Short, focused tasks for one person.
CaptureExtractAnswer
Lightweight assistant lane
Private AI for real work
Private AI,
sized to the work.
Keep sensitive work close. Start with the smallest profile that feels useful, then scale when the workload earns it.
Private by design6 profiles
More room for context.Every step keeps the same private boundary.
Your prompts stay local.Privacy
10+ output tokens/sec.Warm, single stream*
Scale when work grows.Fit
01 / The capacity map
Six ways to keep
work close.
Choose by the shape of the work, not a badge.
Edge
02Structured work close to the source.
RouteSearchSummarise
Everyday assistant lane
Studio
03Content, code, and internal documents.
DraftCodeRetrieve
Strong general work lane
Pro
04More nuance, steadier turnaround.
ReviewAnalyseBuild
Complex work lane
Max
05Large working sets and long briefs.
ResearchSynthesiseReview
Wide-context lane
Workstation
06Shared knowledge work for demanding teams.
AgentsKnowledgeOperate
High-capacity lane
Capability lanes are workload heuristics, not one-to-one hosted model claims.
A practical floor
10+
tokens/sec
* single active stream, warm decode, fixed prompt mix. Final acceptance is validated on the chosen machine.
02 / How we size it
Work first.
Machine second.
Bring one real workflow. We map the boundary, context, and response standard.
- 01Workload
Read, write, retrieve, classify.
- 02Boundary
What must stay close?
- 03Shape
One person, team, or service?
- 04Validate
Test real prompts before sign-off.
03 / Clear boundaries
Private,
with a plan.
Local reduces data exposure and recurring usage costs.
Access controls, updates, monitoring, and a human owner still matter.
Ready when the work is clear
Tell us what should stay close.
Share the workflow. We’ll help you find the smallest setup that earns its place.