15 open-source orchestrators, one table of same-day 2026-09-22 numbers
This page does one job: it puts the same-day numbers from 15 repositories side by side, so you can see who still pushes, which licence family a tool is in, and whether it is a desktop app or a command line, without opening each README. Stars and forks move daily, so every number is pinned to its fetch day, 2026-09-22; re-check against today's API before quoting. Orca's own row is measured with the same ruler and placed first. Checked 2026-09-23.
Updated 2026-09-23
Four dimensions you can count on the fetch day
Stars and activity
Ranked by 2026-09-22 stars: Orca first at ★75,454, vibe-kanban second at ★28,168, cmux third at ★27,335; terragon-oss last at ★259. Only 1 of 15 clears 50,000. For upkeep look at the last push: 3 pushed before 2026-06-01 — crystal (2026-02-26), uzi (2025-06-04), terragon-oss (2026-02-10). Stars are a stock signal; the push date is the "does anyone still tend this" signal, and you need both. Checked 2026-09-23.
A licence is not a two-way open/closed switch
7 are MIT (Orca included), 5 Apache-2.0, 1 AGPL-3.0 (claude-squad); Worktrunk is MIT OR Apache-2.0, your pick. For 3 more the GitHub API says NOASSERTION — its classifier did not match the file, there is still a licence: cmux's README says GPL-3.0-or-later, Superset's says Elastic License 2.0 (ELv2), Worktrunk's says MIT OR Apache-2.0, all three verifiable word for word. If you plan to commercialise or distribute in closed form, AGPL and ELv2 are a different matter from MIT, so that column deserves its own look. Checked 2026-09-23.
Form and model source: only where a README says it
For the form, model-source and pricing columns this batch has verbatim anchors for 11 repositories — every phrase here sits behind a sentence from that repository's README, kept only after a literal string comparison. Desktop app, terminal-native, command line, cloud: each wording is one repository's own. The other 4 (nimbalyst、parallel-code、uzi、terragon-oss) had no README fetched, so those cells stay empty. The cheap lie in a comparison page is writing «so is this one» from «most of its kind are». Checked 2026-09-23.
The cost of parallelism: a user's own numbers, no vendor benchmark
Git itself gets slower as parallelism grows — in public sources you can find users' own accounts, not vendor benchmarks. An open issue in the Orca repository (#10769, filed 2026-07-26 by AmethystLiang, 2 comments, fetched 2026-09-22) reports: with 297 live worktrees in one repository, git commit from the Source Control panel went from about 1.5s to a median of 8.9s (peak 21.6s), with .git at 1.1G. It is one repository on one day, self-reported, not a reproducible benchmark, and it describes Orca alone. It is on the table because people choosing an orchestrator ask exactly this — read it with its own qualification. Checked 2026-09-23.
How to read this table
15 rows, five parts each: stars (GitHub API of 2026-09-22), licence, last push date, latest release with its publish date, then form, model source and pricing. The first four come straight from the API archive — a machine read them, nobody transcribed them. The last three are filled only where that repository's README has been checked word for word: this batch fetched and verified 11 READMEs, and for the remaining 4 the cells say "not in the README" rather than borrowing a neighbour's description. One licence caveat: GitHub's API returns NOASSERTION for 3 of them, which means its classifier did not recognise the file — not that there is no licence. cmux, Superset and Worktrunk each state their licence in the README verbatim, so that is what the column says, with the source marked in parentheses. Checked 2026-09-23.
Why a fetch day is mandatory
Within the same day the largest star gap among these 15 is not at the top but at the tail: terragon-oss has ★259 while the second-placed vibe-kanban has ★28,168 — a factor of 108. Speed of change matters more: Orca went from 73,754 on 09-21 to 75,454 on 2026-09-22, 1,700 in a day. A star leaderboard without a fetch day is already wrong the next day. So this page appends the fetch day (2026-09-22) to every number and lists each latest release's publish date as well — release dates show whether a project is still moving this week better than stars do: orca 2026-09-22, vibe-kanban 2026-04-24, cmux 2026-09-17, superset 2026-09-21, agent-orchestrator 2026-09-12. Checked 2026-09-23.
