summaryrefslogtreecommitdiffstats
path: root/2026/captions
diff options
context:
space:
mode:
authorSacha Chua <sacha@sachachua.com>2026-11-16 14:03:19 -0500
committerSacha Chua <sacha@sachachua.com>2026-11-16 14:03:28 -0500
commitf8c32a938c6be615a7fb10978ca1b13d9777fd73 (patch)
tree844a29c3ee7783f044e8c0200ed7fec88b2e0958 /2026/captions
parent0d63edb3564c947e31fa0b29d8241b020da546a8 (diff)
downloademacsconf-wiki-master.tar.xz
emacsconf-wiki-master.zip
really publish remoto and hutchHEADmaster
Diffstat (limited to '')
-rw-r--r--2026/captions/emacsconf-2026-hutch--hutch-code-reviews-for-the-mildly-disenfranchised--akshay-gupta--main--chapters.vtt41
-rw-r--r--2026/captions/emacsconf-2026-hutch--hutch-code-reviews-for-the-mildly-disenfranchised--akshay-gupta--main.vtt758
-rw-r--r--2026/captions/emacsconf-2026-remoto--el-remoto-browse-any-github-repos-without-cloning-them--ag--main--chapters.vtt32
-rw-r--r--2026/captions/emacsconf-2026-remoto--el-remoto-browse-any-github-repos-without-cloning-them--ag--main.vtt645
4 files changed, 1476 insertions, 0 deletions
diff --git a/2026/captions/emacsconf-2026-hutch--hutch-code-reviews-for-the-mildly-disenfranchised--akshay-gupta--main--chapters.vtt b/2026/captions/emacsconf-2026-hutch--hutch-code-reviews-for-the-mildly-disenfranchised--akshay-gupta--main--chapters.vtt
new file mode 100644
index 00000000..afb34b6a
--- /dev/null
+++ b/2026/captions/emacsconf-2026-hutch--hutch-code-reviews-for-the-mildly-disenfranchised--akshay-gupta--main--chapters.vtt
@@ -0,0 +1,41 @@
+WEBVTT
+
+
+00:00:00.071 --> 00:02:14.025
+Introduction
+
+00:02:14.825 --> 00:03:25.462
+Typical coding agent review
+
+00:03:28.261 --> 00:04:18.130
+Hutch
+
+00:04:19.220 --> 00:04:50.950
+Tools
+
+00:04:51.291 --> 00:05:12.425
+Report
+
+00:05:14.317 --> 00:05:34.800
+Comments
+
+00:05:37.385 --> 00:06:11.953
+Additional reviews
+
+00:06:13.194 --> 00:07:01.263
+Storing review notes
+
+00:07:03.876 --> 00:07:36.482
+Formal eval and Perfetto
+
+00:07:38.342 --> 00:09:29.750
+Tracing performance with Perfetto
+
+00:09:32.417 --> 00:09:53.294
+SQL queries
+
+00:09:55.100 --> 00:11:07.694
+Model comparison
+
+00:11:09.356 --> 00:11:43.462
+Summary
diff --git a/2026/captions/emacsconf-2026-hutch--hutch-code-reviews-for-the-mildly-disenfranchised--akshay-gupta--main.vtt b/2026/captions/emacsconf-2026-hutch--hutch-code-reviews-for-the-mildly-disenfranchised--akshay-gupta--main.vtt
new file mode 100644
index 00000000..2fb37039
--- /dev/null
+++ b/2026/captions/emacsconf-2026-hutch--hutch-code-reviews-for-the-mildly-disenfranchised--akshay-gupta--main.vtt
@@ -0,0 +1,758 @@
+WEBVTT captioned by sacha
+
+NOTE Introduction
+
+00:00:00.071 --> 00:00:00.572
+I'm Akshay.
+
+00:00:01.112 --> 00:00:03.694
+I'm going to talk about a local code review tool
+
+00:00:03.774 --> 00:00:06.477
+I've been building for Emacs for the past five
+
+00:00:06.557 --> 00:00:08.219
+months or so.
+
+00:00:08.239 --> 00:00:10.661
+I've previously been an Emacs user for about 10
+
+00:00:10.701 --> 00:00:13.704
+years, and I use it as a daily driver for most of
+
+00:00:13.724 --> 00:00:15.205
+my code and prose editing.
+
+00:00:16.306 --> 00:00:19.509
+When I say a local review tool, I mean local
+
+00:00:19.529 --> 00:00:22.272
+to Emacs and not necessarily local LLMs on
+
+00:00:22.292 --> 00:00:24.334
+device, even though I think that's theoretically
+
+00:00:24.374 --> 00:00:25.014
+possible today.
+
+00:00:26.475 --> 00:00:28.797
+By code reviews, I specifically mean the
+
+00:00:28.837 --> 00:00:32.339
+thing we do on pull requests and on tools like
+
+00:00:32.378 --> 00:00:34.581
+Gerrit and Phabricator in the past.
+
+00:00:36.242 --> 00:00:38.924
+In other words, a system for gating code before
+
+00:00:38.984 --> 00:00:40.405
+it's merged to be released.
+
+00:00:42.778 --> 00:00:45.321
+I should say that there have been many good
+
+00:00:45.381 --> 00:00:48.384
+talks about LLMs in AI and what that means for
+
+00:00:48.564 --> 00:00:51.266
+Emacs, so I'm not going to relitigate any of
+
+00:00:51.307 --> 00:00:52.207
+those arguments again.
+
+00:00:52.728 --> 00:00:55.551
+I approach this, as many others in the community,
+
+00:00:55.611 --> 00:00:58.794
+by letting go of semantic boundaries under which
+
+00:00:58.934 --> 00:01:02.918
+Emacs is considered an editor, but just rather a
+
+00:01:03.158 --> 00:01:05.239
+mutable productivity environment instead.
