From 51b2621d58633aa5c0f5cc7b64616d70d41acc91 Mon Sep 17 00:00:00 2001 From: Paul Buetow Date: Sat, 27 Sep 2025 02:19:22 +0300 Subject: Support multi-provider fan-out across CLI and completions --- docs/configuration.md | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) (limited to 'docs/configuration.md') diff --git a/docs/configuration.md b/docs/configuration.md index 6db7a27..6e42172 100644 --- a/docs/configuration.md +++ b/docs/configuration.md @@ -27,7 +27,7 @@ Environment overrides Per-surface models - Use the `[models]` table in `config.toml` to tailor individual entry points (completion, code actions, chat, CLI) without changing the global provider default. -- Each key accepts either a string (shortcut) or a table with `model` / `temperature` fields, e.g.: +- Each key accepts either a string (shortcut) or one or more tables with `model` / `temperature` fields, e.g.: ```toml [models] @@ -43,6 +43,8 @@ Per-surface models provider = "openai" ``` +- Repeating the table (`[[models.]]`) configures multiple provider/model pairs. Completion requests and the Hexai CLI fan out to every configured entry concurrently and label the responses with `provider:model`. Code actions continue to use the first entry only; any extra [[models.code_action]] tables are ignored at runtime and the loader logs a warning so you know an additional entry was skipped. + - When a per-surface value is omitted, Hexai falls back to the provider’s configured default. Temperatures inherit from `coding_temperature` unless explicitly set, and OpenAI `gpt-5*` models automatically raise an unspecified coding temperature to `1.0` for exploratory behavior. Provider overrides support `"openai"`, `"copilot"`, or `"ollama"` and read the matching credential variables. Runtime reloads -- cgit v1.2.3