ci: run workflows on GitHub-hosted runners

.github/workflows/ci-mobile-native.yml

@@ -1,9 +1,9 @@
 name: CI Mobile Native
 
 # Split out of ci.yml so it can be path-filtered. The job needs a macOS runner
-# for swiftlint, and macOS is the most expensive tier we buy, but the check it
+# for swiftlint, and macOS runners are the slowest to provision, but the check it
 # runs only ever reads .swift/.kt/.kts under apps/mobile -- roughly four seconds
-# of actual linting behind about forty seconds of runner setup. Filtering on
+# of actual linting behind a runner boot many times longer. Filtering on
 # apps/mobile/** would still be wrong: only 3 of the last 40 merged PRs touched
 # native sources, while 18 touched something under apps/mobile, so the broad
 # filter would boot macOS six times more often than the check can do any work.
@@ -43,7 +43,7 @@ permissions:
 jobs:
   mobile_native_static_analysis:
     name: Mobile Native Static Analysis
-    runs-on: blacksmith-6vcpu-macos-26
+    runs-on: macos-26
     timeout-minutes: 10
     steps:
       - name: Checkout

.github/workflows/ci.yml

@@ -13,7 +13,7 @@ concurrency:
 jobs:
   check:
     name: Check
-    runs-on: blacksmith-8vcpu-ubuntu-2404
+    runs-on: ubuntu-24.04
     timeout-minutes: 10
     steps:
       - name: Checkout
@@ -67,7 +67,7 @@ jobs:
   # limit stays at the default 4 so peak load per runner is unchanged.
   test:
     name: Test
-    runs-on: blacksmith-8vcpu-ubuntu-2404
+    runs-on: ubuntu-24.04
     timeout-minutes: 10
     steps:
       - name: Checkout
@@ -97,7 +97,7 @@ jobs:
   # isolation that flag buys is preserved exactly.
   test_server:
     name: Test Server ${{ matrix.shard }}
-    runs-on: blacksmith-8vcpu-ubuntu-2404
+    runs-on: ubuntu-24.04
     timeout-minutes: 10
     strategy:
       fail-fast: false
@@ -174,7 +174,7 @@ jobs:
   # for checks that take under 3s, on the critical path of every PR.
   rust:
     name: Rust
-    runs-on: blacksmith-4vcpu-ubuntu-2404
+    runs-on: ubuntu-24.04
     timeout-minutes: 10
     steps:
       - name: Checkout
@@ -198,7 +198,7 @@ jobs:
 
   release_smoke:
     name: Release Smoke
-    runs-on: blacksmith-8vcpu-ubuntu-2404
+    runs-on: ubuntu-24.04
     timeout-minutes: 10
     steps:
       - name: Checkout

.github/workflows/deploy-relay.yml

@@ -24,7 +24,7 @@ jobs:
     # credentials. Set the DEPLOY_RELAY repository variable to "true" once they
     # exist, so merges to pylon do not fail on a stack that cannot deploy.
     if: ${{ vars.DEPLOY_RELAY == 'true' }}
-    runs-on: blacksmith-8vcpu-ubuntu-2404
+    runs-on: ubuntu-24.04
     timeout-minutes: 15
     environment:
       name: production

.github/workflows/desktop-macos-preview.yml

@@ -19,7 +19,7 @@ jobs:
       github.event.pull_request.head.repo.full_name == github.repository &&
       contains(github.event.pull_request.labels.*.name, 'preview:mac') &&
       (github.event.action != 'labeled' || github.event.label.name == 'preview:mac')
-    runs-on: blacksmith-12vcpu-macos-26
+    runs-on: macos-26
     timeout-minutes: 30
     steps:
       - name: Checkout

.github/workflows/mobile-eas-preview.yml

@@ -10,7 +10,7 @@ jobs:
     if: |
       contains(github.event.pull_request.labels.*.name, '🚀 Mobile Continuous Deployment') &&
       (github.event.action != 'labeled' || github.event.label.name == '🚀 Mobile Continuous Deployment')
-    runs-on: blacksmith-8vcpu-ubuntu-2404
+    runs-on: ubuntu-24.04
     concurrency:
       group: mobile-eas-preview-${{ github.event.pull_request.number }}
       cancel-in-progress: true

.github/workflows/mobile-eas-production.yml

@@ -33,7 +33,7 @@ on:
 jobs:
   production:
     name: EAS Production ${{ inputs.mode }}
-    runs-on: blacksmith-8vcpu-ubuntu-2404
+    runs-on: ubuntu-24.04
     permissions:
       contents: read
     # Pylon Connect's public config lives in this environment, the same source