+
+00:01:07.072 --> 00:01:09.074
+But I do want to briefly show you my experience
+
+00:01:09.114 --> 00:01:11.415
+with existing code review tools that are around
+
+00:01:12.296 --> 00:01:15.718
+because I have a bit of a love-hate relationship
+
+00:01:15.759 --> 00:01:16.079
+with them.
+
+00:01:16.819 --> 00:01:19.741
+It can be pretty overwhelming to keep up with the
+
+00:01:19.841 --> 00:01:23.324
+speed at which the amount of code gets produced
+
+00:01:23.384 --> 00:01:26.366
+daily these days, especially if you work in a
+
+00:01:26.426 --> 00:01:31.109
+team, but also the amount of review text that now
+
+00:01:31.169 --> 00:01:33.291
+gets attached to that code.
+
+00:01:34.639 --> 00:01:37.942
+In my experience, even though the noise to
+
+00:01:37.962 --> 00:01:39.442
+signal for these tools worst case
+
+00:01:39.443 --> 00:01:42.123
+is like 10 to 1, I don't feel like
+
+00:01:42.124 --> 00:01:44.345
+we can get rid of them entirely
+
+00:01:44.346 --> 00:01:45.787
+because those ones can sometimes be
+
+00:01:45.827 --> 00:01:47.828
+pretty sharp and important.
+
+00:01:49.330 --> 00:01:53.593
+But I do think there's value in sort of shifting
+
+00:01:53.613 --> 00:01:56.495
+the process to the left to borrow a term from the
+
+00:01:56.555 --> 00:01:58.116
+DevRel industry.
+
+00:01:59.997 --> 00:02:01.818
+Because it's useful to see the work you did
+
+00:02:02.699 --> 00:02:05.980
+before you create more work for others, for your
+
+00:02:06.020 --> 00:02:07.001
+colleagues and your peers.
+
+00:02:07.821 --> 00:02:09.942
+Many of these flows can now be easily
+
+00:02:10.002 --> 00:02:12.744
+unlocked in Emacs given the excellent building
+
+00:02:12.784 --> 00:02:14.025
+blocks we have today.
+
+NOTE Typical coding agent review
+
+00:02:14.825 --> 00:02:17.587
+But before I walk you through how a Hutch
+
+00:02:17.627 --> 00:02:20.328
+workflow works, I want to show you what a typical
+
+00:02:20.408 --> 00:02:22.749
+coding agent review looks like today.
+
+00:02:23.490 --> 00:02:25.891
+These examples are from a real open source
+
+00:02:25.951 --> 00:02:28.032
+product that I've been building for the past
+
+00:02:28.092 --> 00:02:28.653
+three years.
+
+00:02:29.671 --> 00:02:31.792
+This is one pull request from the project.
+
+00:02:32.432 --> 00:02:35.214
+It's a single review pass by a review agent.
+
+00:02:35.834 --> 00:02:37.035
+Has 23 comments.
+
+00:02:37.775 --> 00:02:42.317
+Every single one of them flagged as a potential issue.
+
+00:02:42.397 --> 00:02:44.178
+Five of them are marked critical.
+
+00:02:45.359 --> 00:02:48.621
+This is technically a review and it's probably a
+
+00:02:48.661 --> 00:02:49.261
+decent one.
+
+00:02:50.041 --> 00:02:53.343
+But it's also a bit of a DOS attack on my attention.
+
+00:02:54.982 --> 00:02:57.105
+If we look behind the scenes on one of those
+
+00:02:57.145 --> 00:03:00.870
+critical flags, it ran 40-something shell scripts.
+
+00:03:00.930 --> 00:03:01.991
+It ran web searches.
+
+00:03:02.452 --> 00:03:04.895
+It wrote a little prompt for another AI to go and
+
+00:03:04.955 --> 00:03:05.536
+fix the thing.
+
+00:03:06.297 --> 00:03:08.540
+All of it to tell me that one method was
+
+00:03:08.600 --> 00:03:10.302
+undefined and would crash at runtime.
+
+00:03:11.687 --> 00:03:14.650
+But at the end of all of it, it was still incorrect.
+
+00:03:14.730 --> 00:03:17.373
+The method comes from a Ruby gem, and this was
+
+00:03:17.413 --> 00:03:20.156
+simply my reply in six words, which to its credit
+
+00:03:20.336 --> 00:03:23.059
+it apologized and updated its memory for.
+
+00:03:23.960 --> 00:03:25.462
+This is not a one-off thing.
+
+NOTE Hutch
+
+00:03:28.261 --> 00:03:30.962
+So I built Hutch for myself.
+
+00:03:31.022 --> 00:03:31.602
+It's up here.
+
+00:03:31.742 --> 00:03:34.383
+It lives in Emacs and on top of Magit.
+
+00:03:35.403 --> 00:03:38.183
+You decide when you want to trigger it and any
+
+00:03:38.303 --> 00:03:41.604
+obvious to relatively bigger issues get caught up
+
+00:03:41.684 --> 00:03:44.065
+front before you push your code up.
+
+00:03:45.305 --> 00:03:47.786
+When it does have something to say, it's usually
+
+00:03:47.806 --> 00:03:50.286
+a full-size diff you can directly apply.
+
+00:03:51.107 --> 00:03:53.927
+But even if it doesn't, it's prose for you to
+
+00:03:53.987 --> 00:03:56.708
+read and not five other people collectively.
+
+00:03:58.338 --> 00:04:00.479
+So here's how you kick it off.
+
+00:04:00.519 --> 00:04:02.681
+When you're ready to commit, you open up the
+
+00:04:02.701 --> 00:04:06.343
+change diffs in Magit, you stage your changes as
+
+00:04:06.503 --> 00:04:09.285
+usual, and then select the hutch action and
+
+00:04:09.345 --> 00:04:10.345
+trigger a new review.
+
+00:04:11.666 --> 00:04:13.627
+Once this is done, you'll see a new hutch buffer
+
+00:04:13.667 --> 00:04:13.987
+show up.
