Dev tools

Catenary launches a spatial IDE that wires Claude Code, Codex and Cursor onto one canvas

September 8, 2026 at 6:30 AM PT

Product Hunt launch card for the Catenary spatial IDE

Image: Catenary via Product Hunt

Why it mattersA developer running three coding agents in parallel today keeps them in separate terminal windows and copies context by hand, and Catenary is one of the first tools that treats the coordination as a wiring problem the IDE handles.

Catenary launched on Product Hunt on 7 September, a desktop application that puts multiple AI coding agents on the same infinite canvas with visual wires between them. The company describes it as a spatial IDE, and the pitch is that a developer running Claude Code on one branch and Codex on another today juggles terminal windows and copies context between them by hand.

The supported agents listed on the product site are Claude Code, Codex, Cursor, Grok, OpenCode and Pi. Downloads exist for Windows as .exe and .zip, for macOS on Apple Silicon and Intel, and for Linux as .AppImage, .deb and .tar.gz. The site notes that the installer is not code-signed yet, so the OS will block it on first run, and the docs give the two clicks around it.

What is on the canvas

Terminals running agent CLIs, Monaco editors, browser previews and parallel git worktrees sit as tiles on an infinite pan-and-zoom canvas. The New Task button on the sidebar creates an isolated branch, its own working directory and its own agent tile in one click, so a developer can spin up three parallel experiments without touching the code in front of them. Each task gets a colour that follows it through the sidebar and the dock.

Context wires and Maestro mode

The two features that make it a spatial IDE rather than a fancier tab bar are Wires and Maestro. Wires are directed lines drawn between two agent tiles: a developer can wire the output of a lead agent into a helper's input, either as a standing permission or as a relay that fires when the source goes quiet. Catenary's product site shows an example where a Claude tile asks a Codex tile to fix an N+1 query and the answer arrives back on the wire.

Maestro is off by default. A human click flips it on, and one agent then recruits, briefs and wires a team of up to five, adjustable up to ten. Every action Maestro proposes goes through an editable approval card before anything appears on the canvas, so the recruitment loop stays under human review.

Local-first and what it sends

Catenary lists exactly two outbound hosts. It hits github.com every 15 minutes for a latest.yml update-check file on its public releases, and it hits thecatenary.app only when a user pastes a licence key, removes one, or claims the trial. The site names each field in the trial payload (an sha-256 machine id, no name, no email, no hostname) and states that the analytics endpoint ships empty and the crash-reporter DSN was removed from the distribution.

The agent CLIs a developer chooses talk to Anthropic, OpenAI or Google with the user's own API keys, or to a local Ollama or LM Studio process. Catenary is not in that path. Pull the Wi-Fi and the app keeps working, because the Pro key is verified on the device by Ed25519 signature and the seven-day trial is a signed token verified offline.

Source

Source: Catenary

This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.

More from AI News

Armature measured 16,893 coding agent sessions to see which tools they pick

Armature ran 16,893 sandboxed sessions across Claude Code, Codex, and Cursor on 75 repositories to see which third-party tools each agent installs, and the three agents agreed on the same pick in only 42 percent of categories.

Source: Hacker NewsProductivity

Penelopa.ai reads real Codex and Claude Code sessions and turns repeated work into reusable skills

Penelopa.ai is a new tool that hooks into Codex and Claude Code, records real sessions, and reports back with skills, checks and prompts extracted from the patterns it finds.

Source: GitHubDev tools

Google publishes Artemis, an Android automation agent with MCP servers for Claude Code and Codex

Google published Artemis, an Apache-licensed Python framework that drives real Android phones from natural-language instructions and exposes an MCP server so Claude Code, Codex and Antigravity can dispatch tests to a connected device.

Source: GitHubDev tools