.github/workflows/mobile-showcase-screenshots.yml

@@ -45,7 +45,7 @@ jobs:
   ios:
     name: iPhone 6.9, iPhone 6.5, and iPad 13
     if: inputs.platform == 'all' || inputs.platform == 'ios'
-    runs-on: blacksmith-12vcpu-macos-26
+    runs-on: macos-26
     # Capturing every palette multiplies the device matrix by six, and only the
     # one native build is shared between them.
     timeout-minutes: ${{ inputs.theme == 'all' && 300 || 60 }}
@@ -94,7 +94,7 @@ jobs:
   android:
     name: Android phone, 7-inch tablet, and 10-inch tablet
     if: inputs.platform == 'all' || inputs.platform == 'android'
-    runs-on: blacksmith-16vcpu-ubuntu-2404
+    runs-on: ubuntu-24.04
     # Capturing every palette multiplies the device matrix by six, and only the
     # one native build is shared between them.
     timeout-minutes: ${{ inputs.theme == 'all' && 300 || 60 }}
@@ -151,7 +151,7 @@ jobs:
           arch: x86_64
           profile: pixel_7_pro
           avd-name: Pixel_10_Pro
-          cores: 8
+          cores: 4
           ram-size: 4096M
           disable-animations: false
           script: pnpm screenshots:mobile --platform android --appearance "${{ inputs.appearance }}" --theme "${{ inputs.theme }}"

.github/workflows/release.yml

@@ -42,12 +42,11 @@ permissions:
   id-token: none
 
 jobs:
-  # Runner split: the four platform builds and the preflight suite stay on
-  # Blacksmith, where the compute is worth paying for. Everything else here is
-  # short and I/O-bound — API calls, artifact uploads, a git push, a curl — and
-  # those jobs repeatedly sat unassigned in Blacksmith's queue while heavier
-  # jobs in the same run picked up immediately. GitHub-hosted runners took them
-  # in seconds every time, so light orchestration runs there.
+  # Every job runs on GitHub-hosted standard runners, which are free on this
+  # public repository. Heavy jobs pin an OS version (`ubuntu-24.04`,
+  # `windows-2025`, `macos-26`) so a build cannot drift when `-latest` moves;
+  # the short I/O-bound orchestration jobs use `ubuntu-latest` because nothing
+  # in them is OS-sensitive.
   check_changes:
     name: Check for changes since last nightly
     if: github.event_name == 'schedule'
@@ -91,7 +90,7 @@ jobs:
     if: |
       !failure() && !cancelled() &&
       (github.event_name != 'schedule' || needs.check_changes.outputs.has_changes == 'true')
-    runs-on: blacksmith-8vcpu-ubuntu-2404
+    runs-on: ubuntu-24.04
     # Runs the whole repo's check, typecheck, and test suites.
     timeout-minutes: 30
     outputs:
@@ -188,7 +187,7 @@ jobs:
     name: Release quality checks
     needs: [preflight]
     if: ${{ !failure() && !cancelled() && needs.preflight.result == 'success' }}
-    runs-on: blacksmith-8vcpu-ubuntu-2404
+    runs-on: ubuntu-24.04
     timeout-minutes: 10
     steps:
       - name: Checkout
@@ -348,7 +347,7 @@ jobs:
     name: Build WSL node-pty (linux-x64)
     needs: [preflight]
     if: ${{ !failure() && !cancelled() && needs.preflight.result == 'success' }}
-    runs-on: blacksmith-8vcpu-ubuntu-2404
+    runs-on: ubuntu-24.04
     timeout-minutes: 30
     steps:
       - name: Checkout
@@ -419,32 +418,28 @@ jobs:
       matrix:
         include:
           - label: macOS arm64
-            runner: blacksmith-6vcpu-macos-26
+            runner: macos-26
             platform: mac
             target: dmg
             arch: arm64
             rust_target: aarch64-apple-darwin
             resource_key: darwin-arm64
           - label: macOS x64
-            runner: blacksmith-6vcpu-macos-26
+            runner: macos-26
             platform: mac
             target: dmg
             arch: x64
             rust_target: x86_64-apple-darwin
             resource_key: darwin-x64
           - label: Linux x64
-            # GitHub-hosted deliberately. Every stuck job so far has been on a
-            # blacksmith ubuntu runner, including this build sitting unassigned
-            # for 43 minutes while the macOS and Windows jobs queued after it
-            # ran and finished. Slower here, but it actually starts.
-            runner: ubuntu-latest
+            runner: ubuntu-24.04
             platform: linux
             target: AppImage
             arch: x64
             rust_target: x86_64-unknown-linux-gnu
             resource_key: linux-x64
           - label: Windows x64
-            runner: blacksmith-8vcpu-windows-2025
+            runner: windows-2025
             platform: win
             target: nsis
             arch: x64
@@ -821,7 +816,7 @@ jobs:
     # trusted publishing configured for it. Set the PUBLISH_CLI_TO_NPM
     # repository variable to "true" once both exist.
     if: ${{ !failure() && !cancelled() && vars.PUBLISH_CLI_TO_NPM == 'true' && needs.preflight.result == 'success' && needs.relay_public_config.result == 'success' && needs.quality.result == 'success' && needs.build.result == 'success' }}
-    runs-on: ubuntu-24.04 # blacksmith-8vcpu-ubuntu-2404
+    runs-on: ubuntu-24.04
     timeout-minutes: 10
     permissions:
       contents: read

AGENTS.md

@@ -23,7 +23,7 @@ T3-named environment variables may still be passed to the bundled compatibility
 
 Pylon is a long-lived independent product, not a temporary reskin or a patch queue intended to collapse back into T3 Code. Build in Pylon's direction while preserving the upstream qualities described below.
 
