Stories
Slash Boxes
Comments

SoylentNews is people

posted by hubie on Sunday November 01, @10:28PM   Printer-friendly
from the Microsoft-free-for-now dept.

Now that agents are writing so much code, Entire is trying to reinvent the coding platform itself:

AI has sped up the pace of coding, putting new demands on infrastructure and changing how developers work. Entire, a developer platform from former GitHub CEO Thomas Dohmke, aims to be the solution to those changes.

The platform, available for use today, is built on Entire’s decentralized Git network, announced in July, which is designed to withstand the sharp increase in AI coding activity that has caused mounting GitHub outages.

Entire aims to be “a GitHub alternative for the speed that teams require to merge changes as fast as coding agents create them,” the company said in its announcement. The new git forge aims to make human-agent coding work more seamless by redesigning several features developers have become accustomed to. Here’s how it works.

The first change Entire makes to the standard coding experience is that it replaces pull requests with what it calls Trails.

“Unlike the classic Pull Request model of reviewing code changes with a diff, Trails offer developers a heads-up display similar to an airline pilot, with customized evals that are scored by agents on every code push,” the company said.

“All data is available not only on Entire's web interface, but also through the open-source Entire CLI. This allows coding agents to address findings in a loop, while the human developer validates the progress. When all evals are satisfied, they can merge the code without needing to look at the code itself.”

Entire said that the agents running this process are customizable, and each trail shows how agents are spending tokens.

"With Trails, Entire treats the idea, the agent trace, and the outcome of a developer's sessions as first-class artifacts for review alongside the code,” Dohmke said. “It uses multiple agents to generate evals that verify the quality and success of AI-generated code to let developers confidently merge without reviewing every line in the diff."

The company also claims it offers Git hosting “up to 89x faster than GitHub” and six times faster than Cursor origin, thanks to that aforementioned decentralized network. The platform lets developers create, migrate, and host private repositories in multiple locations across the world, currently starting with the US, India, Australia, and the EU.

Another differentiating factor is Continuous Memory, which keeps developer session logs next to code in the repo as a data source. Agents can use them to search for answers, which, Entire said, helps them use up to 71% fewer tokens. This is similar to other memory features in AI chatbots that reduce time spent gathering context by using past conversations or other agent interactions, where provided, to speed query response time. Entire said Continuous Memory is made up of a codebase graph and an “agentic brain” that uses previous sessions and prompts as lessons so agents don’t repeat mistakes.

Entire is available to try today for all developers.


Original Submission

 
This discussion was created by hubie (1068) for logged-in users only. Log in and try again!
Display Options Threshold/Breakthrough Mark All as Read Mark All as Unread
The Fine Print: The following comments are owned by whoever posted them. We are not responsible for them in any way.
(1)
  • (Score: 4, Insightful) by hendrikboom on Sunday November 01, @11:22PM (1 child)

    by hendrikboom (1125) on Sunday November 01, @11:22PM (#1458275) Homepage Journal

    Might this be exactly what we don't need?

  • (Score: 3, Informative) by JoeMerchant on Sunday November 01, @11:29PM (1 child)

    by JoeMerchant (3937) on Sunday November 01, @11:29PM (#1458276) Journal

    The newly launched GitHub competitor, Entire (founded by former GitHub CEO Thomas Dohmke), opened its decentralized Git network to everyone and is explicitly "free for now."

    While the platform currently allows users to host private repositories at no cost during its launch phase, its core focus is handling massive volumes of AI-agent coding activity through an advanced, evaluation-driven code review system called "Trails." Given the heavy computational costs associated with continuous AI-agent execution and decentralized infrastructure, the "free for now" status indicates that a tiered pricing model or paid enterprise features will likely be introduced as the platform matures.

    --
    🌻🌻🌻🌻✌️ [google.com]
  • (Score: 4, Insightful) by c0lo on Monday November 02, @01:47AM (16 children)

    by c0lo (156) Subscriber Badge on Monday November 02, @01:47AM (#1458281) Journal

    ... Trails offer developers a heads-up display similar to an airline pilot, with customized evals that are scored by agents on every code push,” the company said.
    ...
    When all evals are satisfied, they can merge the code without needing to look at the code itself.

