Your AI. Your keys.
Works offline.
The admin picks the engine. Native local models by default, running with the internet unplugged. Your own cloud key when you want more power, with automatic fallback to native the instant the cloud fails. The intelligence lives in your building, on your terms.
Native by default · your key optional · fallback always on
The problem with renting
Rented intelligence can be switched off.
Almost every AI feature you have seen is a tenant's deal: it runs on someone else's servers, under someone else's pricing, reachable only while the line stays up. A price hike, a revoked key, an outage or a border is all it takes. Own the engine and none of that can reach you.
- Their pricing, changes without you
- Their key, revocable at any time
- Their servers, your data over there
- Line drops? Everything stops
- No meter, native calls are free
- Your keys, encrypted, capped, yours
- Your servers, data never leaves
- Line drops? It keeps thinking
Why own it
The intelligence lives in your building, on your terms.
Air-gapped safe
Behind a DMZ or fully off the grid, the AI still works. Your data never leaves the room it is in.
Zero token bills
Native models cost nothing per call. Use AI as much as you like without watching a meter tick.
One setting, everywhere
Pick the engine once in global settings. Every feature that uses AI inherits it. No wiring, no repeats.
One setting, everywhere
One switch. Native by default, your key when you want it.
Pick the engine once in global settings and every feature inherits it, no wiring, no repeats. Point it at a cloud provider with your own key for more power, and the moment that key caps out or the line drops, it falls back to native. Nobody gets paged.
Every AI feature inherits the setting:
The actual setting
Downstream
Everything downstream just calls it.
Rewrite, translate, chatbot, journeys, campaign copy, RAG: none of them know or care which engine sits behind the one interface. Change the engine and they all follow, together, instantly.
The guard that never phones home
Every message checked. Spam flagged, not blocked.
Spam-checking is the one job too important to outsource. It runs native, on the message hot path: fastText and rules checking every message in about a tenth of a millisecond, flagging what looks like spam without ever blocking a send or making a network call.
Flag, don't block: you keep the final call, and the check never leaves the box.
Under the hood
For the engineer in the room.
Present, not primary: the specifics a technical evaluator will want to confirm.
One internal interface, OpenAI-compatible
Built to the /v1/chat-completions shape, so native, OpenAI and most providers work through the same contract. One extra adapter covers Anthropic.
Embeddings are a separate setting
The embeddings model is chosen independently from the chat model: swap either without touching the other.
Spam is native, always
fastText and rules, on the hot path, flag-don't-block. Never an LLM, never a network hop.
Keys are encrypted at rest
Third-party keys are encrypted, redacted in logs, and capped. Set a hard cost cap per engine: hit it, and it falls back to native.
Assistive, not autonomous
Campaigns stay human-approved. RAG answers carry a “not in the documentation” guardrail instead of inventing.
Runs on commodity hardware
Native serves through llama.cpp / Ollama (e.g. Phi-4-mini) on plain CPU: built for offline, edge and low-concurrency boxes.
Before you ask
What hardware does native AI need?
Native runs on commodity server hardware: no GPU farm required for the assistive tasks (rewrite, translate, chatbot, spam). Sizing guidance is in the platform docs.
Which cloud providers can I use?
Any OpenAI-compatible provider through the standard interface, plus Anthropic via its adapter. You bring the key; you stay in control of the account and the spend.
What happens when a key hits its cost cap?
The engine falls back to native automatically. The feature keeps working. You get your data, never a broken screen.
Can the AI leak my data to a vendor?
On native, nothing leaves your servers. On cloud, only what you choose to send goes out, under your own account, and you can run fully native if you never want that.
Own the intelligence.
Your AI, your keys, on servers you own, that keeps thinking with the world switched off.