T3 Code on GitHub: 57,212 stars on November 12, 2026

pingdotgg/t3code had 57,212 GitHub stars at 03:45:28 UTC on November 12, 2026. The open queue beside that number is mostly pull requests: 3,426 of 5,178. And 759 of 823 releases are marked prerelease.

The repository is the project behind T3 Code. The README at commit 7462f234 calls it an agent-harness control surface. A harness, on this site, is the loop, tools, memory, permissions, and stop rules around a model. The FAQ on the same commit says T3 Code is not that loop. It drives a harness you already installed. What the product is, in the longer November 8 reading, is What is T3 Code?. This note is the GitHub census.

How many GitHub stars does T3 Code have?

57,212, in the repository response whose HTTP Date header was 2026-11-12T03:45:28Z.

The same object listed 10,539 forks, open_issues_count 5,178, the SPDX id MIT, TypeScript as the primary language, archived: false, and created_at 2026-02-08T23:46:22Z. That timestamp is 276.17 days before the response. The homepage field was https://t3.codes. The description field was null, so the phrase on the GitHub card is not where “agent harness control surface” lives. That phrase is the README. pushed_at was 2026-11-12T03:42:08Z, three minutes and twenty seconds before the star field was read. subscribers_count was 107. 57,212 divided by 107 is 534.7 stars per subscriber. Subscribers are people watching the repository. Stars are bookmarks. The ratio is arithmetic, not a count of operators.

The languages endpoint, in the same pass, returned 209,777,771 bytes across languages. TypeScript was 202,386,383 of them, which is 96.5 percent. The next share was MDX at 2.0 percent. Kotlin and Swift are in the list, which matches a repo that also ships phone apps. The byte share is not a claim about which language the agent loop is written in. T3 Code does not contain that loop.

The numbers in the tables are licensed CC BY 4.0. Credit AI Agents Info if you reuse them.

How fast did the star field move after November 8?

It rose by 5,810 stars from the November 8 reading to this one, and the latest stretch ran at 56.3 stars per hour.

This site already printed two earlier responses from the same endpoint. At 03:46:26 UTC on November 8 the star field was 51,402. At 11:48:27 UTC on November 10 it was 54,963. Those readings stay on What is T3 Code?. They were not requested again.

Source: aiagentsinfo.com. The first two rows are the published API readings. The third row is this pass. 54,963 minus 51,402 is 3,561. 3,561 divided by 56.0336 hours is 63.55, printed here as 63.6. 57,212 minus 54,963 is 2,249. 2,249 divided by 39.9503 hours is 56.29, printed here as 56.3. Per day, those rates are 1,525 and 1,351 if the pace held for 24 hours. It may not.

Across the whole repository life, 57,212 stars over 276.17 days is 207.2 stars per day. The latest interval, at 1,351 per day, is 6.5 times that lifetime average. 1,351 divided by 207.2 is 6.52. The average hides a quiet stretch. The longest gap between releases with prerelease: false is 671.65 hours, from v0.0.28 on June 29, 2026 to v0.0.29 on July 27. The star average is not a flat line, and this note does not have the daily star series that would draw one.

Forks rose by 370 between November 8 and November 10, and by 276 between November 10 and November 12. The combined open field fell by 17, then by 611. The 611 is issues and pull requests together. This pass did not split the November 10 field, so it cannot say whether issues or pull requests moved.

What is inside the open queue?

1,752 issues and 3,426 pull requests. The pull requests are 66.2 percent of the headline open count.

GitHub’s open_issues_count adds the two. A buyer who reads 5,178 as 5,178 bug reports is reading a pull-request queue as a defect queue. 3,426 divided by 5,178 is 0.6616, which is 66.2 percent. The issue share is 33.8 percent.

Source: aiagentsinfo.com. Search totals at 2026-11-12T03:45:31Z. incomplete_results was false on each of the four queries. 6,322 divided by 8,074 is 0.7830. 19,432 divided by 22,858 is 0.8501.

The closed share is high, and the open remainder is still mostly code waiting to merge, not mostly tickets. v0.1.9’s own notes, published the day before, open with “198 changes from 31 contributors.” That sentence is the release notes’ count for one tag. It is not the 477.

Who shows up in the contributor list?

477 GitHub users, and five accounts hold the largest contribution counts on the endpoint.

The Link header at 03:45:32 UTC named page 477 as last for a one-user page, and that page contained one user. At 03:47:27 UTC the top of the list was:

Source: aiagentsinfo.com. GET /repos/pingdotgg/t3code/contributors?per_page=5 at 2026-11-12T03:47:27Z. Contributions are the field GitHub returns for that list. They are not a count of merged pull requests, and they are not the 10,434 commits in the separate commits listing.

The README on 7462f234 says the project is very early, that bugs are expected, and that the maintainers are mostly not accepting contributions. It says small fixes may be considered and big features will not. The API list and that paragraph are both dated the same morning. A contributor count of 477 does not mean 477 feature proposals were accepted. It means 477 GitHub users have commits the endpoint is willing to list. The commits listing itself, from a Link header at 03:45:32Z, runs to page 10,434. This pass did not open that page, so 10,434 is the header’s last-page number, not a commit this desk read.

How often does a non-prerelease tag ship?

Ten tags, v0.1.0 through v0.1.9, shipped in 26.51 days. The full release list is mostly previews and nightlies.

