CLI
Two commands. That is the whole tool.
The CLI reads LAZAPI_KEY from the environment, defaults to the gateway base URL, and prints raw JSON so it composes with jq.
| Command | What it does |
|---|---|
lazapi chat | Sends a prompt to a named model, streams or prints the reply |
lazapi bench | 20 samples against one model, prints p50 latency and tok/s |
Install
# pip pip install lazapi # or straight from the repo git clone https://lazarus-th.duckdns.org/lazapi.git cd lazapi && pip install -e .
Set the key
export LAZAPI_KEY=sk-... export LAZAPI_BASE=https://lazarus-th.duckdns.org/v1
Chat
lazapi chat --model zai-org/GLM-5.3-Flash "explain paged attention in two lines" # stream lazapi chat --model deepseek-ai/DeepSeek-V4-Flash-0731 --stream "count to 50" # raw JSON for jq lazapi chat --model MiniMaxAI/MiniMax-M2.7 --json "hi" | jq '.usage'
Benchmark a route
lazapi bench --model zai-org/GLM-5.3-Flash --n 20 --tokens 512 model zai-org/GLM-5.3-Flash samples 20/20 ok ttft p50 0.31s p95 0.48s tok/s 112 cost $0.0004 per call
Exit codes
0— the call succeeded and the body parsed1— transport failure, no response2— the gateway answered with an error object3— the model was not in the key's allow-list