Parallel Agents & Spec-Driven Work
Run several coding agents at once without them stepping on each other, and have them work from a written spec instead of a one-line prompt.
Trends
Public counts each month: pull requests on GitHub, package downloads, stars, Product Hunt followers. They show direction, not market share.
The real choice
More agents at once, or a better plan for each one. Parallel-agent workspaces give every task its own git worktree or container and one place to watch, review and merge the results, which multiplies output but also the diffs you have to read; spec-driven methods add a written spec, plan and task list the agent works through, which slows the start but keeps long tasks on track. Neither brings a model: they drive the agents and subscriptions you already have.
Pick by situation
Tap the ones that are you — the tools that fit light up below.- You use Claude Code or Codex on a Mac and want several tasks running at once, each in its own copy of the repo
Conductor
Superset - You live in the terminal and want parallel agents with free, open-source tooling
Claude Squad - You want the agent to write a spec and plan you approve before it codes, inside the agent you already use
Spec Kit
OpenSpec - You want plans and specs written as tickets that any agent can then pick up
Traycer - You want to follow and answer your agents from your phone while away from the desk
Happy - You want to see what your agents' token usage would cost
ccusage
Compare the tools
Grouped by the side of the choice they answer, not ranked. Open a row for when to use it, the trade-off and what makers say.
Conductorapp · free tierRunning several Claude Code, Codex, Cursor or OpenCode agents at once on a Mac, each in an isolated copy of the repo, then reviewing and merging their changes.


23+6 open source
Use it whenYou already use a terminal agent and want to work on several tasks in parallel.
Trade-offMac only, and it adds no model of its own; cloud workspaces, multiplayer and the API need the $50/month Pro plan.
llms.txtUsed by Evidence, Faces, Fluxmail, MindPal, Mito Health and 18 more · in 6 open-source projects.
As we progressed from traditional IDEs to Cursor and finally to Conductor we finally found the tool that lets us build with the speed of agents. Highly recommend!
Supersetapp · free tierA desktop workspace that runs Claude Code, Codex, OpenCode and other CLI agents in parallel worktrees, with a diff viewer, automations and a mobile app, keeping code and sessions on your machine.

8★ 15.1k
Use it whenYou want parallel agents with a free single-user plan, and may add teammates, Linear or Slack later.
Trade-offSource-available (ELv2), not open source; team members, remote access and automations need the Pro plan, and the product says it is about to be renamed.
Used by Pixero AI, Slashy, Sutra, CADAM, Checksum AI and 3 more.
Superset lets us build internal dashboards to track the metrics that matter: email processing latency, draft acceptance rates, inbox triage accuracy. We needed something powerful and self-hosted, and Superset delivered without forcing us into a SaaS analytics bill.
Claude Squadapp · open source · free tierAn open-source terminal app that runs Claude Code, Codex, Gemini CLI or Aider side by side in tmux, each in its own git worktree.No evidence yet★ 8.6k
Use it whenYou work in the terminal and want parallel agents without a desktop app or subscription.
Trade-offNeeds tmux and the GitHub CLI, and there is no visual diff review — you review in your own tools. AGPL-licensed.
No maker's product we track shows it for this yet.
Nimbalystapp · open source · free tierA free, open-source visual workspace for Claude Code and Codex sessions in worktrees, with a task tracker, editors for Markdown, diagrams and mockups, and a wiki the agents keep in your repo.No evidence yet★ 1.9k
Use it whenYou want parallel sessions plus a place for the documents and plans around them, on a free open-source app.
Trade-offRenamed from Crystal in early 2026, so older guides use the old name; team wiki features are a paid plan after the beta.
No maker's product we track shows it for this yet.
OpenSpeclibrary · open source · free tierA lighter spec framework that keeps a proposal, specs and task list per change in your repo, with commands to propose, apply, verify and archive each change.18 open source532.8k/wk
Use it whenYou want specs for changes to an existing codebase without a full planning process.
Trade-offStill adds a step to every change, and the specs only help if you read and correct them.
No maker's product we track shows it for this yet · in 18 open-source projects.
Spec Kitlibrary · open source · free tierGitHub's open-source toolkit whose `specify` CLI adds commands that take a feature from spec to plan, tasks and implementation as Markdown files, in Copilot, Claude Code, Gemini CLI, Codex and 30+ other agents.5 open source19k/wk
Use it whenYou want a structured spec-first process that works whichever agent you use.
Trade-offSeveral documents per feature before any code; heavy for small changes.
No maker's product we track shows it for this yet · in 5 open-source projects.
Traycerapp · open source · free tierA desktop workspace that plans work as specs and tickets, then hands them to Claude Code, Codex, Cursor, OpenCode or another agent, with terminals and diffs in one place.No evidence yet★ 1.6k
Use it whenYou want planning done once, in one tool, and executed by whichever agent you prefer.
Trade-offRecently moved from an editor extension to a desktop app; cloud sync and team features are paid.
No maker's product we track shows it for this yet.
Happyapp · open source · free tierFollowing and steering Claude Code, Codex or Grok sessions from your phone or another computer, on the subscriptions you already have, with end-to-end encrypted sync.No evidence yet8.2k/wk
Use it whenAgents run long tasks and you want to answer their questions away from your desk.
Trade-offAnother layer between you and the agent; sync runs through Happy's relay server (encrypted end to end, and open source like the rest).
No maker's product we track shows it for this yet.
Before you choose
Start with two agents on unrelated tasks before trying five; the limit is how many diffs you can review well, not how many agents you can start. Give each task its own worktree or workspace and its own branch, and merge one at a time with tests passing. For features bigger than an afternoon, have the agent write the spec and task list first, read them, and only then let it implement — fixing a plan is cheaper than fixing the code it produced.
- Starting more agents than you can review, so unread diffs pile up and get merged in bulk.
- Running parallel agents in the same checkout, where they overwrite each other's files and break each other's dev servers and databases.
- Writing specs for one-line fixes, spending more time on the method's documents than the change itself would take.
Other options
Real choices most makers here won't need to weigh.
Decided alongside
What the 30 makers' products here chose for their other decisions.