How Many GitHub Stars Does OpenClaw Have? 394,222 on November 9, 2026

OpenClaw had 394,222 GitHub stars at the end of an API pass at 07:41:51 UTC on November 9, 2026. Hermes Agent had 261,856. Claude Code had 155,043, and the license file in that repository reserves all rights.

AI Agents Info pulled eleven repositories in that pass. Ten are projects whose own text calls them an agent a person runs. The eleventh, LangChain, is the library that led this site’s November 5 framework snapshot at 149,047 stars. On this pull LangChain had 149,241. The framework page is a release table. It earns the impression for “agent framework” stars and does not answer how many stars OpenClaw has. This note answers that, and it prints the clock.

How many GitHub stars does OpenClaw have?

394,222, in the repository response for openclaw/openclaw when the pass finished at 07:41:51 UTC.

The same object listed 82,839 forks, open_issues_count 8,006, the SPDX id MIT, TypeScript as the primary language, and created_at 2025-11-24T10:16:47Z. That timestamp is 349 days and about 21 hours before the pass finished. The homepage field was https://openclaw.ai. The description, with the trailing emoji removed, was “The AI that really does things. Any OS. Any Platform. The lobster way.” The latest release was tag v2026.11.2, name “openclaw 2026.11.2,” published 2026-11-08T19:31:04Z. pushed_at was 2026-11-09T07:41:03Z, forty-eight seconds before the pass finished, so the branch was moving while the count was being taken. The star field is the field from that response. It is not a promise about the next minute.

The README on main, fetched the same morning, says OpenClaw is an open-source AI assistant that runs on your own computer and meets you in the channels you already use. It names Discord, iMessage, Slack, Teams, Telegram, WhatsApp, and others, plus native apps. It says state, memory, and credentials live on your hardware, and that models and agent harnesses are plugins. A harness, on this site, is the loop, tools, memory, permissions, and stop rules around a model. The README’s claim that OpenClaw has no paid tier is the README’s claim. This note did not audit a billing page.

The eleven star fields sum to 1,594,452. Without LangChain, the ten agent repositories sum to 1,445,211. Without LangChain and without Claude Code, the nine repositories that returned an SPDX id sum to 1,290,168. Those sums are arithmetic on one pass. They are not a market size.

How many stars does Hermes Agent have, and what is the gap?

261,856. OpenClaw’s lead on that number is 132,366 stars.

Hermes Agent’s repository object listed 58,267 forks, open_issues_count 53,817, the SPDX id MIT, Python, and created_at 2025-07-22T22:22:28Z. That timestamp is 474 days and about 9 hours before the pass, so the repository record is older than OpenClaw’s and the star field is smaller. The description was “The agent that grows with you.” The homepage was https://hermes-agent.nousresearch.com. The latest release was tag v0.23.0, name “Hermes Agent v0.23.0 — The Hearth Release,” published 2026-11-05T21:32:33Z. pushed_at was 2026-11-09T07:40:08Z.

The README calls it the self-improving AI agent built by Nous Research, and says it runs from Telegram, Discord, Slack, WhatsApp, Signal, and a terminal, through one gateway process. The cost of running it is not in the star field. A separate November 9 note, Garry Tan’s harness evals, put a pinned Hermes Agent and a pinned OpenClaw on the same public tasks. On Opus 5.5 in that matrix, OpenClaw cost $0.4435 per solved task and Hermes cost $0.2434. Those are that report’s figures. They are not restated here as a new measurement, and they used older pins than the release tags in the table above. OpenClaw’s pin in that matrix was 2026.10.6. The tag on this pull was v2026.11.2.

Is Claude Code open source?

The repository is public. The license file does not grant an open-source license.

anthropics/claude-code returned 155,043 stars, 25,345 forks, and open_issues_count 16,907. The license object was null. A request to the repository license route returned 404. LICENSE.md on main is two lines. It says “© Anthropic PBC. All rights reserved.” The second sentence says use is subject to Anthropic’s commercial terms. This note did not parse those terms. It records what the file says.

The description field calls Claude Code an agentic coding tool that lives in the terminal. The homepage was https://code.claude.com/docs/en/overview. created_at was 2025-02-22T17:41:21Z. The latest release was v2.1.319, published 2026-11-06T17:04:01Z, and pushed_at was the same second. The product page on this site is Claude Code. A star rank that files this repository under “open source” is filing it by the fact that the code can be read, not by the license GitHub returned.

OpenClaw’s lead over Claude Code is 239,179 stars. That gap is a pair of counts. It is not a verdict on which coding agent to install. Price and limits for Claude Code and Codex are on Claude Code vs Codex pricing and on the November 9 coding-agent comparison.

