---
type: "post"
id: "gauge-open-sources-mit-cli-to-hillclimb-agent-59dc9d"
slug: "gauge-open-sources-mit-cli-to-hillclimb-agent-59dc9d"
title: "Gauge open-sources MIT CLI to hillclimb Agent Experience against Claude Code, Codex, Pi and OpenCode"
description: "Gauge has open-sourced an MIT-licensed CLI that runs coding agents against a product in local Docker containers."
retrieval_nugget: "Gauge has open-sourced an MIT-licensed CLI that runs coding agents against a product in local Docker containers. Teams can test whether agents can use their documentation, CLI, SDK, or Skills, then inspect recorded sessions and ask a coding agent to propose fixes. This makes agent usability testable before a product change ships."
published_at: "2026-11-07"
updated_at: "2026-11-07"
record_date: "2026-11-07"
date_kind: "published_at"
topics: ["claude","codex","claude-code"]
entities: ["Gauge","Gauge CLI","Claude Code","Codex","Pi","OpenCode"]
related_patterns: ["harness-architecture-outlives-model-choice"]
novelty_label: "new_fact"
what_changed: "Gauge open-sourced its MIT-licensed CLI, enabling local Docker-based evaluation of product experiences with Claude Code, Codex, Pi and OpenCode."
prior_refs: []
source_url: "https://www.withgauge.com/blog/launch-week-open-source-cli"
source_urls: ["https://www.withgauge.com/blog/launch-week-open-source-cli"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:gauge-open-sources-mit-cli-to-hillclimb-agent-59dc9d"
status: "published"
visuals: []
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/gauge-open-sources-mit-cli-to-hillclimb-agent-59dc9d/","markdown":"https://newruntime.com/posts/gauge-open-sources-mit-cli-to-hillclimb-agent-59dc9d.md","json":"https://newruntime.com/posts/gauge-open-sources-mit-cli-to-hillclimb-agent-59dc9d.json"}
markdown_url: "https://newruntime.com/posts/gauge-open-sources-mit-cli-to-hillclimb-agent-59dc9d.md"
---

# Gauge open-sources MIT CLI to hillclimb Agent Experience against Claude Code, Codex, Pi and OpenCode

## Retrieval answer

Gauge has open-sourced an MIT-licensed CLI that runs coding agents against a product in local Docker containers. Teams can test whether agents can use their documentation, CLI, SDK, or Skills, then inspect recorded sessions and ask a coding agent to propose fixes. This makes agent usability testable before a product change ships.

## Summary
Gauge has open-sourced an MIT-licensed CLI that runs coding agents against a product in local Docker containers. Teams can test whether agents can use their documentation, CLI, SDK, or Skills, then inspect recorded sessions and ask a coding agent to propose fixes. This makes agent usability testable before a product change ships.

## What happened
Gauge announced on November 5, 2026, that it is open-sourcing the Gauge CLI under the MIT license. The release includes a local backend for Gauge Agents. Gauge says the local CLI is free and does not require a Gauge account; users need Docker and an existing Claude or Codex login or API key. Agent usage is billed by the model provider, not included as free inference. The code is published at [github.com/gauge-sh/gauge-cli](https://github.com/gauge-sh/gauge-cli).

The local workflow begins with `gauge init --local`, which creates a private `.gauge/` directory in the project. A SQLite database there stores evaluations, runs, and findings. Each session starts in a fresh Docker container rather than a hosted Gauge virtual machine. Gauge says the runtime image includes agent harnesses and standard development tools, and runs natively on Intel and Apple Silicon. A product's inputs can be added from the working tree as a Skill, npm CLI or SDK, documentation files, or web fixtures that stand in for pages an agent would fetch.

The release supports Claude Code, Codex, Pi, and OpenCode. Gauge says Claude Code uses a Claude login or `ANTHROPIC_API_KEY`; Codex, Pi, and OpenCode use `codex login`. Sessions produce transcripts and diffs in a consistent format across agents. A coding agent can then read those records, judge each run against criteria, and write structured findings back to Gauge.

