Installation

Quorum is a TypeScript program that runs under Bun, from a clone of its repository or as a standalone binary built from that clone. This page installs it, builds the binary, puts it on your $PATH, and checks that it works before any model is running.

There is no npm package yet: a TypeScript program that uses quorum as a library imports src/index.ts from the clone, as the Library API shows.

Prerequisites

• Bun 1.3.0 or newer. The engines field of package.json asks for it, and it installs the dependencies, runs the CLI and builds the binary.
• A model endpoint, for the reviews themselves. With no config file and no base URL, preset or protocol variable it is the Ollama on your machine, through its native API at http://localhost:11434, the address Ollama listens on, on 127.0.0.1, by default. QUORUM_RESEARCH_BASE_URL on its own points quorum at an OpenAI-compatible /chat/completions endpoint instead, a self-hosted vLLM, TGI or llama.cpp server or a hosted API, QUORUM_RESEARCH_PRESET names a vendor's preset, such as anthropic or vllm, and a config file can name an Anthropic, Gemini or Ollama endpoint as well: Configuration. The check at the end of this page needs none.
• Claude Code, for a claude-code provider only. It is the model's endpoint then: claude on your PATH, signed in or with a key, run once per request, headless and isolated: claudeCode. Quorum requires Claude Code 2.1.311 or newer.

From source

git clone https://github.com/this-is-tobi/quorum.git
cd quorum
bun install

bun install installs quorum's two runtime dependencies, zod, which validates every model reply, and yaml, which parses a YAML config file that loadConfig reads, and the development tools.

From the root of the clone, bun run start runs the CLI: bun run start ./my.diff. Bun first prints the script it runs, $ bun run ./src/cli/bin.ts, on stderr, and bun run --silent start leaves that line out.

A standalone binary

bun run compile
install -m 0755 ./bin/quorum /usr/local/bin/quorum

bun run compile writes ./bin/quorum: one executable with the Bun runtime inside, for the system you build it on, which runs without Bun and without the clone. The second line copies it onto your $PATH (with sudo where /usr/local/bin is not yours; any directory on your $PATH will do).

The binary reads no .env and no bunfig.toml from the directory it runs in, which a Bun-compiled binary does by default. Bun itself reads both from its working directory, so bun run start reads the clone's own, a bunfig.toml can name code to run before quorum starts, and a .env can set any QUORUM_* variable, QUORUM_CONFIG included, which names the config file quorum reads. In a directory you do not control, such as a pull request's checkout, run the binary, or Node as node /path/to/quorum/src/cli/bin.ts: Nothing read from the working directory.

The pages that follow write the command as quorum, which is that binary. Without it, read quorum as bun run start run from the clone, except after a --: the binary and Node read quorum -- --help as a file named --help, and Bun itself drops the first --, so bun run start -- --help prints the help.

Check that it works

With no diff and stdin a terminal, quorum prints its usage and exits 2 instead of waiting for input:

quorum
echo $?

Error: no diff given
Usage: quorum [options] [diff-file]
       git diff | quorum [options]

Options:
  -c, --config <path>     Read this config file, over QUORUM_CONFIG
  -p, --profile <id>      Run only this profile; repeat it to run several
      --allow-incomplete  Exit 0 instead of 4 when the review is incomplete, except for a request too large for its model
      --print-config      Print the resolved config, keys redacted, and exit
  -h, --help              Print this help and exit
  -v, --version           Print the version and exit
2

An empty diff on stdin, with a research model and its context size named, is reviewed without calling any model, so this works before you have one running. The JSON result goes to stdout, sent to /dev/null here:

printf '' | QUORUM_RESEARCH_MODEL=qwen3.6:27b QUORUM_RESEARCH_CONTEXT_TOKENS=32768 quorum > /dev/null
echo $?

Research (qwen3.6:27b @ http://localhost:11434)...
Panel will use qwen3.6:27b @ http://localhost:11434
empty diff: nothing to review
0 kept, 0 discarded, 0 unreviewed by panel
0

The first two lines name the model of each role and the origin of the endpoint it would call. QUORUM_RESEARCH_CONTEXT_TOKENS is the model's context window in tokens, which quorum checks every request against. With no endpoint variable set, the endpoint is the local Ollama's native API, which takes num_ctx, truncate: false and shift: false, so the server is told the context size and refuses an overflow, where Ollama's OpenAI-compatible /v1 would cut the prompt in silence (refusing needs Ollama 0.12.6 or later): Layering. Without QUORUM_RESEARCH_MODEL, the same command prints Error: no model configured: set QUORUM_RESEARCH_MODEL, or name a config file that sets one and exits 2: a research model must be named even for an empty diff. With the model but without QUORUM_RESEARCH_CONTEXT_TOKENS, it prints Error: no context size configured: set QUORUM_RESEARCH_CONTEXT_TOKENS, or name a config file that sets contextTokens and exits 2 too, for the same reason.

Next

Quick start — point it at a model, review a diff, and read the progress, the result and the exit code.