AI News HubLIVE
站内改写3 min read

Show HN: ctx is now open-source, a hackable desktop workbench for coding agents

ctx is an open-source desktop workbench for coding agents like Claude Code and Cursor. It provides isolated containers, task management, merge queues, and a hackable environment for agent-based development.

SourceHacker News AIAuthor: ripped_britches

Notifications You must be signed in to change notification settings

Fork 5

Star 80

BranchesTags

Open more actions menu

Folders and files

NameName

Last commit message

Last commit date

Latest commit

History

6 Commits

6 Commits

.github

.github

assets

assets

core

core

docs

docs

.bazelignore

.bazelignore

.bazelrc

.bazelrc

.bazelversion

.bazelversion

.dockerignore

.dockerignore

.gitignore

.gitignore

BUILD.bazel

BUILD.bazel

CONTRIBUTING.md

CONTRIBUTING.md

LICENSE.md

LICENSE.md

MODULE.bazel

MODULE.bazel

MODULE.bazel.lock

MODULE.bazel.lock

README.md

README.md

SECURITY.md

SECURITY.md

user.bazelrc.example

user.bazelrc.example

Repository files navigation

curl -fsSL https://ctx.rs/install | sh

Platform support: macOS and Linux today. Windows is on the roadmap.

Website: https://ctx.rs

Blog: https://ctx.rs/blog

Install guide: https://ctx.rs/getting-started/install-and-launch/

What ctx helps you do

Use Claude Code, Codex, Cursor, and other coding agents in a desktop workbench instead of flickering terminal panes

Run agents in isolated containers with explicit disk and network controls

Let agents run in yolo mode inside explicit disk and network boundaries instead of choosing between approval prompts and full access to your shell, files, and network

Keep tasks, sessions, diffs, transcripts, and artifacts in one review surface

Run work locally or on remote machines you control

Keep parallel tasks isolated in separate worktrees and land them cleanly with the agent merge queue

Why ctx exists

A bare diff is not enough: ctx keeps the prompt, transcript, commands, artifacts, and worktree state that produced it

Parallel agents need isolated worktrees and a sane way to land branches without manual branch juggling

Agent sessions should be durable and inspectable instead of trapped in one provider's app state

You should be able to change harnesses and models without rebuilding your whole workflow

Teams can add stronger runtime, review, and provenance controls later without forcing everyone into one agent

How is this different from Codex app?

Codex app, Cursor, and Antigravity are first-party environments for their own agent stacks. ctx is an open-source, local-first workbench around the agent harness you use today and the ones you may want to try later.

If one vendor's app is already your whole workflow, you may prefer that app. ctx is for engineers who want the workbench to stay stable as agents, models, and harnesses change.

The difference is the layer ctx cares about. ctx focuses on task state, worktrees, transcripts, artifacts, diffs, containers, remote machines, review, and landing branches. The workbench itself is open source, so you can inspect it, modify it, script it, and keep your workflow independent of any single agent provider.

The Pi ethos, one layer up

Pi makes the case that an agent harness should be yours: inspectable, adaptable, built from primitives, and shaped around your workflow instead of sealed behind a vendor product.

ctx is trying to bring that same ethos to the Agentic Development Environment. The ADE is the layer where agent sessions run, transcripts accumulate, diffs are reviewed, artifacts are captured, and branches land. If that layer becomes central to software development, it should be open and hackable too.

ctx does not yet have Pi-style extensibility or plugin primitives. Making the workbench more extensible, plugin-ready, and hot-reloadable is an active area of development. If that direction interests you, we would love contributions.

Under the hood

ctx is built around a local Rust daemon, real agent harnesses, and per-task worktrees.

The workbench talks to a local Rust daemon that owns sessions, transcripts, artifacts, diffs, workspace state, provider setup, and merge queue state

ctx runs real agent harnesses instead of replacing them with one internal agent loop; adapters use structured runtime protocols where available instead of scraping terminal output

CRP adapters normalize provider streams into one durable session model for the UI, review surface, local SQLite store, and artifact system

Each task runs in its own worktree, so transcripts, artifacts, diffs, and review state are tied to the exact execution root that produced them

Sandbox runs are materialized into an isolated container data plane instead of mutating a host checkout in place

Containerized runs can use explicit network egress policies such as LLM-provider-only, allowlist, or full access

The local merge queue replays patches against the latest target branch in queue worktrees, runs verification, and only advances the target when the entry applies cleanly

The desktop app is Tauri with a TypeScript UI, but the runtime path is Rust. The repo uses Bazel for the larger validation graph alongside normal Cargo and pnpm workflows

Get started

Install and launch

Connect a provider

Add a workspace

Run your first task

Learn more

Workbench tour

Containerization

Agent Merge Queue Overview

What is a worktree?

ADE vs CLI

ADE vs IDE

Build From Source

Prerequisites:

Rust stable

Node.js 20+ with Corepack

Platform desktop dependencies if building the Tauri app, including GTK/WebKitGTK on Linux

Install dependencies and build the Rust workspace:

git clone https://github.com/ctxrs/ctx.git cd ctx cd core corepack enable pnpm install --frozen-lockfile cargo build --workspace

Run the daemon and web workbench locally in separate terminals:

cd core cargo run -p ctx-http --bin ctx -- serve --bind 127.0.0.1:4399 --data-dir "${CTX_DATA_DIR:-$HOME/.ctx}"

cd core pnpm -C apps/web dev

Launch the desktop app from source:

cd core pnpm desktop:dev

Build the desktop app from source:

cd core pnpm desktop:build

About

Open source Agentic Development Environment (ADE): a hackable desktop workbench for coding agents.

ctx.rs

Topics

rust

ai

developer-tools

cursor

agents

codex

ade

claude

tauri

worktrees

coding-agents

agentic-development-environment

Resources

Readme

License

GPL-3.0 license

Contributing

Contributing

Security policy

Security policy

Uh oh!

There was an error while loading. Please reload this page.

Activity

Custom properties

Stars

80 stars

Watchers

1 watching

Forks

5 forks

Report repository

Releases

1 tags

Packages 0

Uh oh!

There was an error while loading. Please reload this page.

Uh oh!

There was an error while loading. Please reload this page.

Contributors

Uh oh!

There was an error while loading. Please reload this page.

Languages

Rust 60.1%

TypeScript 37.9%

CSS 0.9%

Starlark 0.8%

Shell 0.2%

JavaScript 0.1%