+
+00:04:14.628 --> 00:04:17.228
+I also have a debug log buffer that can be
+
+00:04:17.309 --> 00:04:18.130
+optionally enabled.
+
+NOTE Hutch tools
+
+00:04:19.220 --> 00:04:21.642
+This review has been run on the Hutch codebase
+
+00:04:21.682 --> 00:04:24.224
+itself and the agent is sifting through the code
+
+00:04:24.404 --> 00:04:26.126
+and you can see a bunch of tools being called
+
+00:04:26.206 --> 00:04:28.608
+like `read_file` and `search_codebase` and stuff.
+
+00:04:29.649 --> 00:04:32.771
+There's around five tools that Hutch exposes, one
+
+00:04:32.812 --> 00:04:35.554
+of which is called `surrounding_context` that uses
+
+00:04:35.694 --> 00:04:37.636
+TreeSitter if you have that installed to
+
+00:04:37.836 --> 00:04:40.698
+increasingly fetch more context around certain
+
+00:04:40.718 --> 00:04:43.741
+suspicious lines without needing to read whole
+
+00:04:43.781 --> 00:04:44.822
+files and waste tokens.
+
+00:04:45.927 --> 00:04:48.169
+You'll see that it took around five rounds to
+
+00:04:48.229 --> 00:04:50.950
+finish this review this time.
+
+NOTE
+Hutch walkthrough
+#+SCREENSHOT: emacsconf-2026-hutch--hutch-code-reviews-for-the-mildly-disenfranchised--akshay-gupta--reencoded-00-04-51-300.jpg
+#+SCREENSHOT_TIME: 00:04:51.300
+
+00:04:51.291 --> 00:04:53.532
+Once the Hutch buffer is ready, you'd see
+
+00:04:53.572 --> 00:04:54.253
+something like this.
+
+00:04:54.553 --> 00:04:56.554
+It'll report the total tokens at the top.
+
+00:04:57.055 --> 00:04:59.236
+You'll see the findings organized under
+
+00:04:59.276 --> 00:05:01.538
+suggestions, comments, and LGTM
+
+00:05:01.998 --> 00:05:02.878
+(which is "looks good to me").
+
+00:05:03.759 --> 00:05:06.361
+Just like Magit, you can open the submenus up.
+
+00:05:06.961 --> 00:05:09.263
+If you agree with the suggestion and the
+
+00:05:09.343 --> 00:05:12.425
+attached patch, you can hit `m` to queue it up.
+
+NOTE Applying suggestions
+
+00:05:14.317 --> 00:05:16.119
+Comments are just text for you to read and
+
+00:05:16.159 --> 00:05:17.260
+determine if they make sense.
+
+00:05:17.601 --> 00:05:19.723
+These can also be copied and sent off to someone
+
+00:05:19.763 --> 00:05:22.286
+else to take a look or another agent that you
+
+00:05:22.306 --> 00:05:23.007
+might have running.
+
+00:05:24.568 --> 00:05:27.091
+Once all the suggestions are queued up, you
+
+00:05:27.131 --> 00:05:30.395
+can apply them all with the key binding `A`
+
+00:05:31.236 --> 00:05:33.598
+and the patches get applied to the staged files
+
+00:05:33.679 --> 00:05:34.800
+directly, bottom-up.
+
+NOTE Additional reviews
+
+00:05:37.385 --> 00:05:40.727
+But perhaps at this point, I'm not convinced yet.
+
+00:05:40.887 --> 00:05:43.308
+I'd like to run another review on these corrected
+
+00:05:43.368 --> 00:05:45.009
+changes, and I can do so.
+
+00:05:52.573 --> 00:05:55.295
+So here we get another suggestion to fix that
+
+00:05:55.335 --> 00:05:57.596
+wasn't caught the last time.
+
+00:05:57.636 --> 00:06:00.878
+We queue it up, apply it, and verify that it's
+
+00:06:00.938 --> 00:06:04.019
+lodged in, and then run one more final review
+
+00:06:04.059 --> 00:06:04.820
+before we commit.
+
+00:06:07.010 --> 00:06:08.671
+So at this point, I'm happy with the changes.
+
+00:06:08.711 --> 00:06:10.812
+The comments are not actionable, so I go ahead
+
+00:06:11.112 --> 00:06:11.953
+and write my commit.
+
+NOTE
+Storing review notes
+#+SCREENSHOT: emacsconf-2026-hutch--hutch-code-reviews-for-the-mildly-disenfranchised--akshay-gupta--reencoded-00-06-22-900.jpg
+#+SCREENSHOT_TIME: 00:06:22.900
+
+00:06:13.194 --> 00:06:15.155
+Another useful thing you can enable as a
+
+00:06:15.235 --> 00:06:17.897
+post-commit hook is having your review notes be
+
+00:06:17.937 --> 00:06:20.218
+stored within Git as a separate ref.
+
+00:06:21.219 --> 00:06:22.859
+In the terminal pane on the right, you can see
+
+00:06:22.900 --> 00:06:25.821
+our last review was durably kept for posterity.
+
+00:06:26.522 --> 00:06:29.003
+This can be useful for maintaining a related
+
+00:06:29.043 --> 00:06:31.385
+history of reviews along with your commits and
+
+00:06:31.465 --> 00:06:33.666
+for sharing the reviews with your team as well.
+
+00:06:35.874 --> 00:06:36.314
+That's it.
+
+00:06:36.554 --> 00:06:40.075
+It's hopefully pretty simple and intuitive to get
+
+00:06:40.116 --> 00:06:41.356
+started with.
+
+00:06:41.396 --> 00:06:43.257
+There's a lot more things to configure that you
+
+00:06:43.277 --> 00:06:44.197
+can find in the README.
