CNVS alternative: an agent canvas with a foreman and a memory

By Meffysto · updated October 4, 2026

If you're searching for a "CNVS alternative", you're probably after the same thing we are: a native app where several coding agents run side by side and you direct them by voice. On that bet, we agree with them. Here's, plainly, what's shared and what isn't.

What we have in common

Both products start from the same observation (running five agents in a terminal multiplexer doesn't hold up) and reach for the same family of answers:

• a native app, not an editor extension or a web service;
• a canvas where sessions live together, tiled automatically;
• real terminals running the CLIs as they are;
• voice as the main way to give commands, with named agents;
• speech recognition that runs on the device.

On voice, the gap is narrow: both use a local Parakeet model for transcription by default. If any of these points appeals to you, it'll appeal on both sides. The useful question is elsewhere.

Difference 1: a foreman, not just windows

A canvas solves display: you can see your agents. It doesn't solve coordination. You're still the one splitting the work, deciding who does what, and copying one agent's output into another's window.

Fauconnier adds that layer. A request can be split into tasks linked by their dependencies and run in waves: independent tasks start in parallel, the rest wait until their prerequisite has landed. "Split the rest" hands each agent its share of the list. And handoffs fire on an agent's real state: "when Merlin's done, send Milan: pick up the migration" waits for the actual end of Merlin's turn, not a timer.

Difference 2: an indexed memory, not a log

Remembering and logging are different things. Searching history for a string finds what you already remember. What you need is to find what you've forgotten.

Our memory is full-text indexed and organised in three readable layers: session history, the corrections you've made (they become instructions again on the next turn), and repeated procedures in the open SKILL.md format. Everything can be edited live: fix a note and the change counts immediately. A procedure written by an agent stays disabled until you arm it.

Difference 3: approvals in a queue, in batches

When agents touch real machines, the real question isn't "are they autonomous?" but "what happens before anything irreversible?". Our answer: sensitive requests don't block the agent. They queue up while it carries on with the rest, and you handle them together, including in one sentence ("approve everything except the delete"). Exclusion always beats blanket approval, and anything irreversible is refused if you don't answer.

Difference 4: life beyond your Mac

An agent can start on a remote machine, over SSH, inside a tmux session with a stable name. It keeps running with the lid closed, and Fauconnier reattaches it when you're back without creating a duplicate. Perchoir, our iPhone app, attaches to the same sessions: you see where the agent is and answer its question from your phone. The full recipe is in the guide on remote agents.

What you get with Fauconnier

We don't publish a table of CNVS prices or features: their product moves fast, and a side-by-side would be out of date before you read it. Go look at their site, it's well made. Then come back to the four points above, which are structural and will change less.

How to choose

• Mostly want to see your agents together and talk to them? Either will do. Pick the interface you like.
• Want something to split and chain the work for you? That's our main difference.
• Losing time re-explaining the same things every morning? Look closely at memory, on both sides.
• Agents touching production? Look at the approval queue and what's refused by default.
• Agents running on servers, or want to follow them from your phone? Look at SSH, tmux and Perchoir.

The simplest way to decide is to try. Fauconnier is free and asks for no card and no sign-up: run both for a week on your real work. That's the only comparison worth anything.

What carries over when you switch

Almost everything, because neither CNVS nor Fauconnier replaces your agents. Your CLIs stay installed and logged in. Their instruction files (CLAUDE.md, AGENTS.md), settings and SKILL.md procedures belong to the CLIs and the repo, not to the app displaying them. What doesn't carry over is each app's own memory and your window layout. Give yourself half an hour to settle in: name your agents and relearn the two or three phrases you use most.

What Fauconnier doesn't do

To be complete, here are our limits. Fauconnier only runs on macOS 14 or later, on Apple silicon. The code isn't open. There are no team features: no shared accounts, no shared task board, and that's on purpose. Conversational voice, the mode that goes through a realtime provider, needs your own OpenAI or xAI key; basic voice commands stay local and free. If any of that is a dealbreaker, better to know now.

Frequently asked questions

Is Fauconnier free?

Yes. No licence, no trial, no account, as many Macs as you like. The code stays closed. You only pay for your own AI subscriptions, as you already do.

Is Fauconnier a fork of CNVS?

No. It's a separate app, written in Swift. The two share the same bet (native, canvas, voice), not code.

Does it work with the same CLIs as CNVS?

Fauconnier supports thirteen CLIs, including Claude Code, Codex, Gemini, Grok and Cursor. They're the real tools with your own accounts: nothing is reimplemented.

Is the app available in English?

Yes. The interface and voice commands work in English and French.

Read next

• Conductor alternative
• Voice control for coding agents
• Running several coding agents at once