Models

GDPR-compliant model matrix.

Every provider we offer is contractually covered for EU use. The orchestrator picks the right model for each task; you get per-request observability and per-org cost.

Model-agnostic

No vendor lock-in

Choose the best model for each task. Switch any time. Bring your own keys, run on-prem when sovereignty matters.

Provider	Modality	Context window	On-prem
Claude Anthropic	Text + vision	200K	—
GPT OpenAI	Text + vision	128K	—
Gemini Google	Text + vision	1M	—
Mistral Mistral AI	Text	128K
Llama Meta (open weights)	Text + vision	128K
Azure OpenAI Microsoft	Text + vision	128K	—

All models are GDPR-compliant via signed DPA and SCCs. Mistral is the only provider whose inference runs fully on EU infrastructure; open-weight models can run on-prem when sovereignty matters.

Switch model per channel, mix providers, run on-prem when sovereignty matters.

Routing

How the orchestrator chooses.

Each request is classified, then routed to the cheapest model that can handle it. You override per knowledge base, per channel, or per workflow.

1
Task class
Retrieval, summarisation, classification, code, vision — different models excel at different things.
2
Latency budget
Voice & chat get fast models. Background jobs (KB compile, weekly digests) can afford slower, cheaper ones.
3
On-prem constraint
If your KB is on-prem only, Mistral or Llama become the only candidates. No silent fallback to cloud.
4
Per-request audit
Every call logs the model, token counts, cost, and the routing rationale. Visible in the dashboard.

Pricing

Token cost is passed through transparently.

We don't mark up provider tokens. Your monthly bill shows a per-model breakdown — see the pricing page for exact rates and quotas, and the trust page for the legal framework.

See pricing See trust