+
+00:06:45.177 --> 00:06:47.678
+If you're the kind that never wants to go back to
+
+00:06:47.718 --> 00:06:50.679
+writing code in editors by hand again, but are
+
+00:06:50.779 --> 00:06:53.400
+still the kind that enjoys programming on their
+
+00:06:53.440 --> 00:06:57.101
+own projects, this might be a nice thin layer to
+
+00:06:57.121 --> 00:06:59.862
+get your work verified without over-relying on
+
+00:07:00.003 --> 00:07:01.263
+LLMs for everything.
+
+NOTE Formal eval and Perfetto
+
+00:07:03.876 --> 00:07:06.498
+On a separate note, Hutch does have
+
+00:07:06.499 --> 00:07:08.121
+a formal eval pipeline.
+
+00:07:09.302 --> 00:07:12.205
+If you set `hutch-trace-dir` and give it a
+
+00:07:12.245 --> 00:07:15.729
+path, it can emit Perfetto traces that can then
+
+00:07:15.789 --> 00:07:18.552
+be piped into some of the eval scripts that we
+
+00:07:18.632 --> 00:07:22.737
+have to get fresh data about agent quality.
+
+00:07:23.812 --> 00:07:29.016
+We can also use the traces to visualize the agent itself.
+
+00:07:29.116 --> 00:07:33.459
+They're useful for both post runtime observability
+
+00:07:33.460 --> 00:07:36.482
+and to fetch data for verifying its correctness.
+
+NOTE
+Tracing performance with Perfetto
+#+SCREENSHOT: emacsconf-2026-hutch--hutch-code-reviews-for-the-mildly-disenfranchised--akshay-gupta--reencoded-00-08-10-767.jpg
+#+SCREENSHOT_TIME: 00:08:10.767
+
+00:07:38.342 --> 00:07:41.603
+Perfetto is an open source performance tracing tool
+
+00:07:41.604 --> 00:07:45.266
+that comes with a bundled SQL query engine as well.
+
+00:07:46.087 --> 00:07:49.129
+You see this typically being used in Android
+
+00:07:49.209 --> 00:07:52.351
+and on Chrome for performance profiling and things.
+
+00:07:53.251 --> 00:07:56.153
+But it turns out it works pretty well as a simple
+
+00:07:56.553 --> 00:07:59.735
+agent tracing tool because it's entirely
+
+00:07:59.875 --> 00:08:00.856
+infrastructure-free.
+
+00:08:01.476 --> 00:08:04.658
+You just emit JSON in its own trace format and
+
+00:08:04.698 --> 00:08:07.480
+you can load it up on ui.perfetto.dev.
+
+00:08:08.823 --> 00:08:10.726
+So once you load it up, this is what the
+
+00:08:10.766 --> 00:08:11.987
+visualization looks like.
+
+00:08:12.228 --> 00:08:15.793
+There's a timeline of what the agent did, what
+
+00:08:15.913 --> 00:08:19.357
+tools it called, the slices of LLM calls, and
+
+00:08:19.558 --> 00:08:22.742
+down below the tokens that were consumed.
+
+00:08:24.710 --> 00:08:26.771
+We can zoom into one of these tool calls and we
+
+00:08:26.791 --> 00:08:29.392
+can check out what all the agent did.
+
+00:08:30.013 --> 00:08:34.475
+So here we see a `search_codebase` tool call with
+
+00:08:34.515 --> 00:08:37.135
+that search string and it looks like it took
+
+00:08:37.196 --> 00:08:39.477
+about 400 milliseconds to finish.
+
+00:08:40.897 --> 00:08:43.719
+If we move further along, we'll see another
+
+00:08:43.779 --> 00:08:48.120
+`search_codebase` call being called with slightly
+
+00:08:48.160 --> 00:08:49.501
+different arguments.
+
+00:08:49.541 --> 00:08:51.562
+It took a little longer to finish.
+
+00:08:53.357 --> 00:08:57.321
+As you can see, as we move along the trace,
+
+00:08:57.361 --> 00:08:59.903
+the number of tokens consumed are increasing over time.
+
+00:09:01.164 --> 00:09:05.928
+If we go all the way to the end, we'll see two markers.
+
+00:09:06.028 --> 00:09:09.171
+The first one is where the LLM calls end.
+
+00:09:10.672 --> 00:09:12.954
+This is where the agent is done with all its findings.
+
+00:09:13.715 --> 00:09:16.337
+At this point, no models are being called and
+
+00:09:16.517 --> 00:09:18.819
+everything that we got back is being now
+
+00:09:18.859 --> 00:09:20.020
+processed by Hutch.
+
+00:09:21.081 --> 00:09:22.923
+A few milliseconds later, we see the final
+
+00:09:22.963 --> 00:09:25.386
+findings that are eventually rendered onto the
+
+00:09:25.426 --> 00:09:26.106
+Hutch buffer.
+
+00:09:26.286 --> 00:09:29.750
+These are verified and are sanitized for correctness.
+
+NOTE Perfetto SQL queries
+
+00:09:32.417 --> 00:09:37.301
+You can also run SQL queries on Perfetto.
+
+00:09:37.521 --> 00:09:40.444
+Here we have a very simple query to fetch all the
+
+00:09:40.484 --> 00:09:41.565
+tool call durations.
+
+00:09:42.185 --> 00:09:45.106
+Here we can see the `search_codebase`
+
+00:09:45.107 --> 00:09:48.470
+tool call seems to have taken the most amount of time
+
+00:09:48.510 --> 00:09:49.571
+in this particular trace.
+
+00:09:50.071 --> 00:09:53.294
+`read_file` seems to have been called the most.
+
+NOTE Comparison of model findings
+
+00:09:55.100 --> 00:09:57.342
+Before I wrap this section up, I wanted to share
+
+00:09:57.382 --> 00:10:00.064
+a couple of quick findings from slightly older
+
+00:10:00.124 --> 00:10:03.386
+models run on Martian's code review dataset
+
+00:10:03.387 --> 00:10:06.890
+of 40 PRs across different difficulty levels.
