AI routing & spend

Run the model you're allowed to run

Point EstiWright at the local model your security team approved. On-prem mode uses it for every task.

EstiWright — Local model
llama-3.1-8bqwen2.5-14bmistral-nemophi-4
On-premModels

A hand-built mock in the site theme, modelled on the real screen and populated with our demo tenant's data — not a screenshot. Hover to pause.

More in AI routing & spend

Keep going

EstiWright — AI & Models · llm_mode
cloud routingon-prem edition — zero egress

Flip to the on-prem edition

On the on-prem edition, one switch moves every AI call in-network — no prompt, document or drawing leaves your own infrastructure.

EstiWright — Routing
TaskModel
extract_requirementslocal · cheap-fast
draft_compliancecloud · premium-reasoning
draft_rfilocal · cheap-fast
translatelocal · cheap-fast

Route each task differently

Send extraction to a local model and only the hardest reasoning to the cloud — per task, not all-or-nothing.

EstiWright — Settings · API keys
provider: openai
source: environment
stored_in_database: false

Your keys stay out of the database

Provider keys are read from your environment or secret store and never written to the application database.

See this on your own tender.

We'll run your RFP through the same pipeline and show you the output.