loading…

𓃀 Settings

Self-calibration · model · prompt budget · platform keys · tier

← back to studio

Model

Open turbo health ↗

Claude Sonnet 4.5turbo · cloud
Anthropic Sonnet via the FastAPI turbo router. Default.
claude-sonnet-4-5-20250929
Claude Haiku 4.5fast · cloud
Lower-latency Haiku via the same turbo pool.
claude-haiku-4-5-20251001
NeterGaiiaM (alias)alias · server-routed
Server picks the best available cloud model under the alias.
netergaiiam-default
Local twin-pass (M1 / iPad)twin · sovereign
Two-pass F+M inference on local Llama 3.1. Requires the sovereign tarball.
local-twin

Prompt budget ratios

The four core layers (System, Instructions, Context, History) sum to 100. Sliders rebalance the others as you move one.

System15%

Persona, model spec, identity frame.

Instructions15%

Per-turn task instructions for the agent.

Context55%

RAG / tool / library context for THIS turn.

History15%

Prior turns in the thread.

Σ(system+instructions+context+history) = 100

Decoding controls

Max tokens4096

Hard cap on a single completion.

Temperature0.20

Sampling spread. 0.0 = deterministic; 1.0+ = expansive.

Platform keys & connected services

Anthropic / Claude (turbo pool)probing…
wings-reasonerprobing…

Platform keys live on the FastAPI server. Sovereign-tarball deploys can edit them via /root/.env.aamt-api on the VPS. The browser never holds a key.

Tier & turbo budget

Membership tierresolving…

Monthly turbo budget—

0 / 0 used this month · usage feed pending GET /api/v1/inference/turbo/usage