    Force those humans into rubber-stamping AI slop; because some human will need to be scapegoated for the ensuing shit.
    What can go wrong?

    --
    https://www.youtube.com/@ProfSteveKeen https://soylentnews.org/~MichaelDavidCrawford
    • (Score: 4, Interesting) by JoeMerchant on Monday November 02, @02:11AM (8 children)

      by JoeMerchant (3937) on Monday November 02, @02:11AM (#1458283) Journal

      After the initial euphoria of the department generating a million lines of code per minute wears off (for those who get jazzed by such metrics), the humans driving the LLM agents should be refocusing on using those LLM token credits to generate good code, less code where possible, and encasing that code in automated validation suites to reduce future regression bugs.

      For one thing, if you let the LLMs write redundantly repetitive copy pasta bug bandage slop diarrhea as your deployed code, that's going to take a lot more tokens and time to maintain than if you invest the time in refining the code to concise spare accurate, complete but minimal implementations.

      Backing up a layer, examine the requirements and refine them for efficient implementation before even starting - are there un-necessary, wasteful requirements to meet? Does the architecture support modular re-use, DRY SSoT? Now that you have a legion of agent minions at your disposal, actually reduce that technical debt before declaring "ready to ship" - the payoff is real.

      --
      🌻🌻🌻🌻✌️ [google.com]
      • (Score: 4, Insightful) by c0lo on Monday November 02, @03:00AM (7 children)

        by c0lo (156) Subscriber Badge on Monday November 02, @03:00AM (#1458288) Journal

        encasing that code in automated validation suites to reduce future regression bugs.

        Reduce is a good word there.
        Some coding conventions can be used to further reduce the number of unit tests required, otherwise non-trivial projects can blow their test suites to prohibitively expensive to execute.
        I wonder if one can "teach" the agents to adopt such conventions.

        Now that you have a legion of agent minions at your disposal, actually reduce that technical debt before declaring "ready to ship" - the payoff is real.

        I never understood (productivity) metrics related to code, even less so since I started seeing code as a liability which often takes primacy over its investment aspect.
        At best, the productivity metric should point towards "how much fewer lines code can one manage to cover the same set of requirements" rather than "how many LLoC (were implemented)/time".

        (my last 15y in SoftEng were very little about producing code and a lot more towards minimal code designs; I doubt the current level of AI can be "taught" towards the same. Maybe I'm wrong, but I feel I'd prefer to mentor a bunch of juniors - which have chances to actually learn - than to constantly steer an AI to stop producing token fluffiness. Speaking of them tokens, that's how the AI masters get paid, so little chance for the said masters to take the pain of training the model towards code minimalism).

        --
        https://www.youtube.com/@ProfSteveKeen https://soylentnews.org/~MichaelDavidCrawford
        • (Score: 2) by JoeMerchant on Monday November 02, @04:07AM (5 children)

          by JoeMerchant (3937) on Monday November 02, @04:07AM (#1458292) Journal

          > non-trivial projects can blow their test suites to prohibitively expensive to execute.

          I was never good at efficient unit tests... I could write a test suite that would find all kinds of edge cases and bugs, but it would generally take hours to run and once it found its bugs it wasn't the kind of thing you'd want to run before every commit. Claude has been much better about covering with 1000 tests that run in a minute or two and do seem to still have some ongoing value catching the occasional regression.

          >I wonder if one can "teach" the agents to adopt such conventions.

          Well, our department just dropped the updated coding standards document... I expect we will be dumping that in the project folder for compliance testing...

          > productivity metric should point towards "how much fewer lines code can one manage to cover the same set of requirements" rather than "how many LLoC (were implemented)/time".

          Yeah, people inside the software team "get that" - outside, mostly not at all.

          > I doubt the current level of AI can be "taught" towards the same.

