There was a time when intelligence had no owner.

Age Antique is an AI agent installed on your premises and built on an open or open-weight model, picked from the family that fits your task. You pay for the install once, then a monthly cost close to zero. Your data never leaves and you depend on no one. A legend to tell it, a method to deliver it.

WhatsApp
Timeline3 to 6 weeks
PaymentOnce, at installation
SubscriptionNone
I
Chapter one

The forgotten century.

Long before data centers, a civilisation whose archives were lost had solved a question we are rediscovering: how to put an intelligence to work without renting it from a distant oracle.

Every city kept its own, in stone, sheltered. It learned from local texts, answered in the local tongue, and owed nothing to anyone.

It is a legend. We tell it because it describes, word for word, what we install for our clients.
II
Chapter two

What the pyramids kept.

What the builders had understood fits in three principles. We have not found better ones.

I
Intelligence lives where the data is.

No trip to a foreign server. The model runs next to your files, inside your network, and keeps going without a connection.

II
It depends on no trade route.

No API key, no contract someone else can cancel, reprice or read. The day you stop paying, nothing switches off.

III
It can be handed down, repaired, improved.

Everything is documented and open: your team can understand it, change it, and swap the model when a better one appears.

III
Chapter three

The detour through the oracle.

Then came the central oracle. For a few years, almost all the world's intelligence passed through three or four companies. Every question sent away. Every answer billed by the word. Every business tied to a contract it does not control.

Convenient, at first. Then expensive. Then risky: your contracts, your customers, your prices, read by a machine that is not yours.

The oracle
Closed models, billed per use
  • Billed every month, per seat or per token
  • Your data flows through the vendor
  • Price grows with usage
  • Cut off the day you stop paying
Age Antique
Open model, on your server
  • Paid once, at installation
  • Your data stays inside your walls
  • Monthly cost close to zero, whatever the usage
  • Yours forever, even offline
IV
Chapter four

The return.

We travel back in time to reach that future.

Open models have caught up with the oracles on the work an agent actually does: reading a letter, sorting a request, extracting an amount, finding the right document, drafting a reply. Open models, freely published, runnable on an office machine, in every family: text, vision, speech, documents, forecasting.

Age Antique installs them on your premises, plugs them into your documents and tools, and hands you the keys. We come back when a genuinely better model appears. Not before.

V
What you get

One agent, its documents, its tools, and the keys.

  • The model chosen for your hardware, from 8 GB of RAM
  • Installed on your server or VM, works offline
  • RAG on your documents with PostgreSQL and pgvector
  • Connected to your tools: CRM, ERP, messaging
  • Documentation and training so you stay autonomous
  • A planned upgrade when a better open model ships
Starter8 GB RAM, CPU onlyBusiness16 to 32 GB RAMEnterpriseGPU recommended
Textsmall model, 1 to 4 Bmedium model, 7 to 14 Blarge model, 30 B and up
Searchmultilingual embeddingsembeddings + rerankingembeddings + reranking
Vision / OCRsmall vision-languagemedium vision-languagelarge vision-language
Voicetranscription, synthesistranscription, synthesis, diarizationreal time, diarization
StorePostgreSQL + pgvectorPostgreSQL + pgvectorpgvector, Qdrant

Everything is free and open source. Real throughput (tokens per second) is measured on your machine and written into the proposal.

VI
The method

Six weeks, at most.

  1. Week 1
    Hardware and use-case audit

    We measure your machine, list the documents and tools involved, and pick with you the three tasks the agent takes over first.

  2. Weeks 1 to 2
    Model choice and throughput tests

    Three open candidates tested on your real documents. You see the answers, response times and memory usage before deciding.

  3. Weeks 2 to 4
    Install, RAG and connections

    Model, vector store and agent installed inside your network. Hooked up to the chosen CRM, ERP or messaging.

  4. Weeks 4 to 5
    Acceptance with your teams

    Users test on their real cases. We fix prompts, rules and access rights.

  5. Weeks 5 to 6
    Training and handover

    Full documentation, training session, upgrade procedure. You are autonomous: the contract ends there.

VII
Questions

What we get asked before signing.

What if a better model ships in six months?
Planned, and the only reason to see us again. The model-swap procedure is part of the documentation; do it yourselves or hand it to us, quoted, with no subscription.
What machine do I need?
From 8 GB of RAM on CPU for specialised tasks (sorting, extracting, answering on your documents). 16 to 32 GB for team use. A GPU only if you want the largest models. We measure before we promise.
What about voice and phone?
The honest exception: real-time conversation needs a latency local models do not reach yet. If you need it, you pay the speech provider directly, with no margin, and we tell you before you sign.
What happens if Wikolabs disappears?
Nothing switches off. The model is public, the code and configuration are on your side, so is the documentation. That is precisely the point of the offer.
How much does it cost?
A fixed price, announced after the week-1 audit, paid once. No licence, no monthly fee, no billing per token.

Bring your intelligence home.

Thirty minutes to measure your machine, list your documents and tell you honestly what a sovereign agent can take over for you.

WhatsApp
+261 38 66 261 00team@wikolabs.com