+
+00:10:07.911 --> 00:10:10.193
+You can find more detailed benchmarks and eval
+
+00:10:10.213 --> 00:10:12.635
+runs in the README if anyone's curious.
+
+00:10:14.137 --> 00:10:16.479
+This first graph shows that different models tend
+
+00:10:16.499 --> 00:10:17.480
+to find different issues.
+
+00:10:18.445 --> 00:10:21.307
+So not all of them necessarily agree on what a
+
+00:10:21.367 --> 00:10:22.208
+real bug is.
+
+00:10:23.188 --> 00:10:25.690
+But if you theoretically combine their results,
+
+00:10:25.970 --> 00:10:29.192
+you would get about a 50% jump in the bug hit rate.
+
+00:10:30.853 --> 00:10:34.016
+The second graph shows how many rounds each model
+
+00:10:34.076 --> 00:10:36.077
+took to grind through the dataset.
+
+NOTE
+Rounds vs. token cost
+#+SCREENSHOT: emacsconf-2026-hutch--hutch-code-reviews-for-the-mildly-disenfranchised--akshay-gupta--reencoded-00-10-36-700.jpg
+#+SCREENSHOT_TIME: 00:10:36.700
+
+00:10:36.697 --> 00:10:40.360
+Here, GPT takes many more rounds and Opus is far fewer.
+
+00:10:41.881 --> 00:10:43.922
+But rounds can be misleading because they don't
+
+00:10:43.962 --> 00:10:44.723
+track cost.
+
+00:10:46.221 --> 00:10:49.142
+Opus finishes in the fewest rounds but burns the
+
+00:10:49.182 --> 00:10:52.604
+most input tokens by far, likely because of the
+
+00:10:52.644 --> 00:10:54.105
+growing conversation context.
+
+00:10:55.065 --> 00:10:56.806
+GLM does it in under 2 million.
+
+00:10:57.627 --> 00:10:59.968
+So on the axis that actually costs money, GLM is
+
+00:10:59.988 --> 00:11:02.169
+the cheapest of the three, which is the reason
+
+00:11:02.209 --> 00:11:04.210
+why it's my default model of choice.
+
+00:11:05.891 --> 00:11:07.694
+That's it from me for today.
+
+NOTE Summary
+
+00:11:09.356 --> 00:11:12.480
+Rich Hickey wrote a post recently arguing that LLMs
+
+00:11:12.741 --> 00:11:15.184
+aren't tools, they're automation.
+
+00:11:15.985 --> 00:11:19.090
+They do the work and they take the thinking with it.
+
+00:11:20.961 --> 00:11:23.724
+Hutch is my attempt at keeping as much of this
+
+00:11:23.804 --> 00:11:27.087
+process under your control as possible.
+
+00:11:27.327 --> 00:11:30.590
+You choose when to run it, you read what it says,
+
+00:11:30.950 --> 00:11:33.132
+and then you apply the fixes yourself.
+
+00:11:33.993 --> 00:11:38.678
+It's a small local "tool-enough" for people still
+
+00:11:38.718 --> 00:11:41.140
+reading what they write, and it's available to
+
+00:11:41.180 --> 00:11:43.462
+download on MELPA and on GitHub.
diff --git a/2026/captions/emacsconf-2026-remoto--el-remoto-browse-any-github-repos-without-cloning-them--ag--main--chapters.vtt b/2026/captions/emacsconf-2026-remoto--el-remoto-browse-any-github-repos-without-cloning-them--ag--main--chapters.vtt
new file mode 100644
index 00000000..0179993b
--- /dev/null
+++ b/2026/captions/emacsconf-2026-remoto--el-remoto-browse-any-github-repos-without-cloning-them--ag--main--chapters.vtt
@@ -0,0 +1,32 @@
+WEBVTT
+
+
+00:00:01.067 --> 00:00:31.214
+Introduction
+
+00:00:31.914 --> 00:01:25.042
+Remote files
+
+00:01:26.282 --> 00:05:04.913
+Accessing GitHub
+
+00:05:05.353 --> 00:06:43.708
+Opening Github
+
+00:06:44.008 --> 00:09:23.457
+Embark
+
+00:09:23.837 --> 00:09:42.364
+Codeberg
+
+00:09:42.484 --> 00:09:47.025
+Issues
+
+00:09:47.085 --> 00:11:20.761
+Plans
+
+00:11:21.382 --> 00:11:44.288
+Why not a TRAMP backend?
+
+00:11:45.008 --> 00:12:10.382
+Why not plain HTTP?
diff --git a/2026/captions/emacsconf-2026-remoto--el-remoto-browse-any-github-repos-without-cloning-them--ag--main.vtt b/2026/captions/emacsconf-2026-remoto--el-remoto-browse-any-github-repos-without-cloning-them--ag--main.vtt
new file mode 100644
index 00000000..00df4ccd
--- /dev/null
+++ b/2026/captions/emacsconf-2026-remoto--el-remoto-browse-any-github-repos-without-cloning-them--ag--main.vtt
@@ -0,0 +1,645 @@
+WEBVTT captioned by sachac
+
+NOTE Introduction
+
+00:00:01.067 --> 00:00:03.628
+Okay, hello people of Emacs, my favorite kind of people.
+
+00:00:04.268 --> 00:00:06.869
+I would like to show you this thing I've been
+
+00:00:06.909 --> 00:00:08.829
+messing around with lately.
+
+00:00:09.080 --> 00:00:11.830
+So I hope this is big enough.
+
+00:00:13.270 --> 00:00:16.911
+Most of you probably know that you can do this in Emacs.
+
+00:00:17.071 --> 00:00:19.512
+When you do `C-x C-f`, which is
+
+00:00:19.712 --> 00:00:21.953
+normal `find-file` dialog,
+
+00:00:22.233 --> 00:00:25.531
+you can browse folders and files
+
+00:00:25.532 --> 00:00:30.294
+and that's a built-in feature of Emacs.
+
+00:00:30.414 --> 00:00:31.214
+Everybody knows about it.
