Providers

Home / Providers / Overview

Providers

CommitBrief supports eleven review backends. Eight call HTTPS APIs directly; three shell out to a host CLI tool. One additional mock provider is registered for tests and not user-relevant.

Name	Type	API key	Page
anthropic	HTTPS API	required	Anthropic
openai	HTTPS API	required	OpenAI
gemini	HTTPS API	required	Gemini
deepseek	HTTPS API (OpenAI-compatible)	required	OpenAI-compatible
mistral	HTTPS API (OpenAI-compatible)	required	OpenAI-compatible
cohere	HTTPS API (OpenAI-compatible)	required	OpenAI-compatible
openai-compatible	HTTPS API (any OpenAI-compatible endpoint)	optional	OpenAI-compatible
ollama	HTTPS API (local)	not needed	Ollama
claude-cli	host-CLI subprocess	not needed (uses Claude Code's auth)	CLI tools
gemini-cli	host-CLI subprocess	not needed (uses Gemini CLI's auth)	CLI tools
codex-cli	host-CLI subprocess	not needed (uses Codex CLI's auth)	CLI tools

Picking a provider

# Interactive wizard (asks which one + API key + model).
commitbrief setup

# Switch the active default later without re-entering keys.
commitbrief providers use openai

# Override for one invocation.
commitbrief --provider gemini --model gemini-3.1-flash-lite --staged

# Shorthand for the CLI-tool-backed providers.
commitbrief --cli claude --staged

Measured review quality

Two models have been measured on a corpus of 129 fixtures (96 with a planted bug, 33 clean) across Go, Java, JavaScript, PHP, Python and TypeScript, three runs each, measured 2026-10-26 (OpenAI) and 2026-10-31 (Gemini); COMMITBRIEF_EVAL_PROVIDER=<name> make eval-live reproduces them. A model passes if recall is at least 0.90, the false-positive rate on clean fixtures is at most 0.10, and recall in every language is at least 0.70.

Model	Recall	False-positive rate	Cost per review	Passes
Gemini gemini-3.8-flash	0.9731	0.0303	$0.006407	yes
OpenAI gpt-5.4-mini	0.9259	0.2929	$0.003674	no (false-positive rate)

Cost per review is the list rate with no cache discount; for Gemini it includes thinking tokens. Anthropic and every other provider are not measured yet, so none of them is ranked. commitbrief setup does not show these results and still labels every model "not measured".

API providers vs CLI-tool providers

Property	API providers	CLI-tool providers
Auth	API key in CommitBrief config	Host CLI's own auth (claude login, gemini auth)
Output shape	Structured findings JSON (schema v1)	Pre-formatted plain text from the host CLI
Cards / JSON / Markdown rendering	All three available	Plain-text emit only
--fail-on severity gate	Honored	Not applicable (no severity field)
Per-call cost	Billed via the provider	Counts against your host CLI subscription
--json / --markdown / --sarif allowed	Yes	No — rejected before any call
--output <file>	Yes	Yes (UC-07, since v0.9.2)

CLI-tool providers are a thin subprocess wrapper around the host's own CLI; CommitBrief feeds the combined prompt in and streams stdout out verbatim. Their Pricing() returns zero, so the cost preflight short-circuits silently, and no verbose footer is printed (plain-text output is streamed verbatim).

Provider name vs binary expectation

Provider	Expected binary on PATH	API endpoint
anthropic	—	api.anthropic.com
openai	—	api.openai.com
gemini	—	generativelanguage.googleapis.com
deepseek	—	api.deepseek.com
mistral	—	api.mistral.ai/v1
cohere	—	api.cohere.ai/compatibility/v1
openai-compatible	—	<base_url> (you set it)
ollama	ollama runtime running on base_url	<base_url> (default http://localhost:11434)
claude-cli	claude (Claude Code)	(subprocess, no HTTPS from CommitBrief)
gemini-cli	gemini (Google Gemini CLI)	(subprocess, no HTTPS from CommitBrief)
codex-cli	codex (OpenAI Codex CLI)	(subprocess, no HTTPS from CommitBrief)

For Ollama, the binary running the server is detected via the base_url (default http://localhost:11434). The Ollama daemon must be running.

Where models come from

For the eight API providers (Anthropic, OpenAI, Gemini, DeepSeek, Mistral, Cohere, openai-compatible, Ollama), the model is whatever you set in config (providers.<name>.model or --model). If unset, each provider falls back to its DefaultModel().

For CLI-tool providers, the host CLI manages model selection — the model config field is ignored.

See also

• Setup command — initial provider configuration.
• Providers command — list / switch / test providers.
• Configuration files — the YAML keys.

See also: Providers