects errors and rate limits and retries the request on an equivalent model from another provider, following the rules you define. Your application keeps responding.
See what smarter routing could save you
Compare models, try language-aware optimizations and see the Clara discount on a sample workload. No account required.