+
+NOTE Remote files
+
+00:00:31.914 --> 00:00:37.317
+Now, you can also type /ssh: and
+
+00:00:37.337 --> 00:00:39.898
+then if you type the name of the machine and you
+
+00:00:39.938 --> 00:00:43.079
+have access to it, you can actually browse things
+
+00:00:43.859 --> 00:00:45.199
+on that remote machine
+
+00:00:45.200 --> 00:00:47.561
+as if you're browsing things locally.
+
+00:00:48.001 --> 00:00:51.783
+Similarly, you can do sudo.
+
+00:00:52.583 --> 00:00:55.264
+You can do Docker.
+
+00:00:55.344 --> 00:00:57.685
+You can do FTP.
+
+00:00:58.345 --> 00:00:59.705
+You can do SFTP.
+
+00:00:59.805 --> 00:01:03.346
+You can do SMB to access some other sources.
+
+00:01:03.446 --> 00:01:07.208
+You can do Podman, Kubernetes, some other things.
+
+00:01:07.268 --> 00:01:08.128
+You can do SCP.
+
+00:01:09.108 --> 00:01:12.269
+This feature is built-in into Emacs.
+
+00:01:14.070 --> 00:01:15.030
+It's been there forever.
+
+00:01:15.630 --> 00:01:16.890
+And it's called TRAMP.
+
+00:01:17.111 --> 00:01:20.712
+TRAMP stands for Transparent Remote File Access
+
+00:01:20.752 --> 00:01:21.512
+Multiple Protocol.
+
+00:01:22.421 --> 00:01:23.141
+And it's fantastic.
+
+00:01:23.201 --> 00:01:25.042
+It's one of the killer features of Emacs.
+
+NOTE Accessing GitHub
+
+00:01:26.282 --> 00:01:32.084
+I've been sitting thinking, well, can I use
+
+00:01:32.324 --> 00:01:35.405
+Tramp to access GitHub stuff?
+
+00:01:35.865 --> 00:01:39.587
+So if somebody gives me a URL, I have a URL in my
+
+00:01:39.627 --> 00:01:40.267
+browser open.
+
+00:01:40.367 --> 00:01:42.327
+Let me just bring it here.
+
+00:01:42.447 --> 00:01:44.028
+I have a handy command for that.
+
+00:01:44.468 --> 00:01:48.270
+So it inserted it in Org mode format because I'm
+
+00:01:48.330 --> 00:01:49.491
+in an Org mode buffer.
+
+00:01:50.011 --> 00:01:53.253
+Let's convert this into a plain URL.
+
+00:01:53.373 --> 00:01:58.016
+So if somebody gives me the URL like this, and I
+
+00:01:58.076 --> 00:02:02.699
+want just to see what's on in this repo, and I
+
+00:02:02.759 --> 00:02:07.161
+can do `C-x C-f`, paste that URL,
+
+00:02:07.622 --> 00:02:09.523
+press RET, and browse it.
+
+00:02:10.933 --> 00:02:13.474
+Turns out that is possible and that is not that
+
+00:02:13.514 --> 00:02:14.695
+difficult to achieve.
+
+00:02:15.495 --> 00:02:19.016
+And I started writing a package and I called it
+
+00:02:19.257 --> 00:02:19.897
+el-remoto.
+
+00:02:20.417 --> 00:02:24.019
+el-remoto is a pun on the EL extension.
+
+00:02:24.299 --> 00:02:25.799
+So it's basically self-recursive.
+
+00:02:25.939 --> 00:02:31.262
+It's el-remoto, remoto.el. And that's what it does.
+
+00:02:31.582 --> 00:02:36.384
+But we typically don't operate with URLs,
+
+00:02:36.404 --> 00:02:37.824
+specifically like full.[??]
+
+00:02:40.890 --> 00:02:42.531
+Let's say you want to look up something in, let's
+
+00:02:43.052 --> 00:02:44.513
+say Linux kernel.
+
+00:02:57.423 --> 00:02:59.123
+And Linux kernel is,
+
+00:02:59.124 --> 00:03:01.004
+it's kind of big, I think, right?
+
+00:03:01.064 --> 00:03:02.505
+Let's check how big it is.
+
+00:03:02.705 --> 00:03:06.646
+So I'm going to send https, I'm going to send a
+
+00:03:06.686 --> 00:03:10.627
+request to api.github.com.
+
+00:03:11.607 --> 00:03:16.688
+I think it's repos and then, well, I think,
+
+00:03:16.689 --> 00:03:19.530
+is it Torvalds? Linux, right?
+
+00:03:19.570 --> 00:03:23.451
+So if I sent this, if I sent this request...
+
+00:03:27.432 --> 00:03:28.974
+Okay, it gave me some data.
+
+00:03:30.275 --> 00:03:32.437
+There should be some numbers, I think.
+
+00:03:33.138 --> 00:03:33.799
+Was it size?
+
+00:03:34.059 --> 00:03:36.221
+Okay, let's just focus on size.
+
+00:03:36.241 --> 00:03:41.226
+I'm going to pipe it into jq with size.
+
+00:03:41.827 --> 00:03:44.850
+So this, I believe this is something like six
+
+00:03:44.890 --> 00:03:46.291
+gigabytes, not 60.
+
+00:03:46.731 --> 00:03:49.794
+And just to make sure, let's pipe it to, how do I?
+
+00:03:50.395 --> 00:03:53.337
+I think there is numfmt, but if I just do it like
+
+00:03:53.357 --> 00:03:54.918
+this, this is not, nothing going to happen.
+
+00:03:54.938 --> 00:03:58.400
+I think I have to convert it to, well, I don't
+
+00:03:58.460 --> 00:03:59.800
+remember exact parameters.
+
+00:03:59.840 --> 00:04:02.602
+Let's look them up.
+
+00:04:03.182 --> 00:04:08.725
+Why things are..., Oh yeah, numfmt,
+
+00:04:08.786 --> 00:04:19.672
+to maybe this one, let's try that.
