// guide · cli reference

jointchiefs — the CLI.

Same engine as the MCP server. Streams live. Exits with meaningful codes. Pipes stdin. Emits JSON on request.

// commands

Commands.

`jointchiefs review <path>`

Runs a full multi-model review of the file or directory. Streams to stdout; every provider token lands live.

$ jointchiefs review src/auth.swift
$ jointchiefs review src/auth.swift --goal "security audit"
$ jointchiefs review src/auth.swift --context "prefer pure functions; Swift 6 concurrency"

`jointchiefs review --stdin`

Reads code from stdin. Great for piping a git diff in as a pre-commit gate.

$ git diff | jointchiefs review --stdin --goal "pre-commit check"
$ cat pr.patch | jointchiefs review --stdin --json > review.json

`jointchiefs models`

Lists every configured provider and its model. Use --test to probe each provider's API with a round-trip connection check.

$ jointchiefs models
$ jointchiefs models --test

// flags

Flags.

Flag	Description
`--goal <text>`	What you want the debate to focus on. "security audit", "performance review", "readability pass". Passed to every provider.
`--context <text>`	Free-form context — style rules, conventions, framework versions, caller behavior. Feeds the spokes alongside the code.
`--rounds <n>`	Max debate rounds. Default 5. Debate breaks early when positions converge.
`--timeout <sec>`	Per-provider request timeout in seconds. Default 120.
`--stdin`	Read code from stdin instead of a file argument.
`--format <fmt>`	`summary` (default), `json`, or `full`. `full` adds the transcript ID after the summary.
`--quiet`	Suppress streaming output. Only the final consensus prints when the review completes.

// exit codes

Exit codes.

Code	Meaning
`0`	Review completed successfully.
`1`	Review failed — every provider errored, no providers configured (no API keys found in environment or Keychain), or consensus could not be produced.
`64`	Usage error — missing file argument, empty stdin, malformed flag, or validation failure from `ArgumentParser`.

// configuration

Environment.

API keys resolve in order: environment variable first (CI escape hatch), then the Keychain via jointchiefs-keygetter. Model overrides always come from env vars.

`OPENAI_API_KEY`	Enables OpenAI as a spoke.
`GEMINI_API_KEY`	Enables Gemini as a spoke.
`GROK_API_KEY`	Enables xAI Grok as a spoke.
`ANTHROPIC_API_KEY`	Enables Claude — also the default moderator/decider.
`OLLAMA_ENABLED`	`1` to force-include / `0` to force-exclude the local Ollama general (overrides `StrategyConfig.ollama.enabled`).
`OLLAMA_MODEL`	Ollama model override. Default `llama3`.
`OPENAI_COMPATIBLE_BASE_URL`	Force-enable an OpenAI-compatible local server (LM Studio, Jan, llama.cpp-server, Msty, LocalAI). CI override for `StrategyConfig.openAICompatible`.
`OPENAI_COMPATIBLE_MODEL`	Model identifier as the local server exposes it.
`OPENAI_MODEL` etc.	Override the default model for any cloud provider (`GEMINI_MODEL`, `GROK_MODEL`, `ANTHROPIC_MODEL`).
`CONSENSUS_MODEL`	Override the Claude model used specifically for consensus synthesis.

Prefer the tool-call flow?

The MCP server exposes the same engine to any MCP-aware client — one JSON snippet to wire it up.

MCP setup guide Source on GitHub