Time anchors of this page
2026-09-22: every star, fork, licence, pushed_at and latest_release on this page comes from that day's GitHub API archive, 15 repositories in total; the same day 10 full READMEs were fetched, and they are where the form and model-source anchors come from. That day Orca stood at 75,454 stars, 1,700 more than its 09-21 value of 73,754.
2026-07-26: issue #10769 filed in the Orca repository, «Slow Git auto-maintenance is invisible in the UI, and workspace cleanup…», reporting the git-commit latency change at 297 live worktrees; still open on the fetch day. This page uses it only to show that parallel cost exists, not as a benchmark number.
2026-09-23: the four languages of this page were finalised. Stars and releases move with versions, so every number is locked to the 2026-09-22 fetch day; if you come back in a few days, re-fetch today's API instead of reusing this page.
What you can compare yourself vs what only the vendor can say
What a sentence in the archive backs
Checked 2026-09-23; these match the GitHub API archive word for word: 1) stargazers_count, forks_count, license.spdx_id, pushed_at and the tag plus published_at of the latest release for all 15 repositories; 2) Orca being MIT with 75,454 stars and 4,939 forks; 3) three verbatim licence sentences in READMEs (cmux "cmux is open source under [GPL-3.0-or-later]", Superset "full source is on GitHub under Elastic License 2.0 (ELv2)", Worktrunk "License: MIT OR Apache-2.0"); 4) the issue #10769 sentences 297 live worktrees, ~1.5s to 8.9s median (peak 21.6s) and .git size: 1.1G. All of that is at the "the archive contains this sentence" level; whether the numbers make one tool better than another is not a claim this page makes.
⚠️ Three things people assume
One, «more stars means it fits me» among these 15: stars measure attention, not your repository size, whether you want a GUI, or whether you can live with AGPL — and on this page the newest release date does not belong to the tool with the most stars. Two, «GitHub says NOASSERTION so there is no licence / no commercial use»: that only means the automated classifier failed; all three spell their terms out in the README, and Superset's ELv2 is precisely a use-restricted licence — the opposite direction. Three, «if it is on the table it is maintained»: crystal's last push is 2026-02-26 and uzi's README states it is deprecated; this page copies the API and README values as they are, without calling them active. Checked 2026-09-23.
Middle 5 and last 5
Middle 5, starting with claude-squad
claude-squad: ★8,522 (GitHub API 2026-09-22) · licence AGPL-3.0 (API) · last push 2026-08-20 · latest release v1.0.20 (2026-08-20) · form terminal TUI, several sessions in one window · model drives local agent CLIs; Codex brings its own key · pricing pricing not mentioned; source is AGPL-3.0 worktrunk: ★8,339 (GitHub API 2026-09-22) · licence MIT OR Apache-2.0 (README) · last push 2026-09-22 · latest release v0.79.0 (2026-09-21) · form CLI (the wt command plus shell integration) · model drives external CLI agents (example: claude) · pricing open source, MIT OR Apache-2.0 emdash: ★5,806 (GitHub API 2026-09-22) · licence Apache-2.0 (API) · last push 2026-09-22 · latest release v1.2.6 (2026-09-21) · form desktop app (three platforms) · model reuses the CLI agents already installed · pricing open source under Apache-2.0 container-use: ★4,046 (GitHub API 2026-09-22) · licence Apache-2.0 (API) · last push 2026-09-21 · latest release v0.4.2 (2025-08-19) · form CLI tool plus a stdio MCP server · model model comes from the host agent; no key or subscription mentioned · pricing open source (Apache-2.0), pricing not mentioned crystal: ★3,120 (GitHub API 2026-09-22) · licence MIT (API) · last push 2026-02-26 · latest release v0.3.5 (2026-02-26) · form deprecated (2026-02; form not stated) · model model source not stated in the README · pricing pricing not stated in the README
Last 5 (star 232–1,762)