+
+00:04:20.753 --> 00:04:21.553
+Let's write that.
+
+00:04:22.874 --> 00:04:23.815
+So I do this.
+
+00:04:24.915 --> 00:04:25.515
+And see?
+
+00:04:26.176 --> 00:04:27.036
+No, this is wrong.
+
+00:04:27.897 --> 00:04:29.017
+It can be six megabytes.
+
+00:04:29.918 --> 00:04:31.478
+I think we need
+
+00:04:32.179 --> 00:04:36.861
+from 1024, something like this?
+
+00:04:49.010 --> 00:04:50.730
+Oh, I need from unit, not from.
+
+00:04:53.671 --> 00:04:56.411
+So yeah, it's roughly six gigs, like I said.
+
+00:04:56.791 --> 00:04:59.452
+So you don't want to clone the entire six
+
+00:04:59.472 --> 00:05:01.432
+gigabytes just to look up something, right?
+
+00:05:02.773 --> 00:05:04.913
+And I wanted the browsing experience.
+
+NOTE Opening Github
+
+00:05:05.353 --> 00:05:08.474
+So if I press C-x C-f, and then type
+
+00:05:08.674 --> 00:05:15.015
+/github:, and you can see it shows my orgs.
+
+00:05:15.853 --> 00:05:17.855
+That's available because I'm using my own token.
+
+00:05:18.655 --> 00:05:21.877
+And if I type "torvalds",
+
+00:05:21.878 --> 00:05:25.320
+you see it's matching on the names.
+
+00:05:25.921 --> 00:05:29.042
+And at any point, if I press RET right here,
+
+00:05:29.043 --> 00:05:29.744
+it will open.
+
+00:05:30.064 --> 00:05:32.105
+Well, it matched on Torvalds,
+
+00:05:32.106 --> 00:05:33.807
+but let's go back again.
+
+00:05:34.548 --> 00:05:39.211
+Let me just explain what's happening.
+
+00:05:39.231 --> 00:05:40.812
+So if I type "tor",
+
+00:05:42.714 --> 00:05:47.536
+It will open all the matching authors and
+
+00:05:47.556 --> 00:05:50.697
+maintainers and orgs, but we need specifically
+
+00:05:50.777 --> 00:05:51.357
+"torvalds".
+
+00:05:52.478 --> 00:05:54.498
+And if I press enter here, and you can see this
+
+00:05:54.619 --> 00:05:59.180
+is normal Dired experience.
+
+00:05:59.341 --> 00:06:02.562
+It doesn't feel like I'm browsing something remote.
+
+00:06:02.722 --> 00:06:04.382
+It doesn't feel like
+
+00:06:04.383 --> 00:06:06.803
+it's sending requests somewhere else,
+
+00:06:06.804 --> 00:06:09.465
+but it's indeed remote repositories.
+
+00:06:10.125 --> 00:06:11.485
+If you want to look something in a file,
+
+00:06:11.486 --> 00:06:12.807
+you don't have to clone it.
+
+00:06:13.247 --> 00:06:14.328
+You can copy this file locally.
+
+00:06:14.348 --> 00:06:16.169
+There are other things you can do.
+
+00:06:16.210 --> 00:06:17.410
+For example, you can copy the URL of it,
+
+00:06:35.243 --> 00:06:37.024
+remoto-url.
+
+00:06:37.044 --> 00:06:39.645
+You can copy all sorts of different URLs.
+
+00:06:39.685 --> 00:06:43.708
+You can copy its raw URL, blame URL, etc.
+
+NOTE Embark
+
+00:06:44.008 --> 00:06:48.650
+But here, the point of Embark integration comes.
+
+00:06:48.850 --> 00:06:52.532
+Embark is a fantastic package and I have
+
+00:06:52.612 --> 00:06:53.893
+integrated into Remoto.
+
+00:06:53.973 --> 00:06:59.876
+So if I press my embark-act key, which is bound
+
+00:06:59.916 --> 00:07:02.898
+to `C-;`,
+
+00:07:03.859 --> 00:07:07.973
+It knows that I am at the remoto file,
+
+00:07:08.823 --> 00:07:13.406
+and it gives me some actions to copy the raw URL,
+
+00:07:13.446 --> 00:07:15.547
+copy the URL, blame history URL.
+
+00:07:15.587 --> 00:07:21.771
+So if I press H, I can send this to anybody,
+
+00:07:21.891 --> 00:07:23.332
+like somebody on my team, right?
+
+00:07:23.992 --> 00:07:28.755
+And if I am at the higher level, can I go higher?
+
+00:07:28.955 --> 00:07:30.316
+Oh, not from here, I guess.
+
+00:07:30.856 --> 00:07:36.760
+But if I go, if I'm here
+
+00:07:37.420 --> 00:07:41.442
+at this level, and if I press embark-act key again,
+
+00:07:41.482 --> 00:07:44.583
+you can see at this point it is... It knows that
+
+00:07:44.663 --> 00:07:48.065
+it's acting on... The target is a repo, not a file,
+
+00:07:48.085 --> 00:07:52.066
+so you can clone the repo. You can get
+
+00:07:52.286 --> 00:07:56.028
+specific different kind of urls.
+
+00:07:56.068 --> 00:07:57.649
+This also works while you...
+
+00:07:58.449 --> 00:08:00.350
+If you type add here,
+
+00:08:21.477 --> 00:08:24.857
+it will show the tags and branches
+
+00:08:24.858 --> 00:08:25.959
+in specific repo.
+
+00:08:26.719 --> 00:08:33.900
+If you type `#`, it will show issues and
+
+00:08:34.642 --> 00:08:35.241
+pull requests.
+
+00:08:35.302 --> 00:08:39.342
+Well, Linux doesn't seem to have any pull requests.
