Summarise this 40-page supplier contract.
Response:Three renewal clauses auto-extend unless cancelled 90 days out. Penalty cap is 12% of annual value.
Built in El Salvador · Served in El Salvador
Local and Sovereign AI inference infrastructure
Your tools, your data, your models and your custom intelligence, with not a single request out of the country.
- Tier III certified
- Data never leaves SV
- OpenAI-compatible API
Inference cluster · sv-cscl-1Hover to replay
Reconcile these 1,204 ledger lines against the bank export.
Response:1,197 matched. 7 flagged: 4 duplicated references, 3 outside the posting window.
Draft the risk memo for the Q3 credit committee.
Response:Memo drafted. Exposure concentrated in two sectors; recommended provisioning attached.
Request pathyour networkSan Salvador · Tier IIIyour networkus-east-10 packets crossed the border
Trusted by teams building El Salvador's digital economy
Banco Cuscatlán
Grupo Agrisal
Superintendencia Financiera
Puerto de Acajutla
Aeroman
Grupo Calvo
Dirección de Innovación
Solutions
A supercomputer you rent by the token
Frontier AI models do not run on ordinary servers. They need racks of tightly coupled accelerators, high-bandwidth interconnects, and power and cooling envelopes that almost no building in the region can sustain. Until now, that meant every Salvadoran prompt had to travel abroad to find a machine big enough to answer it.
That machine is now here. Indigho operates it inside the only Tier III enterprise-grade data center in the country — minutes from your office instead of thousands of kilometres from your jurisdiction. You do not buy hardware, you do not size a cluster, and you do not sign a capacity commitment to get started. You send a request to a standard API and you pay for the tokens you used.
Pay per token
Serverless access to every model on the platform. No minimum, no idle cost, per-second metering, one invoice in USD.
Dedicated deployments
A model pinned to your own isolated accelerators, on demand. Predictable latency and throughput that nobody else shares.
Reserved capacity
Contracted quotas with guaranteed availability, so a regional demand spike never becomes your outage.
Fine-tuning & RL
Train on your own corpus. Every checkpoint deploys straight back onto the same inference platform — weights included.
- <15 ms
- Round trip, metro San Salvador
- Tier III
- Concurrently maintainable facility
- 100%
- Requests served in-country
- 1:1
- OpenAI-compatible endpoints
Models
Whoever's open model is best that week
A neutral capacity layer, not a bet on one lab. Prices are in USD per one million tokens, billed per second of use.
| Model | Input / 1M | Output / 1M | Context | Best at |
|---|---|---|---|---|
| Kimi K3FlagshipMoonshot AI | $3.00 | $15.00 | 1M tokens | Long-horizon agentic work, native vision |
| GLM 5.2Z.ai | $1.40 | $4.40 | 1M tokens | Project-scale software engineering |
| DeepSeek v4 ProBest valueDeepSeek | $0.44 | $0.87 | 1M tokens | Frontier reasoning at the lowest cost |
| Paddle OCR v3.7Per-page billingBaidu | On request | On request | Per page | Document extraction across 50 languages |
Cached input is billed at a reduced rate. Paddle OCR is billed per processed page rather than per token. Published rates are indicative and confirmed at contract; new open-weight releases are onboarded as they ship.
Why us
The new blue gold
For nearly three centuries, the most valuable blue on Earth came from a small strip of land in Central America. Salvadoran añil — indigo drawn from the jiquilite plant, xiquilitl in the Nahuat of our ancestors — dyed the uniforms of empires and the denim of revolutions. At its peak this small country was not a participant in the market for blue. It was the market.
Añil worked because of a simple pattern: take an abundant local resource, apply a refinement process nobody else had mastered at scale, and export the concentrated result to a world that could not get enough of it. Raw leaf in, pure pigment out. That pattern never died. It was waiting for a new resource.
Today that resource is computation, and the refinement process is inference. Raw data in, pure intelligence out.
- 01
Adoption is no longer a decision
Every process in your company that reads, writes, classifies or reconciles is being rebuilt around models right now — by your competitors if not by you. The open question stopped being whether to adopt AI. It is now only where the inference happens, and under whose law.
- 02
The obvious place to put it
El Salvador has exactly one enterprise-grade Tier III data center. Concurrently maintainable power and cooling, carrier-diverse fibre, physical access control, inside national jurisdiction. If sovereign inference is going to exist here, it belongs there. We put it there.
- 03
Intelligence is too strategic to rent from a landlord
Closed frontier APIs give you a capability you can never own. Prices move, models are deprecated under you, terms change, and your accumulated context lives in someone else's jurisdiction. We serve open-weight models. Your fine-tuned checkpoints are yours, portable, and inspectable — no lock-in on the one asset that will define your next decade.
- 04
Salvadoran engineering is more than enough
This platform was designed, deployed and is operated by a Salvadoran team. Distributed training, MoE serving, KV-cache routing, speculative decoding, autoscaled GPU scheduling — the state of the art is documented and our engineers work at it daily. The talent was never the constraint. The infrastructure was, and now it is not.
The añil trade ended when a synthetic dye was invented somewhere else. This time we are the ones doing the inventing.
Contact
Start with a conversation, not a credit card
Tell us what you are trying to run and where your data has to stay. An engineer — not a sales team — replies within one business day.
- hola@indigho.ai
- Phone
- +503 7795 1839
- Facility
- San Salvador, El Salvador
- Response
- Within 1 business day