Charlie Holtz, the founder of Conductor, announced today (November 4) that Conductor is joining Meta:
Conductor was one of the first agentic development environments I used, and a huge inspiration for a new way of working with multiple models without swapping context between them. Theyâre a pioneer, and I canât wait to see what they do next.
In the meantime, here is the timeline theyâve laid out in the announcement:
- November 4New sign-ups and downloads stop.
- November 13Conductor Cloud becomes read-only.
- November 20Cloud shuts down and data deletion begins.
- Early 2027The local app winds down. Model updates and critical fixes continue until then.
Export your Cloud data, or move those workspaces to local, before the 13th.
A few people have asked me for help navigating this, so here is a summary of the research Iâve done and shared with them. If thereâs anything youâd like me to discuss, or anything I missed, DM me on X.
Iâve spent the past few months on agent-to-agent tools, and I keep comparison sheets on these products as I go. This post folds them together. I checked each claim below against the makerâs own pages on November 4, and listing a tool is not an endorsement.
Majors and up-and-comers
I sort these into two groups. The split is my judgment, not a measured ranking. Cloud means the maker runs agents on its own computers; every tool here also has a desktop, phone or web client. Remote means you can reach your own other machines, such as a server or a box over SSH. Free means a plan you can keep using without paying; a trial doesnât count. License is the open-source license in the public repository, and a dash means there is no public repository. A dot means the toolâs own pages back it, as of November 4; a blank means they didnât say. Stars are from GitHub on the same day, rounded.
| Tool | Stars | License | Cloud | Remote | Worktrees | Review | Phone | Free | Conductor import |
|---|---|---|---|---|---|---|---|---|---|
| Majors | |||||||||
| Orca | 115k | MIT | No | Not stated | |||||
| T3 Code | 46k | MIT | No | Not stated | |||||
| Paseo | 23k | Apache 2.0 | No | Not stated | |||||
| Paradev | 16k | MIT | |||||||
| Up-and-comers | |||||||||
| cmux | 29k | GPL 3.0 | Not stated | Not stated | Not stated | ||||
| Orchestrator | 14k | Apache 2.0 | Not stated | ||||||
| Emdash | 6.2k | Apache 2.0 | No | Not stated | Not stated | ||||
| Traycer | 1.7k | MIT | No | Not stated | |||||
| Cloudroom | 1.4k | Apache 2.0 | Not stated | ||||||
| Boxes.dev | â | â | Not stated | Not stated | Not stated | Not stated | |||
| Capy | â | â | Not stated | Not stated | |||||
| Replicas | â | â | Not stated | Not stated | Not stated | ||||
| Superconductor | â | â | Not stated | Not stated | Not stated | ||||
Stars measure attention, not quality, and a hosted toolâs paying users never show up in them.
Where the agents run
Most of these tools have desktop, phone or web clients, and a client doesnât tell you where the agents execute. Each entry below says it: local (your machines), cloud (the makerâs computers), or both. Paradev, cmux, Orchestrator, Cloudroom and Capy do both. Replicas, Boxes.dev and Superconductor run only on the makerâs computers.
A closer look at the majors
Desktop and mobile clients, worktrees, review, and access to your own remote machines. Remote access here means your machines, not managed compute. Repository.
A desktop, mobile and web agent interface with review and remote access to your own hosts. Worktrees are covered in its user docs. Remote access can go over your LAN, Tailscale, another VPN, SSH or a public URL. T3 Connect is the easy route; it uses T3âs cloud tunnels and a sign-in, and a paid plan raises its machine limit. It is MIT licensed. The repository warns that the project is early.
Desktop, phone and web clients for daemons running on your machines. The app is free and open source under Apache 2.0. The optional hosted Hub manages triggers, not agent compute. Hub Pro is $15 per seat per month, and model and machine costs stay yours.
Formerly Superset. Local worktrees and a beta cloud. The Free plan covers one user with local workspaces, the desktop app and the CLI. Pro is $20 per user per month and adds remote access, automations and the phone app. Cloud workspaces start on Team, at $60 per user per month with 150 hours per seat. Models cost extra (pricing). The pricing page changed on November 4; earlier that day it listed 40 cloud hours on Pro. Its Conductor guide adopts your Conductor worktrees in place, with their branches and uncommitted changes. Chats donât come across; see below.