sculptor: ★232 (GitHub API 2026-09-22) · licence MIT (API) · last push 2026-09-22 · latest release sculptor-v0.48.0 (2026-09-21) · form desktop app: runs coding agents in parallel · model through external agents such as Claude Code/Pi · pricing calls itself open source, pricing not mentioned nimbalyst: ★1,762 (GitHub API 2026-09-22) · licence MIT (API) · last push 2026-09-21 · latest release v0.77.5 (2026-09-09) · form / model / pricing: this repository's README was not fetched in this batch, so left unwritten parallel-code: ★1,018 (GitHub API 2026-09-22) · licence MIT (API) · last push 2026-09-22 · latest release v3.0.1 (2026-09-19) · form / model / pricing: this repository's README was not fetched in this batch, so left unwritten uzi: ★583 (GitHub API 2026-09-22) · licence MIT (API) · last push 2025-06-04 · latest release v0.0.2 (2025-06-03) · form / model / pricing: this repository's README was not fetched in this batch, so left unwritten terragon-oss: ★259 (GitHub API 2026-09-22) · licence Apache-2.0 (API) · last push 2026-02-10 · GitHub API lists no release · form / model / pricing: this repository's README was not fetched in this batch, so left unwritten
Top four, counting Orca
orca: ★75,454 (GitHub API 2026-09-22) · licence MIT (API) · last push 2026-09-22 · latest release v1.4.207 (2026-09-22) · form desktop app plus a phone companion, orchestrating parallel worktrees · model model source not stated in the README · pricing free and open source (MIT) vibe-kanban: ★28,168 (GitHub API 2026-09-22) · licence Apache-2.0 (API) · last push 2026-09-19 · latest release v0.1.44-20260424091429 (2026-04-24) · form local desktop app with a browser interface · model model source: the coding agent you already authenticated · pricing pricing not stated in the README cmux: ★27,335 (GitHub API 2026-09-22) · licence GPL-3.0-or-later (README) · last push 2026-09-22 · latest release v0.64.25 (2026-09-17) · form native macOS terminal (not Electron/Tauri) · model model source: the agent CLIs you installed yourself · pricing free and open source under GPL, plus a paid Founders tier superset: ★14,501 (GitHub API 2026-09-22) · licence Elastic License 2.0 (ELv2) (README) · last push 2026-09-22 · latest release desktop-v1.30.1 (2026-09-21) · form desktop app plus CLI/SDK/MCP · model runs any CLI agent; you can plug in OpenRouter · pricing desktop is free forever; anything paid is an optional service agent-orchestrator: ★12,267 (GitHub API 2026-09-22) · licence Apache-2.0 (API) · last push 2026-09-22 · latest release v0.13.0 (2026-09-12) · form desktop app plus a local daemon · model plugs into the coding agents you already have · pricing open source under Apache-2.0, no price stated (every star above is the GitHub API value for 2026-09-22)
What this has to do with QCode
An orchestrator answers «how many agents run at once and where do their edits land»; models and billing sit on the other side: across the 11 READMEs on this page the model source is almost always stated as «the agents you already installed or authenticated» (Orca included — it ships no model). So what sets the cost is the token price and cache behaviour on that other side, not the orchestrator's star count. QCode is on that side: one key, per-token billing, several models, and the CLIs above point at it after you configure their own endpoint, while the orchestrator runs them in parallel. The quota and price arithmetic lives on the parallel-agent quota-and-cost page; this page does not repeat prices. Checked 2026-09-23.
Questions
Whose stars are these, and can I still use them?
All of them are the GitHub API values for 2026-09-22, with the fetch day written after each number. Whether you can use them depends on what you want them for: as an order of magnitude and a relative position, one day does not change the conclusion (Orca 75,454 against second-placed 28,168); as an exact figure, no — Orca alone gained 1,700 within that same day. To quote, go back to the 2026-09-22 archive or fetch it yourself today. Checked 2026-09-23.
GitHub shows NOASSERTION — does that project have no licence?