          At the rate of progress I have experienced over the last 18 months - I'm not doubting much at this point. What I'm struggling with in reduction of code volume at this point with LLMs isn't so much that they are producing "bad bloated code" - at least not after a few cleaning / review passes - but they're excessively thorough in consideration and coverage of edge cases - and I find it hard to fault them there. Looking back at my code of the past 10 years in a single project, the early stuff was very minimalist, but it started needing more and more and more edge case coverage as time went on, and eventually it was modularized into common helpers that covered all the edge cases and they have (mostly) replaced the simpler stuff that had less error handling, less result logging, etc. That error handling and logging pays off when things go sideways. Of course there are a few simple cases still lurking in my older code because simple was good enough, works reliably - in that case, but in many others having the error handling and logging really accelerated getting things reliable.

          > I feel I'd prefer to mentor a bunch of juniors - which have chances to actually learn

          I have very mixed feelings about this. Most of our juniors live in India, and they're still people with feelings that can get hurt. I've never had a problem with an AI sulking, at least not one that couldn't be cured with a reset / context flush. You can read the documents and "see what it's thinking" and spot the occasional misconception.

          > Speaking of them tokens, that's how the AI masters get paid

          Yeah, I don't know where that's going to land, I suspect it will amount to "whatever the market is willing to pay." Right now, my work tokens are costing about 40x what my Claude Code home subscription tokens cose, and the home isn't on the bulk discount plan. I hold hope that local hosting will serve as competition to keep prices reasonable, if we ever recover from the component supply constraints.

          > little chance for the said masters to take the pain of training the model towards code minimalism.

          I used to be that cynical. I think, honestly, they're harvesting our usage - determining what works for people - and training the next gen models to do more of that and less of the stuff that goes nowhere.

          --
          🌻🌻🌻🌻✌️ [google.com]
          • (Score: 2) by c0lo on Monday November 02, @08:21AM (4 children)

            by c0lo (156) Subscriber Badge on Monday November 02, @08:21AM (#1458299) Journal

            I could write a test suite that would find all kinds of edge cases and bugs, but it would generally take hours to run and once it found its bugs it wasn't the kind of thing you'd want to run before every commit.

            1. doesn't mean such a suite is useless
            2. select them on tiers - smoke test for every build, some more for EoB tests, some extra for "over weekend", go full set with release candidates

            At the rate of progress I have experienced over the last 18 months - I'm not doubting much at this point. What I'm struggling with in reduction of code volume at this point with LLMs isn't so much that they are producing "bad bloated code" - at least not after a few cleaning / review passes - but they're excessively thorough in consideration and coverage of edge cases - and I find it hard to fault them there.

            Interesting

            ... and eventually it was modularized into common helpers that covered all the edge cases...

            (oh, that passive tense) did you manage to goad Claude to do it itself after some time (ie isolate reusable error checking code) or do you need to prod it every time there's a new corner-case (or set of)?

            That error handling and logging pays off when things go sideways.

            But of course. As also pays off the formalizing the logging beyond just [log_level_prefix] and then having a set of means to filter out the hay and give you the needles (regexp-es were the most common. Maybe log analysis could involve AI nowadays)

            I have very mixed feelings about this. Most of our juniors live in India,

            I had to offload some of my work to Vancouver to above junior level. 'Twas awfully hard.
            After 6y of WFH, I know onboarding anyone remote is harder than in person, especially over diff timezones.

            and they're still people with feelings that can get hurt.

            Thanks God I haven't had those lately.

            Never had a problem with an AI sulking,

            That's not the risk I had in mind. More on the discipline required on the each-and-the-confluence-between DRY, error checking/contract programming, design and architectural manageability/reusability/flexibility (APIs, cross-cutting concerns, evolutions in data structs/formats/comms, configuration management).

            Right now, my work tokens are costing about 40x what my Claude Code home subscription tokens cose, and the home isn't on the bulk discount plan. I hold hope that local hosting will serve as competition to keep prices reasonable, if we ever recover from the component supply constraints.

            The cynic in me says the masters will take care of it by increasing the model parameters until one can't justify the investment in hardware able to run it locally (I can't justify investing on a personal local setup able to run models for non-trivial tasks).