A closer look at the up-and-comers
A free Mac terminal workspace, and local agents use your own API keys at no cost. Persistent cloud VMs are paid early access. Pro starts at $50 a month, and the pricing page has also shown a $15 Plus tier with one saved cloud VM. On cloud VMs, the agents use your teamâs own accounts through CodeRouter. I couldnât confirm automatic worktrees per task.
Local worktrees with PR and CI feedback loops. Hosted execution needs an enabled build and account access. I couldnât establish whether cloud includes model costs.
Parallel CLI agents on macOS, Windows and Linux, with worktrees, diffs and a PR workflow, locally or over SSH. The stable v1 and the 2.0 waitlist are separate things. Repository.
A desktop workspace with worktrees, review and your own remote hosts, under an MIT license. Free already shares tasks with people you invite, and Sync shares a task with the whole team for $20 per user per month (pricing). Provider costs are separate. Optional Traycer inference adds 20% and includes no monthly credits, and the automatic judge can spend credits.
Stable 1.5.3 shipped on November 3. An open report describes memory and queue issues; I didnât reproduce it.
Runs agents on your own computer or on Cloudroom Cloud, where threads keep working with the laptop closed. Locally it runs Claude Code, Codex, Pi and any ACP agent, such as Cursor or OpenCode (home). Its open-source Core runs on your own Ubuntu 24.04 server, while GPUs, Cloud loops and computer use run only on Cloudroom Cloud (FAQ). The Free plan includes 3,000 cloud credits a month and unlimited local agents. Pro is $50 a month for 50,000 credits, Super $150 for 150,000 and Ultra $500 for 500,000, and Cloudroom estimates an hour of agent work at about 100 credits (pricing). Models stay on your own Claude or Codex plan with no markup, so credits pay only for the cloud computers. Its local diff and worktree features donât mean the cloud has the same ones (local threads).
A thread runs on a cloud VM or, with the desktop app, in a worktree on your own computer, and it can move between the two (desktop). capy dock pairs a server to run threads without the app (Dock). Capy opens the pull request, fixes failing CI and answers review comments, and a separate review agent posts inline findings. There is no phone app; capy.ai works in a phone browser (FAQ).
Plans have no seat fees. Lite is $20 a month with $20 of credits, Pro is $100 and Max starts at $200, and Liteâs first week costs $1 (pricing). Threads on your own computer have no machine-time charge, and your own subscriptions or keys cost nothing on any plan. The docs mention a free plan, but the plans table doesnât list one.
Hosted Linux workspaces. It has desktop and mobile clients, but the agents run on its computers. The plan includes $50 of usage (billing), and you bring your own model credentials. It documents a move from Conductor Cloud for workspaces, setup scripts, secrets, routines and team.
Hosted Linux computers for Claude Code and Codex. Starter includes 50 box-hours and four awake devboxes plus a Template. A standard hour is 4 vCPU and 8 GiB, and extra usage is $0.60 per box-hour. Larger boxes and an awake Template use more, and models cost extra (billing).
Hosted agent environments and PR review. Pro starts at $50 a month for 20,000 credits and rises with the credits you pick (docs); the pricing page opens on the $100 tier. The free plan includes 20,000 credits, and compute costs 100 credits an hour at 1 vCPU and 2 GB. Model tokens can come out of credits; bring your own tokens and credits go only to the cloud computers.
The maker posted âFree Pro until end of the year for CC customers!â, and the pricing page gives no terms for it.
A shout-out to Amp
Amp plays a different game from the tools above: it is a coding agent with its own development environment. It still answers the same need. A thread can run in an orb, a machine Amp sets up with your code, tools and plugins, so the agent keeps working after you close your laptop (manual). It can also run on a runner or in a local CLI session (threads). Orbs can be Amp-hosted or self-hosted, and your Mac can serve as a runner. You can start one from your phone and follow up from your Mac or the terminal.
