sashiko 0.3.3

Agentic code review system for Linux kernel
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18

[ai]
provider = "vllm"
model = "Qwen/Qwen3-8B"
max_input_tokens = 32768

[ai.vllm]
base_url = "http://localhost:8000"
# Should match the vLLM server's --max-model-len
context_window_size = 32768
# Leave max_tokens unset to let vLLM use the remaining context
#max_tokens = 4096
# Disable thinking for reasoning models such as Qwen3
enable_thinking = false
# Enforce JSON via guided decoding; leave disabled unless the backend supports it
#guided_json = true
# Forward tool definitions; requires a server started with
# --enable-auto-tool-choice and --tool-call-parser
#enable_tools = true