            Then, if you read the Anthropic IPO (I didn't, just trusted Patrick Boyle with this one [youtube.com]), they positioned themselves as the ones responsible with the fate of humanity in the face of AI - doesn't sound like a position conducive to "yeah, forever release all models to the public so anyone able can run it locally".

            I used to be that cynical. I think, honestly, they're harvesting our usage - determining what works for people - and training the next gen models to do more of that and less of the stuff that goes nowhere.

            Do you trust you'll be able to keep this position in the relation with them over the (next, let's say, 5) years and they won't alter the deal?

            --
            https://www.youtube.com/@ProfSteveKeen https://soylentnews.org/~MichaelDavidCrawford
            • (Score: 3, Informative) by JoeMerchant on Monday November 02, @02:15PM (3 children)

              by JoeMerchant (3937) on Monday November 02, @02:15PM (#1458329) Journal

              >1. doesn't mean such a suite is useless
              >2. select them on tiers - smoke test for every build, some more for EoB tests, some extra for "over weekend", go full set with release candidates

              Preaching to the choir my man. Every time I float such an idea it's "too complicated, what we do meets our regulatory requirements - that extra complexity just slows us down." Nevermind that the latent bugs reaching the field have occupied 60%+ of our last 5 years...

              >did you manage to goad Claude to do it itself after some time (ie isolate reusable error checking code)

              So... I was speaking of my hand written code which Claude sees as a model while it's working... I did the isolation and reuse more and more over time, and at this stage Claude does it out of the gate with basically no prompting - maybe it's looking at "my style" for examples, but whatever it is doing, it comes out and really embraces error checking, logging, and re-use. It still takes a little nudge to get it to capture source code version info in the executable / logs, but once you have it there it uses it quite effectively and frequently.

              I guess I have DRY, SSoT, uniquely identified requirements and specifications, traceability from documents to code, modularity, version tracking and all that in some kind of CLAUDE.md or similar since 16 months ago. Back then it still ignored (lost track of) a lot of those things during implementation, these days it doesn't need reminding it just does it - I think those things are migrating from local context window instructions back into the model training - and that's likely what Anthropic is doing with our usage data, improving the models for our use cases based on how we shape their behavior through prompts / config files.

              > onboarding anyone remote is harder than in person, especially over diff timezones.

              That's a mixed bag for me. I don't really want 8 full hours of availability for training every day, I like to see what they can do on their own, and have quiet time of my own as well - you would think 9.5 hours of time-shift would give you that but I still get a lot of calls for my attention from India all through my morning and even into early afternoon, when it is 2am there.

              > More on the discipline required on the each-and-the-confluence-between DRY, error checking/contract programming...

              I have been giving the LLM agents' code reviews directly to the juniors, along with a 100-ish word summary of my top concerns. Some learn and improve quickly, others still haven't started doing their own LLM reviews before issuing the pull requests. Of course I give the LLM a short laundry list of top concerns to look for and tell it to get very specific about where the problems it's "finding" lie: source files and line numbers and clear, concise descriptions of the problems.

              > the masters will take care of it by increasing the model parameters until one can't justify the investment in hardware able to run it locally

              So far, competition out of China and some academic labs doesn't seem to care about Anthropic/Meta/X/Google and the rest's quest for hyper-large-scale as a system requirement, they're continuing to demonstrate 90% of the functionality on 1% or less of the processing power - said numbers being meaningless and in continual flux, but the real point it: today's smaller not-quite-frontier models are continuing to surpass the practical performance of the frontier models of 12 months ago. Of course you can use the frontier models to help develop the lighter weight versions and they are doing so... where it all lands is the $64,000 question (in 1785 US$s).

              >Do you trust you'll be able to keep this position in the relation with them

              Never, but... things have worked out mostly O.K. so far...

              --
              🌻🌻🌻🌻✌️ [google.com]
              • (Score: 2) by c0lo on Monday November 02, @02:42PM (2 children)

                by c0lo (156) Subscriber Badge on Monday November 02, @02:42PM (#1458332) Journal

                what we do meets our regulatory requirements - that extra complexity just slows us down

                If they test just for compliance with the regulation and not the quality of the software they produce... I wish you to enjoy playing w/ AI, will certainly help your resume when the time comes.