The Hobby plan is free. Amp-hosted orbs are pay-as-you-go. Self-hosted orbs and your own runners are free on every plan except Enterprise. You can bring your ChatGPT or Claude subscription or your own keys, with no Amp token fees. Megawatt is $20 a month with 45,000 minutes on Amp-hosted orbs (pricing).
Amp is also the one tool here that imports Conductor chats. Its co-founder Karl Clement posted the steps: update Amp and run amp threads import conductor --all, or open Settings, Imports, Import in the Mac app. Move Conductor Cloud chats to your Mac first (right-click, Transfer to local) before November 20. Amp is also offering three free months of Megawatt to people without a paid Amp plan.
What moves over
Two tools document a path from Conductor.
Paradev adopts local Conductor workspaces where they sit (local migration). Add the same repository folder Conductor used, not a new clone, then right-click the project and choose Import untracked worktrees. Nothing is copied or moved, and each workspace keeps its path, branch and uncommitted changes. Leave Run setup script off, since theyâre already set up. A worktree on a detached HEAD needs a branch before it shows up in the list.
Scripts move from conductor.json or .conductor/settings.toml into .paradev/config.json. Setup stays setup, run stays run, and the archive script becomes teardown. The CONDUCTOR_* path and name variables become SUPERSET_*. CONDUCTOR_PORT has no equivalent: Paradev detects the ports each workspace listens on instead of assigning them. Once a workspace is imported, donât archive it in Conductor. That deletes the workspace directory, which is now your Paradev workspace too. Quit Conductor instead.
Conductor Cloud has the shorter clock (Cloud migration). Commit everything with git add -A and push every branch. A push leaves Git-ignored files such as .env behind, so copy those out yourself. Then either right-click the workspace, choose Transfer to local and import it as above, or continue in a Paradev cloud workspace. For the cloud route, create an environment with the repositories, variables and secrets your Conductor setup used, and commit .paradev/config.json to every base branch. Paradev starts a new working branch from the one you pushed, so an open pull request on the Conductor branch wonât get the new commits. Routines become automations with Cloud as the device. Cloud workspaces need a paid plan and are in early access.
Replicas moves Conductor Cloud workspaces, setup scripts, secrets, routines and team into its own cloud workspaces. It doesnât take local worktrees.
Only Amp imports Conductor chats; see above. Paradev says it canât, and points you to Conductorâs Settings, Account, Export data for a record of them. On your Mac, Claude Code sessions are a partial exception. Conductor saves them in your usual ~/.claude folder under the workspaceâs path, and an imported workspace keeps that path, so claude --resume in it lists the old conversations. Claude Code deletes sessions older than 30 days by default. Replicas says it canât import chats either. Traycer and Boxes can import local conversation history from your model provider, but that isnât the same as importing Conductor chats or Cloud exports.
What the makers said
Plenty of makers answered Charlieâs post with an offer. Here are the tools in this post, plus Agentastic, in their own words.
Paradevâs account also replied with its guide for moving off Conductor Cloud. Users recommended T3 Code and Paseo. Capy, Cloudroom, cmux and Emdash are here on their documentation alone.
Founders and teams also pitched these in the thread. I havenât checked them yet:
- Aether @pranavctl
- Alan @prasad_pilla
- AQ @anmelano
- Brahma @codeanand
- Codz @codz_com
- Happy @bra1n_dump
- Leverage @eimen0
- Nyte @d2ac__
- Prized @hudson_gri
- Reason Machines @svemyh
- TalentOS @Win22James
- Tembo @benjaminakar
- Vorflux @harithaelango
Add your tool
If you build one of these, or one I missed, DM me on X with the details below. Link your own page for each answer, and Iâll add the tool once those pages back it.
Tool:
Website:
Public repository and license:
Where agents run (local, your cloud, or both):
Reach your own remote machines (SSH, a server):
A worktree or sandbox per task:
Review diffs or pull requests in the app:
Phone client (iOS, Android, web):
Free plan (not a trial), and paid prices:
Moving from Conductor (link to the docs):
Agents it runs (Claude Code, Codex, others):
Prices and dates are as each makerâs pages stated them on November 4, 2026. They will change.