+
+00:08:39.503 --> 00:08:41.202
+Well, what about uemacs,
+
+00:08:41.203 --> 00:08:44.665
+which is a fork of Emacs Torvalds uses?
+
+00:08:45.045 --> 00:08:47.986
+Yeah, this one has pull requests and some open issues.
+
+00:08:48.326 --> 00:08:51.806
+You can even glance at them,
+
+00:08:51.807 --> 00:08:55.407
+see what this pull request is about.
+
+00:08:55.408 --> 00:08:59.190
+This is great. It also works for vim or
+
+00:08:59.390 --> 00:09:02.549
+evil users, so you can do like stuff like this.
+
+00:09:02.550 --> 00:09:08.111
+You can totally type `e` or `v` new and etc etc
+
+00:09:08.112 --> 00:09:10.413
+and it just works as well.
+
+00:09:11.493 --> 00:09:14.874
+You also don't have to type the whole GitHub thing.
+
+00:09:15.394 --> 00:09:20.836
+I have shorthand, so `gh`, and voila, we can go to
+
+00:09:20.976 --> 00:09:23.457
+and explore other repositories.
+
+NOTE Codeberg
+
+00:09:23.837 --> 00:09:27.819
+I just recently added CodeBerg support, so you
+
+00:09:27.839 --> 00:09:28.819
+can do CodeBerg.
+
+00:09:29.499 --> 00:09:34.361
+And I know Phil Technomancy.
+
+00:09:34.401 --> 00:09:36.822
+Yeah, Phil has CodeBerg.
+
+00:09:40.043 --> 00:09:42.364
+I have plans to add more support for more things.
+
+NOTE Issues
+
+00:09:42.484 --> 00:09:44.644
+Let's see what things I have remotely.
+
+00:09:44.764 --> 00:09:47.025
+If I press `#`, you can see the open issues.
+
+NOTE Plans
+
+00:09:47.085 --> 00:10:10.775
+I have plans to add support for S3, for browsing NPM,
+
+00:10:11.556 --> 00:10:12.456
+for SourceHut.
+
+00:10:12.496 --> 00:10:14.538
+GitLab probably comes next.
+
+00:10:14.798 --> 00:10:19.702
+I personally want to use the Gists or Gists.
+
+00:10:19.942 --> 00:10:23.966
+I want to be able to browse them.
+
+00:10:24.066 --> 00:10:26.808
+These are the planned features.
+
+00:10:29.733 --> 00:10:33.275
+And if you have any ideas or if you have issues,
+
+00:10:33.735 --> 00:10:34.415
+give it a try.
+
+00:10:34.555 --> 00:10:37.397
+It turns out that this package is very practical
+
+00:10:37.677 --> 00:10:41.619
+and it's actually really nice to be able to do
+
+00:10:42.180 --> 00:10:43.861
+things like that in this way.
+
+00:10:45.442 --> 00:10:45.922
+Check it out.
+
+00:10:46.842 --> 00:10:48.263
+See if it works for you.
+
+00:10:48.543 --> 00:10:50.424
+Let me know if you have any issues.
+
+00:10:50.504 --> 00:10:51.385
+Let's try to...
+
+00:10:51.745 --> 00:10:53.306
+If you have any ideas, let me know.
+
+00:10:53.746 --> 00:10:54.347
+Don't be shy.
+
+00:10:55.167 --> 00:10:56.468
+File a GitHub issue.
+
+00:10:57.549 --> 00:11:02.192
+And let's try to make [browsing remotes]
+
+00:11:02.332 --> 00:11:03.332
+experience in Emacs...
+
+00:11:03.792 --> 00:11:05.673
+Let's try to make it real.
+
+00:11:05.713 --> 00:11:06.974
+Let's try to make it awesome.
+
+00:11:07.755 --> 00:11:09.535
+That's all I have, I think.
+
+00:11:09.596 --> 00:11:10.556
+I still have some time.
+
+00:11:10.916 --> 00:11:15.219
+Maybe I can do Q&A.
+
+00:11:16.959 --> 00:11:19.181
+Some questions I had to answer.
+
+00:11:19.621 --> 00:11:20.761
+Some questions came.
+
+NOTE Why not a TRAMP backend?
+
+00:11:21.382 --> 00:11:24.083
+People asked why this is not a TRAMP backend.
+
+00:11:24.543 --> 00:11:25.324
+Well, TRAMP,
+
+00:11:26.178 --> 00:11:30.020
+turns out its back-end assumes a remote shell on
+
+00:11:30.040 --> 00:11:31.101
+the other end.
+
+00:11:31.281 --> 00:11:33.401
+So for the purposes of Remoto,
+
+00:11:33.402 --> 00:11:34.563
+there is no remote shell.
+
+00:11:34.723 --> 00:11:37.884
+So I just hijacked a file name handler, a list,
+
+00:11:38.565 --> 00:11:40.685
+and this is basically the mechanism
+
+00:11:40.686 --> 00:11:44.288
+the TRAMP mode uses on its own.
+
+NOTE Why not plain HTTP?
+
+00:11:45.008 --> 00:11:49.411
+And another question I had, why are you not using
+
+00:11:49.451 --> 00:11:50.471
+just plain HTTP?
+
+00:11:50.831 --> 00:11:52.692
+And I answered that as well.
+
+00:11:53.633 --> 00:11:57.355
+There's no raw HTTP URL that gives you the
+
+00:11:57.435 --> 00:11:59.596
+directory listing.
+
+00:11:59.796 --> 00:12:01.977
+It's just going to be more difficult.
+
+00:12:02.538 --> 00:12:05.319
+So yeah, there you have it.
+
+00:12:05.599 --> 00:12:06.380
+Here's the package.
+
+00:12:06.660 --> 00:12:07.320
+Give it a try.
+
+00:12:07.660 --> 00:12:08.461
+Let me know how it goes.
+
+00:12:09.301 --> 00:12:09.682
+That's all.
+
+00:12:10.082 --> 00:12:10.382
+Thank you.