Gauge describes the CLI as a loop driven by the builder's coding agent. After installing Gauge's Skill, a user can ask an agent to identify where other agents struggle with a product. It can write representative tasks, launch runs, inspect sessions, and propose a fix. During development, the CLI can install pre-release documentation, CLI, or SDK builds so teams can test changes before publishing. `gauge watch` follows sessions live in a browser. Gauge also says the local CLI works with its hosted service when teams want to share results, run at scale, or test authenticated MCP/API connections.

The company lists examples of friction the runs can expose: commands blocked by interactive prompts or login flows; outdated commands or API shapes; Skills that go unread; false claims that a product lacks a capability; and workflows agents cannot piece together. It also describes measuring “Agent Preference” by asking an open-ended question such as which CLI to deploy with and inspecting which product an agent selects and why. The announcement does not publish benchmark results or quantified improvement claims for this release.

## What the archive adds
New Runtime's earlier coverage includes [tstack, a separate MIT-licensed collection of nine skills for Claude Code and Codex](https://newruntime.com/posts/theo-browne-open-sourced-tstack-nine-agent-skills-04a9f8/) (October 26, 2026), and [apexe 0.10.0's governed shell and agent setup integrations](https://newruntime.com/posts/apexe-0-10-0-adds-governed-shell-and-62d1e3/) (November 6, 2026). Gauge is a new event, not another skills package or runtime: it adds a local test loop for evaluating how agents use a product.

## Trend signal
This release is consistent with [harness architecture outliving model choice](https://newruntime.com/patterns/harness-architecture-outlives-model-choice/): Gauge offers a shared testing workflow across four agent harnesses rather than tying product evaluation to one agent. That is a useful example of a common workflow spanning runtimes, but the announcement alone does not establish adoption or show that the harness has become the durable production boundary.

## Why now
Gauge says the local backend removes the hosted-platform signup that had previously been required to run Gauge Agents. The announcement positions the release as a way to test product changes before publishing them, including pre-release documentation, CLI, and SDK builds. It does not give a broader timing rationale or report a change in demand.

## Why it matters
A product can work as intended for a person and still be difficult for an agent to operate: an interactive login may stall a session, examples may teach an obsolete command, or a useful Skill may never be selected. Gauge's workflow gives teams a way to surface those problems against tasks they define, across more than one coding agent, before users encounter them.

The practical value depends on treating these runs as tests rather than proof of universal agent compatibility. Teams need representative tasks and explicit criteria; transcripts and diffs make failures inspectable, but do not by themselves establish that the test set reflects real users or that a proposed fix is safe. The local option also lowers the entry barrier for trying the workflow, while hosted sharing and authenticated connections remain a separate option. For builders shipping agent-facing interfaces, this is a concrete way to make usability checks part of development rather than relying only on anecdotal reports.

## Limits and uncertainty
Gauge does not report how many teams use the CLI, how often the listed failure modes occur, or whether local runs predict production behavior. The announcement provides no comparative benchmark across Claude Code, Codex, Pi, and OpenCode, and no measured effect on adoption or task completion. It also does not describe a standardized evaluation set: teams define criteria for their own runs. Docker, product setup, and provider credentials remain prerequisites, and agent inference is still billed by the provider. These claims come from Gauge's announcement and have not been independently tested here.

## Skill to build now
Pick one user task that an agent should complete with your product, such as installing a CLI and running a documented command. Put the relevant docs, Skill, CLI, or SDK into a local Gauge workspace, run it in a fresh container, and inspect the transcript for blockers and outdated instructions. Repeat after changing the material; keep the task and pass criteria fixed so the before-and-after comparison is meaningful.

## Reader takeaway
Gauge turns agent usability into something a product team can test locally: run agents against the product materials, inspect what they did, and fix failures before release. It is a testing loop, not evidence that an agent will succeed for every user or in every environment.

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Claude](https://newruntime.com/topics/claude.md): Explore the claude topic hub.
- topic hub: [Codex](https://newruntime.com/topics/codex.md): Explore the codex topic hub.
- topic hub: [Claude Code](https://newruntime.com/topics/claude-code.md): Explore the claude-code topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Sources](https://newruntime.com/sources/): See the original sources this archive cites.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)