                So... I was speaking of my hand written code which Claude sees as a model while it's working... I did the isolation and reuse more and more over time, and at this stage Claude does it out of the gate with basically no prompting - maybe it's looking at "my style" for examples, but whatever it is doing, it comes out and really embraces error checking, logging, and re-use.

                Hmmm... intriguing. I'll probably get a paid sub to an AI and fool around for longer times in non-trivial projects, see how the relation evolves. Thanks for that.

                So far, competition out of China and some academic labs doesn't seem to care about Anthropic/Meta/X/Google

                I might actually start w/ Qwen and Gemini, just because my guts feel they have better chances to survive an AI bubble burst. :fingers-crossed:

                --
                https://www.youtube.com/@ProfSteveKeen https://soylentnews.org/~MichaelDavidCrawford
                • (Score: 3, Informative) by JoeMerchant on Monday November 02, @03:30PM (1 child)

                  by JoeMerchant (3937) on Monday November 02, @03:30PM (#1458342) Journal

                  > I wish you to enjoy playing w/ AI, will certainly help your resume when the time comes.

                  I wish the world were righteous and just and worked as you suggest. In my industry, the little companies die no matter how well they are run no matter how excellent their execution, their only real hope is to be bought by a big player. The big players stumble on - unstoppable in their momentum.

                  >probably get a paid sub to an AI and fool around for longer times in non-trivial projects

                  It wasn't a waste of money for me... time... questionable. I have gotten a lot of personal projects cleaned up and advanced more in the last year than in the previous 10. Do try to avoid the temptation of the $200/month subscriptions that give you 20x the token-resources of the $20/month - using all of that really does interfere with real life.

                  --
                  🌻🌻🌻🌻✌️ [google.com]
                  • (Score: 2) by c0lo on Monday November 02, @08:49PM

                    by c0lo (156) Subscriber Badge on Monday November 02, @08:49PM (#1458380) Journal

                    Do try to avoid the temptation of the $200/month subscriptions that give you 20x the token-resources of the $20/month - using all of that really does interfere with real life.

                    And thank you again.

                    --
                    https://www.youtube.com/@ProfSteveKeen https://soylentnews.org/~MichaelDavidCrawford
        • (Score: 1, Interesting) by Anonymous Coward on Monday November 02, @09:03AM

          by Anonymous Coward on Monday November 02, @09:03AM (#1458300)

          Anecdata: A friend of an acquaintance is a developer with a certain company and his job has been reduced to herding AI agents. He has several windows open with a different agent in each one, doing different stages of the development. All day long he just prompts them in sequence to get the code churned out. The company is absolutely delighted with this. It's cheap and "productive." He's losing his mind.

    • (Score: 2) by JoeMerchant on Monday November 02, @02:52AM (6 children)

      by JoeMerchant (3937) on Monday November 02, @02:52AM (#1458286) Journal

      >What can go wrong?

      Everything, always. New tools, nobody has experience with them - hillarious misuse stories guaranteed.

      One thing that has changed: in the not-so-distant past, the old "you can hire 9 women, but they still can't deliver you a baby in one month" adage has changed. An LLM agent woman can actually deliver your baby in one month. In more direct terms, the process of requirements review that used to take 2 weeks can now be realistically done in 2 days, the processes of code review, unit and regression test construction, formal validation, all of them are accelerated. Where we used to meet schedule by making intelligent tradeoffs, reviewing only "the most critical code", picking and chosing what we build unit tests for, representative samples of expected functionality for validation testing, automated regression tests only when judged "likely to recur in the future" - we can now apply those processess much more liberally, and should - because those processes are what reduce errors from sloppy code generation...

      If management listens to me, we'll still be delivering a baby in 9 months, maybe 7 or 8 if we're lucky, but the baby we deliver will be thoroughly reviewed, completely wrapped in unit and regression tests, it will still have some bugs that get found later - but hopefully less than we used to deliver, and it _should_ be easier to maintain with less surprise regressions and more formal requirements and specifications to base the user facing documentation on - less disconnects between the various layers of documentation and actual functionality.

