Local dashboard. Every saved chat is a callable URL.
api.openai.com + sk-… keys
Your own server. No keys — or Interdata Lab when you scale.
OpenAI's model menu
Any local model — Llama, Qwen, Mistral, 1-bit Bonsai.
Per-token billing
$0 on your hardware. Free on Interdata Lab in alpha.
Drop-in OpenAI replacement
# Configure the bot once in the UI — system prompt, model, sampling.# Then call it from anything that speaks the OpenAI SDK.from openai importOpenAI
client =OpenAI(base_url="http://localhost:8095/v1", api_key="x")
client.chat.completions.create(
model="3", # your saved bot's ID
messages=[{"role": "user", "content": "Triage this ticket..."}],
)
Every saved chat is an addressable URL. One line of code swaps OpenAI
for a model running on your hardware — or, when you scale, on Interdata Lab.
Open Source · MIT
MiniClosedAI — Your Own OpenAI Playground
Save a system prompt, get a stable /v1/chat/completions endpoint — like OpenAI's Playground, but you own it
Each recipe is a battle-tested system prompt, sampling config, and worked few-shot examples.
Customize for your domain, save the chat, and you have a callable URL — no code generation, no boilerplate.
Support Ticket Router
Classify intent, assign team, score urgency, extract entities. Returns structured JSON ready for your CRM.
Inbound Lead Qualifier
Score B2B leads 0–100, detect role and budget signals, route to the right CRM stage.
RAG Query Router
Sub-second pre-classifier — decide cache, fast LLM, light RAG, or deep RAG per query.
JSON Extractor
Universal information-extraction microservice. Strict schema, deterministic, no prose.
Doctor's Office Bot
Conversational front desk. Books appointments, detects red flags, emits structured action JSON.
Reservations Bots
Restaurant, hotel, dental — multi-turn host-stand flows with required-fields gates and JSON action emit.
Why this stack
60s
Prompt to callable API
5 deps
vs. hundreds in LangChain
Drop-in
OpenAI SDK compatible
CSV
Built-in fine-tuning export
Zero
Data leaves your machine
$0
Free in alpha — and forever for open source
Clean compute · Powered by Paraguay
Cleaner, cheaper compute — and a path to your own
Today, Interdata already runs on existing datacenters in Paraguay — one of
the cleanest grids on Earth, powered almost entirely by hydroelectric dams like Itaipú
(≈14 GW). The rest of our compute runs on GPUs we rent by the hour. Either way,
MiniClosedAI runs a lean stack — small models on right-sized hardware — so every bot burns
far less energy than a hosted frontier model. Cheaper compute is cleaner compute, wherever
the GPU lives. Looking ahead, our plan is to scale the same economics down
to a single house: your own always-on, self-serving macro-datacenter — and we're building
the repeatable, low-cost path to get there.
How clean is the compute?Today
Grams of CO₂ per kWh (illustrative — IEA / Our World in Data ranges). Paraguay's
near-100% hydro grid is roughly 20× cleaner than the global average — and a lean
MiniClosedAI stack draws less of it, so the compute we run there is almost carbon-free.
The house as a self-serving datacenterThe plan
The vision: as solar and battery prices keep falling and clean hydro power stays
cheap, the economics that make Interdata's datacenter work scale down to a single
home — turning any house into its own always-on AI compute node.
The repeatable path to your own nodeThe plan
The vision only matters if it's buildable. Interdata's goal is to make it repeatable —
sourcing, and where it pays off manufacturing, the best-cost clean-energy and compute
hardware, so you can self-build a datacenter node at home for a fraction of today's cost.
1
Source the cheapest clean resources
We aggregate — and where it pays off, help manufacture — the best-cost solar panels, batteries, and GPUs.
2
Build from a reference design
A repeatable blueprint to self-build a home compute node, powered by your own solar, storage, and clean grid.
3
Run it self-serving
Power your own always-on AI, and put spare compute to work instead of paying a hosted provider by the token.
Concept · Solar + battery home
A solar-powered home in Ciudad del Este
A rough sketch of where the repeatable path leads: a single house in Ciudad del Este,
Paraguay, running its own compute on rooftop solar and battery storage — one self-serving
node, powered the same clean way as the datacenters we already use today.
A solar + battery home — Ciudad del Este, ParaguayThe plan
Concept sketch of a single self-serving node: rooftop solar — backed by Paraguay's
hydroelectric grid — charges a battery that powers an always-on macro datacenter, the
same clean setup as the datacenters Interdata runs on today.
One network · Many datacenters
Interdata is a network, not a datacenter
Interdata isn't one datacenter — it's a growing network of interconnected macro-datacenters,
some on Paraguay's hydro grid, others on rented GPUs. Connect MiniClosedAI to Interdata and
you reach every model hosted anywhere in the network, through one endpoint — no matter which
datacenter actually runs it.
One connection, every model in the network
The circled nodes run in Paraguay on the hydro grid; the others are rented GPUs — RunPod,
AWS SageMaker, and more. Every node interconnects, so a single MiniClosedAI client can call
any model across the whole network — Llama, Qwen, Mistral, Gemma4, 1-bit Bonsai — from one endpoint.
Free for alpha design partners
MiniClosedAI is open source under MIT — clone it, fork it, run it on a laptop today.
Interdata Lab is free GPU compute for the founders and engineers helping us shape it.
No billing while you're building with us.