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
19

# Direct OpenAI API backend: HTTP calls to https://api.openai.com/v1/chat/completions,
# no codex CLI. Billed per token against an API key rather than a Codex
# subscription.
#
# The key comes from the environment; there is no TOML field for it:
#   export OPENAI_API_KEY=sk-...     # LLM_API_KEY is also accepted
[ai]
provider = "openai"        # official API path; sends max_completion_tokens
model = "gpt-5.6-sol"      # the "gpt-5.6" alias resolves here; "gpt-5.6-terra" is cheaper

# The openai provider reads this table too. base_url and context_window_size
# are optional; set max_tokens as well. It is sent as max_completion_tokens,
# which on a reasoning model covers reasoning tokens as well as the reply, and
# it falls back to 4096 when this table is absent. That leaves a reasoning
# model little room for a reply once it has spent part of the budget thinking.
[ai.openai_compat]
max_tokens = 65536             # output cap, sent as max_completion_tokens
# base_url = "https://api.openai.com/v1/chat/completions"  # auto-derived from model; set only to override
# context_window_size = 128000   # generic fallback for gpt-5.x; raise to the model's real window