      --
      🌻🌻🌻🌻✌️ [google.com]
      • (Score: 2) by c0lo on Monday November 02, @03:22AM (5 children)

        by c0lo (156) Subscriber Badge on Monday November 02, @03:22AM (#1458290) Journal

        completely wrapped in unit and regression tests

        I doubt the completeness claim - things can get wrong in a vastly larger number of ways than the restricted "all good" cases defined by the design/requirements.

        I would accept the claim if you were to reduce it to "one/two order of magnitude better test coverage", but even then I'd point out that automatic testing bears a cost in itself and one absolutely can hit the budget limits for testing (we've hit them twice in the past 7 years - no more physical room for in-house test farms followed by blowing past the budget for the cloud CI infrastructure).

        --
        https://www.youtube.com/@ProfSteveKeen https://soylentnews.org/~MichaelDavidCrawford
        • (Score: 2) by JoeMerchant on Monday November 02, @04:19AM (4 children)

          by JoeMerchant (3937) on Monday November 02, @04:19AM (#1458293) Journal

          Completely wrapped doesn't mean completely protected, you can cover some negative cases but never all of them.

          >one/two order of magnitude better test coverage

          Three or four for me, I'm old-school - barely ever wrote unit tests in the past.

          >one absolutely can hit the budget limits for testing (we've hit them twice in the past 7 years - no more physical room for in-house test farms followed by blowing past the budget for the cloud CI infrastructure).

          Yikes, we're much less resource intensive, our build server of the past 10 years is a fanless Core i7 box - we bought a newer faster box to replace it a couple of years ago but inertia has left the old one as the main build server still today. A full system build takes about 30-40 minutes, including tests. More tests won't be sinking the new server anytime soon.

          --
          🌻🌻🌻🌻✌️ [google.com]
          • (Score: 2) by c0lo on Monday November 02, @07:09AM (3 children)

            by c0lo (156) Subscriber Badge on Monday November 02, @07:09AM (#1458296) Journal

            A full system build takes about 30-40 minutes,...

            Ours takes about 20 mins, except that our build makes use of Ninja build [ninja-build.org] on top of SN-DBS (sony distributed building system) with 200+ dev workstations (pretty beefed up each) enrolled to take part on spare CPU/IO time (which is plenty when you're just coding or incrementally build your changes locally or sleeping into your timezone). There are 10+ builds in the same time, almost all the time except the weekend.

            ... including tests

            Ahem... our yearly budget for CI is in the 8 figures territory. A full test round takes about 8 hours, including the test data "cooking" (generating/transforming once the test assets into formats that are consumable at the level of unit/functional tests).

            --
            https://www.youtube.com/@ProfSteveKeen https://soylentnews.org/~MichaelDavidCrawford
            • (Score: 2) by JoeMerchant on Monday November 02, @01:49PM (2 children)

              by JoeMerchant (3937) on Monday November 02, @01:49PM (#1458326) Journal

              >A full test round takes about 8 hours, including the test data "cooking"

              Yeah, that's my idea of "real testing." All the cool kids think that a unit test should execute in under 100 milliseconds and on the server instead of the target hardware - I ask them: what's the point of having unit tests that miss problems? They just say "it's too expensive to do that kind of testing." I go check the amount of my last direct deposit and sigh... it isn't good for job security to call your colleagues idiot moron children.

              --
              🌻🌻🌻🌻✌️ [google.com]
              • (Score: 2) by c0lo on Monday November 02, @02:14PM (1 child)

                by c0lo (156) Subscriber Badge on Monday November 02, @02:14PM (#1458328) Journal

                All the cool kids think that a unit test should execute in under 100 milliseconds

                When one hits one's head on the reality enough times, the reality ends seeping into the head through the cracks. It's usually called growing up (or old. Why not both, actually? Because some never grow up and finish professionally dead. Or managers, but I repeat myself)

                Now, to be honest, with the test data set being all cooked, about 80% of our unit tests execute under 100 millies, with the maximum (of the rest) being under 5 seconds. Except that we literally have them for execution in the millions.

