typhon-stress-test

typhon-ask

The storm has passed. The data is cold. Now you bring it to the oracle and ask: what did I learn, and what should I do with it?

typhon-ask

Typhon sends your hardware profile and benchmark results to an LLM and streams back a personalized recommendation — optimal --ctx-size, suggested launch flags, and an interpretation of what the numbers reveal.

Configuration

The oracle is reached via environment variables:

Variable	Default	Description
`TYPHON_LLM_URL`	auto-detect	Base URL of the LLM server
`TYPHON_LLM_KEY`	`none`	API key — `none` for local servers
`TYPHON_LLM_MODEL`	`auto`	Model name — `auto` reads from the scan

By default — with no configuration — typhon-ask speaks to the same local server you just benchmarked. If you ran a trial against llama-server on port 8080, the oracle is already there. No key. No ceremony.

Calling the oracle

Default — the server you just measured:

typhon-ask

Ollama, specific model:

TYPHON_LLM_URL=http://localhost:11434 TYPHON_LLM_MODEL=llama3 typhon-ask

OpenAI:

TYPHON_LLM_URL=https://api.openai.com/v1 \
TYPHON_LLM_KEY=sk-... \
TYPHON_LLM_MODEL=gpt-4o \
typhon-ask

Any OpenAI-compatible endpoint:

TYPHON_LLM_URL=https://api.your-provider.com/v1 \
TYPHON_LLM_KEY=your-key \
TYPHON_LLM_MODEL=your-model \
typhon-ask

Who will answer

Any server that speaks the OpenAI Chat Completions protocol (POST /v1/chat/completions):

llama-server (llama.cpp) — default, already known from the scan
Ollama — set TYPHON_LLM_URL=http://localhost:11434 and TYPHON_LLM_MODEL=<name>
LM Studio — set TYPHON_LLM_URL=http://localhost:1234
vLLM — set TYPHON_LLM_URL=http://localhost:8000
OpenAI — set URL + key + model
Any OpenAI-compatible proxy or cloud oracle

What the oracle receives

The scroll sent to the LLM contains:

GPU name, VRAM, CPU, RAM
Model name and server type
Per-benchmark TPS, VRAM usage, and temperature at each context size
Stress test and memory wall results
Zeus extreme context results (128K / 256K), if you have run typhon-zeus at least once

No personal information, file paths, or hostnames are included.

REST API

typhon-ask is also available through the herald when typhon-api is running:

curl http://localhost:8000/ask

This site is open source. Improve this page.