No. NOASSERTION only means GitHub's automated matching could not tie the LICENSE file to a known term. 3 of this batch carry it, and all three state a licence verbatim in their READMEs: cmux "cmux is open source under [GPL-3.0-or-later]", Superset "full source is on GitHub under Elastic License 2.0 (ELv2)", Worktrunk's badge reads "License: MIT OR Apache-2.0". So read that column either from the README or from the repository's LICENSE file — do not turn a failed recognition into an absent licence. Checked 2026-09-23.
Which of the 15 are GPL, AGPL or ELv2 — does commercial use need a separate look?
7 MIT, 5 Apache-2.0, 1 AGPL-3.0 (claude-squad); Worktrunk is MIT or Apache-2.0 as you choose; cmux is GPL-3.0-or-later per its README and Superset is Elastic License 2.0. AGPL is about the source-provision duty when you serve over a network, ELv2 explicitly forbids reselling the product itself as a service, GPL about its propagation terms — none of the three behaves like «just use it» MIT, and running it internally differs from offering it as a service. This page lists clause names only, it does not interpret them: read the original and your legal review. Checked 2026-09-23.
Does running dozens of agents slow git down? Are there numbers?
What is findable in public is a user's own report, not a benchmark. Orca issue #10769 (filed 2026-07-26, still open on the fetch day) says: in one repository at 297 live worktrees the median git commit went from about 1.5s to 8.9s, peak 21.6s, .git at 1.1G, with the cause pointed at degraded automatic maintenance of the shared object store. Single repository, single day, single vendor reported — different repository sizes and worktree counts will differ. To decide, open 20, 50 and 100 worktrees in your own repository and time a commit at each. Checked 2026-09-23.
Do fewer stars mean a worse tool?
This page happens to give the other direction: 3 pushed before 2026-06-01, among them crystal at ★3,120 whose latest release is 2026-02-26, while tail sculptor at ★232 shipped a release as recently as 2026-09-21 — closer to now than several with more stars. Tools are chosen by constraint match (licence, form, your repository size) and by upkeep; stars are a by-product of search popularity. Checked 2026-09-23.
Why are some rows' «form» and «pricing» cells empty?
This batch fetched and word-checked 11 full READMEs; the other 4 (nimbalyst、parallel-code、uzi、terragon-oss) have API metadata only. Those three columns answer the question «does this repository's README contain this sentence» — no sentence, nothing written, and no borrowing from other tools in the same category. The most expensive mistake in a comparison is putting into a subject's mouth something it never said. To fill these in, fetch those READMEs and check them word for word. Checked 2026-09-23.
Sources
Same-day repository metadata: GitHub REST repo and releases/latest endpoints, 15 repositories, fetched 2026-09-22 and stored locally as immutable archives named by response sha16. README text: each repository's default-branch README, fetched the same day (BloopAI_vibe_kanban, 10 of them plus Orca). Issue: the body of github.com/stablyai/orca/issues/10769, same day. No second-hand leaderboards, reviews or aggregators are cited here; «not stated» always means the text fetched in this batch does not contain the sentence.
Use your own constraints to rule most of them out
Three steps: read the licence first (if you distribute closed, pull AGPL and ELv2 out and read their terms); then the last push and the latest release date (half a year without either is an easy cut); then the form (terminal or desktop, per each README's own wording). Install the one or two that survive and try them; do the model bill separately, starting with the quota page.
Related reading
Orca agent development environment guide
What it is, what parallelism saves and what it costs — that page covers the mechanism; this one only puts it among 15 tools and compares the numbers.
Parallel agents quota and cost
An orchestrator does not pay for your models; the shared quota pool, context and cache spread across N parallel agents is accounted on that page.
Claude Code Worktree Parallel Workflow: Multi-Agent Git Isolation Done Right
The path without an orchestrator: managing parallel git worktrees yourself — the flags and commands are on that page.
This page restates public metadata and README content of 15 open-source repositories as of 2026-09-22; it is affiliated with none of them and holds no authorisation or inside information from them. Stars, forks, releases and push dates move with versions; «not stated» means the READMEs this batch fetched do not contain the sentence, not that the vendor denies it. The issue numbers are user-reported and unreproduced. For a selection decision use the official pages as they stand on the day you download.