                (I don't know why I keep using "we/ours", 2 weeks ago I quit with the intention to retire early)

                --
                https://www.youtube.com/@ProfSteveKeen https://soylentnews.org/~MichaelDavidCrawford
  • (Score: 5, Informative) by c0lo on Monday November 02, @02:15AM (3 children)

    by c0lo (156) Subscriber Badge on Monday November 02, @02:15AM (#1458284) Journal

    https://codeberg.org/ [codeberg.org]

    • LLM-generated content is being restricted.
    • Cryptocurrency projects are no longer allowed.

    About [codeberg.org]

    Codeberg e.V. is a registered non-profit association based in Berlin, Germany. You don't have to be a member of the association in order to use Codeberg.org or to contribute to the development of the platform, but if you want you can join Codeberg e.V. to support the project financially, be informed about Codeberg and, optionally, to actively contribute to the association.

    Codeberg members can also take part in the decisions of the platform as explained in the bylaws, and they elect the presidium and board of the platform, thus Codeberg can be considered as community-owned.

    ----

    https://codefloe.com/ [codefloe.com]

    • Hosted in Germany, under EU law. Everything runs and is stored on Hetzner in Germany, so your data falls under German data protection law and the GDPR.
    • Infrastructure you can read The Ansible playbooks, OpenTofu modules, CI pipelines and monitoring configuration that drive CodeFloe are public repositories.
    • No feature gates Private repositories, CI/CD, hosted pages, package registries and preview environments are free for everyone. Paid tiers only cover resource costs for storage and CI/CD.

    Where we stand on AI

    • We see AI as a tool, not a threat - We use AI tools ourselves, in this platform and in our own work. You are welcome to use them here too. Nothing about CodeFloe is built to make that harder.
    • Your code is never training data We do not train models on anything you store here, and we do not sell or hand your data to anyone who would. Not now, and not later: it is not a business we intend to be in.
    • Abusive scrapers are not welcome Aggressive crawlers can flatten a forge. We tackle them at the edge, so the platform stays fast for the people actually using it.

    ----

    https://framagit.org/ [framagit.org] about [devproblems.com]

    Framagit is a Git hosting service run by the French nonprofit Framasoft. It is built on GitLab and is often used by open-source and civic-tech communities that want a European nonprofit operator.

    ---.

    https://pushin.eu/ [pushin.eu] - still in the emerging phase [ycombinator.com]

    --
    https://www.youtube.com/@ProfSteveKeen https://soylentnews.org/~MichaelDavidCrawford
    • (Score: 4, Touché) by Thexalon on Monday November 02, @05:48PM (2 children)

      by Thexalon (636) on Monday November 02, @05:48PM (#1458352)

      And, of course, it's pretty easy to set up your own git repository on your own hardware or your own cloud server. Very much on purpose - this idea that git is supposed to be about centralized repositories under the control of a single organization is the exact opposite of Linus' intent in writing it.

      --
      "Think of how stupid the average person is. Then realize half of 'em are stupider than that." - George Carlin
      • (Score: 3, Funny) by turgid on Tuesday November 03, @10:34AM (1 child)

        by turgid (4318) Subscriber Badge on Tuesday November 03, @10:34AM (#1458443) Journal

        And, of course, it's pretty easy to set up your own git repository on your own hardware

        On Slackware, it's trivial:

        cd <directory>
        git init .

        On Ubuntu it's much harder. You have to mess about with sudo and apt install.

        • (Score: 3, Insightful) by Thexalon on Thursday November 05, @06:41PM

          by Thexalon (636) on Thursday November 05, @06:41PM (#1458659)

          I mean, I was thinking more of instances where you set up a user who owns the repo and ssh keys so that you can use it as an upstream remote, but yes, that's enough for local usage.

          --
          "Think of how stupid the average person is. Then realize half of 'em are stupider than that." - George Carlin
(1)