The other SPDX ids in the table were MIT or Apache-2.0. Codex’s description was “Lightweight coding agent that runs in your terminal,” license Apache-2.0, 134,025 stars, latest tag rust-v0.167.0 with release name “0.167.0,” published November 6, 2026. Gemini CLI’s description was “An open-source AI agent that brings the power of Gemini directly into your terminal,” Apache-2.0, 108,797 stars, tag v0.67.0, published November 4. OpenHands, Cline, goose, Aider, and Browser Use returned MIT or Apache-2.0 as in the table. Browser Use’s description calls it an open-source Python library for AI browser agents, plus a hosted agent API. The glossary entry is browser use. The 120,380 stars are on a library-and-service repository, which is why the row is labeled by its description and not folded into OpenClaw’s “assistant on your own computer.”

Where does LangChain’s count sit now?

At 149,241 stars, 194 above the November 5 snapshot, and 244,981 below OpenClaw on this pull.

The framework release report copied 19 libraries and SDKs on November 5, 2026, and LangChain led them at 149,047. That page is the one in this site’s impressions for framework GitHub stats. Its title is a release table. A person who wanted the largest star count on an agent they can run does not get OpenClaw from it, because OpenClaw was not in the 19. 394,222 divided by 149,241 is 2.64, to two decimal places. The cleaner pair is the subtraction: 244,981 more stars on OpenClaw than on LangChain, same pass.

LangChain’s description on this pull was “The agent engineering platform.” The latest tag was langchain-core==1.7.6, published 2026-11-05T00:22:12Z. In a monorepo, /releases/latest can name one package. This tag names langchain-core. The November 9 framework comparison reads licenses and release tags for those 19 again, including this repository. It does not include OpenClaw or Hermes Agent. The directory listing is LangChain.

Five of the coding-agent rows were already counted on November 8 for the 30-day coding-agent brief. That brief’s question is price and the dates official pages have set through December 8. The star fields moved anyway:

The November 8 brief recorded no release tag for Aider. This pull’s /releases/latest call returned v0.86.0, published 2025-08-09T17:42:19Z. Aider’s pushed_at was still 2026-05-22T14:02:20Z, the same push time the brief recorded. The tag and the empty earlier call are both part of the record. Goose was not in that brief. On this pull block/goose had 56,382 stars, Apache-2.0, and release v1.57.0, published October 28, 2026. Its description calls it “an open source, extensible AI agent that goes beyond code suggestions.”

What does the open-issue number count?

It counts open issues and open pull requests together. Hermes Agent’s 53,817 is mostly pull requests.

A search later the same morning asked for open issues and open pull requests separately. The repository field and the search totals are different requests. For two of the three, they matched.

4,819 plus 3,187 is 8,006. 16,266 plus 641 is 16,907. 18,965 plus 34,850 is 53,815, which is two under 53,817. The miss of two is left as a miss. It is small next to the split that matters: about two thirds of Hermes Agent’s headline issue count, on this search, was pull requests. Codex’s open_issues_count of 26,901 was not split. Read it as the combined field.

OpenClaw’s combined field, 8,006, is the one that matched the search exactly. 4,819 open issues on a repository with 394,222 stars is a different shape from Hermes Agent’s 18,965 open issues and 34,850 open pull requests on 261,856 stars. Neither shape is a defect rate. Issues can be questions, and pull requests can be one-line typos.

What should a star table be used for?

Use it to see which repository accounts have bookmarked, on a dated pull, and read the license field before calling the repository open source.

The sentence the API supports: at 07:41 UTC on November 9, 2026, OpenClaw had 394,222 stars, Hermes Agent 261,856, Claude Code 155,043, and LangChain 149,241, and Claude Code’s LICENSE.md reserved all rights while OpenClaw and Hermes Agent returned MIT.

The sentence the API does not support: OpenClaw is the most-starred agent on GitHub, or the one a team should install, or a better harness than Claude Code. The set stops at eleven repositories. Leaderboards that add other names are answering a wider list, and this note does not import their numbers. Quality, for a harness, is a test of one action and then of the whole job. That method is How do you evaluate an AI agent harness?. The pages that already collect harness notes are on the agent harnesses hub.

Maintenance is a separate fact from stars, and it can run the other way. The same API pass that produced this table also read microsoft/autogen at 61,753 stars with archived false, while that repository’s README says Microsoft is not providing fixes. The write-up is Is AutoGen deprecated?. AutoGen was inside the November 5 library set. It is not one of the eleven rows above.