The nine-page list finished at 03:46:36 UTC. It held 823 releases. Page 9 had 23. 759 had prerelease: true, which is 92.2 percent of 823. 64 had it false. Thirteen of those 64 are named v0.0.0-alpha. Forty-one are numbered v0.0.x. Ten are v0.1.0 through v0.1.9. GitHub did not mark the alpha tags as prereleases. The tag name and the boolean disagree, and this note keeps the boolean as the count.

/releases/latest in the 03:45 pass returned v0.1.9, published 2026-11-11T11:33:49Z. A releases page fetched in that same pass began with v0.1.10-preview.20261112.4560, published 2026-11-12T03:22:30Z. Latest and newest are different tags, because latest skips prereleases. Anyone automating an upgrade off the top of the releases page will install a preview. t3 update, in the install guide, says it moves to the newest release. This note did not run the command, so it does not say which tag that command selects.

The v0.1 tags, which are the current train:

Source: aiagentsinfo.com. Published times from the release objects. The span from v0.1.0 to v0.1.9 is 26.51 days. The nine gaps average 70.70 hours, which is 2.95 days. The shortest gap is 5.40 hours, between v0.1.4 and v0.1.5. The longest is 136.12 hours, between v0.1.7 and v0.1.8. v0.1.8 is the tag this site’s November 10 reading recorded as latest. v0.1.9 landed the next day.

v0.1.9’s notes are a product change, not only a version bump. They name T3 Connect agent sign-in, Claude session approvals for an MCP app, OpenCode goals kept in sync across clients, and GitLab revert of a merged merge request. The first line of the notes is the 198 changes and 31 contributors. The control surface is still shipping provider-specific behavior. It is not a frozen shell in front of untouched CLIs.

Which agent backends do the docs name?

Ten, if you read the install guide. Seven, if you read the README. Eight, if you read the FAQ. The files are the same commit.

The install table is the longest list, and it is the one that says how you sign in: codex login, claude auth login, Cursor’s agent login, grok login, kiro-cli login, opencode auth login, Google sign-in for Antigravity from T3 Code’s settings, pi then /login, muse login, and fx login. The guide says Codex connected through ChatGPT, and Antigravity, can run without a binary on PATH. The others need the CLI on the computer that runs the server, not on the phone.

The FAQ’s answer to “is T3 Code a harness?” is no. It says a harness gives a model tools, runs them, and returns the results, and that Claude Code and Codex are that layer. T3 Code uses them. It does not work without one. That split is the reason a star comparison with Claude Code would be the wrong table. Claude Code is a backend this app drives. Claude Code, Codex, and Cursor are three of the names with directory pages here. Seat prices, which this census does not reprint, are Best AI coding agents.

The README says the project was inspired by the Codex desktop app, Conductor, Claude Desktop, and Cursor Glass, and that none of them met the bar. It also says there is nothing to buy. The FAQ says the same thing about price: no way to pay the project to use T3 Code, MIT licensed, and you bring the subscription you already pay. T3 Connect is described as free for now, with a cap of three machines, and a paid T3 Chat plan is the way the FAQ says that cap rises. The FAQ does not print a dollar price for T3 Chat. This note does not invent one.

How do the other public control surfaces compare?

Two repositories whose own descriptions call them a control surface or mission control for coding agents had 58 and 56 stars. T3 Code’s field was 57,212.

A repository search the same morning, query conductor coding agent control, did not return the Conductor product the README names, and this pass did not match Claude Desktop or Cursor Glass to a repository object. Three public repositories from that search did return, and their descriptions are about controlling coding agents. They were read at 03:47:47 UTC.

Source: aiagentsinfo.com. T3 Code’s row is the 03:45:28Z repository object. The other three rows are 03:47:47Z. Latest-release tags are /releases/latest in that peer pass, which skips prereleases. hyldmo/conductor-remote is a phone panel for Conductor, not a second full control plane. It is in the table because the search returned it and the description names the job. It is not a claim that these four are the market.

57,212 divided by 58 is 986.4. 57,212 divided by 56 is 1,021.6. Those quotients are star fields. conductor-oss was last pushed 2026-10-31T03:28:20Z. ai-code-conductor was last pushed 2026-10-18T06:52:20Z. T3 Code was pushed the morning of this reading. A quiet repository can be finished. A busy one can be unstable. The README still says to expect bugs. The star gap does not settle which control surface to install. It settles which public repository, among the ones this pass could name, is the one people have bookmarked.

What should a builder do with these numbers?

Use T3 Code’s repository as a control surface that is moving, and pin the backend yourself.

The star field gained 2,249 bookmarks in 39 hours and 57 minutes, while a non-prerelease tag, v0.1.9, is a day old and the top of the releases list is a preview. If you install from “latest,” confirm you got v0.1.9 and not a v0.1.10-preview tag. If you install from the top of the releases page, you may get the preview on purpose. Read the tag before you treat the app as the November 10 build this site described, which was v0.1.8.

Treat 5,178 as 1,752 issues plus 3,426 pull requests. File a bug against the issue list. Do not read the combined field as the defect count. The docs you point an agent at should be the install table when you need the backend list, because the README is shorter by Kiro, Pi, and Muse Code, and the FAQ is shorter by Kiro and fx. The three files agree that the harness is the thing you bring. They do not agree on the menu.

The smaller public control surfaces are real repositories with real star counts under 60. They are not a reason to ignore T3 Code, and T3 Code’s star count is not a reason to ignore a license, a last-push date, or a backend you actually run. Prices for Claude Code, Codex, Cursor, and GitHub Copilot, which are the products under some of these buttons, stay on Best AI coding agents.