-- The canonical repository is the private `pylon-code/pylon` repository. Its writable remote is `origin`, and its default product branch is `pylon`. It moved from `rynfar/pylon` to the `pylon-code` organization; GitHub redirects the old path, but use the organization path.
+- The canonical repository is the public `pylon-code/pylon` repository. Its writable remote is `origin`, and its default product branch is `pylon`. It moved from `rynfar/pylon` to the `pylon-code` organization; GitHub redirects the old path, but use the organization path.
 - Base Pylon work on `pylon` or a task branch created from it. Do not treat the inherited `main` branch as Pylon's product branch.
 - `t3code-upstream` (`pingdotgg/t3code`) and `t3code-fork` (`rynfar/t3code`) are reference remotes. They are intentionally fetch-only. Never push to them or re-enable their push URLs.
 - Upstream changes are opt-in. Fetch and inspect upstream commits, then cherry-pick or selectively merge only changes that benefit Pylon. Do not hard-reset, wholesale rebase, or replace Pylon with an upstream branch.

docs/operations/mobile-app-store-screenshots.md

@@ -104,9 +104,9 @@ appearance settings can never drift apart.
 Run the `Mobile Showcase Screenshots` workflow from GitHub's Actions tab, choose `all`, `ios`, or
 `android`, select `light`, `dark`, or `both`, and pick a palette (or `all`, which raises each job's
 timeout from 60 to 300 minutes). The default dispatch captures both appearances of the `t3-code`
-palette and runs iOS and Android concurrently: iPhone and iPad capture on a
-12-vCPU Blacksmith macOS runner, while Android phone, 7-inch tablet, and 10-inch tablet capture on a
-16-vCPU Blacksmith Linux runner with a KVM-accelerated x86_64 emulator.
+palette and runs iOS and Android concurrently: iPhone and iPad capture on a GitHub-hosted
+`macos-26` runner, while Android phone, 7-inch tablet, and 10-inch tablet capture on a GitHub-hosted
+`ubuntu-24.04` runner with a KVM-accelerated x86_64 emulator.
 
 Every job uploads its PNGs even when capture fails, which makes partial runs useful for diagnosis.
 The separate validation step is success-gated: it runs before upload only when capture succeeds. If

docs/operations/release.md

@@ -72,22 +72,21 @@ Release App installation.
 
 ## Runners
 
-Every job runs on Blacksmith runners, which requires the Blacksmith app installed on the
-`pylon-code` organization — Blacksmith does not support personal accounts.
+Every job runs on GitHub-hosted standard runners: `ubuntu-24.04`, `windows-2025`, and `macos-26`
+(Apple Silicon; the x64 desktop leg cross-builds on it). Standard runners are free on public
+repositories, which is the reason this repository is public. Larger tiers (`-large`, `-xlarge`) and
+third-party runner providers are billed regardless of visibility, so do not move a job onto one
+without accepting the cost. The workflows ran on Blacksmith until August 2026; switching providers
+is a pure `runs-on` label swap with no other change required.
 
-Tiers are deliberately conservative: `blacksmith-8vcpu-ubuntu-2404`, `blacksmith-8vcpu-windows-2025`,
-and `blacksmith-6vcpu-macos-26`. Blacksmith's documentation only evidences these sizes, and **an
-unavailable tier does not fail the job — it queues until GitHub's 24-hour limit kills it.** That
-silent hang is the single most confusing failure this workflow can produce, so raise a tier only
-after confirming the plan grants it, and watch the first run after any change.
+**An unknown or unavailable label does not fail the job — it queues until GitHub's 24-hour limit
+kills it.** That silent hang is the single most confusing failure this workflow can produce, so
+watch the first run after any label change.
 
 Job timeouts are generous for the same reason a ceiling is cheap: preflight runs the whole
 repository's check, typecheck, and test suites, and the build matrix packages Electron plus a Rust
 target with the macOS legs cross-building a second architecture.
 
-Switching back to GitHub-hosted runners is a pure label swap — `ubuntu-24.04`, `windows-2025`,
-`macos-26` — with no other change required.
-
 ## Pylon Connect relay deployment
 
 The relay is a shared control plane versioned separately from client releases. Stable and nightly

scripts/mobile-showcase.config.ts

@@ -158,7 +158,7 @@ const config: ShowcaseConfig = {
       platform: "android",
       avd: "Pixel_10_Pro",
       // Apple Silicon uses ARM64 locally; CI overrides this with x86_64 so its
-      // Blacksmith Linux runner can use KVM acceleration.
+      // Linux runner can use KVM acceleration.
       abi: resolveShowcaseAndroidAbi(process.env.T3_SHOWCASE_ANDROID_ABI),
       appearance: "dark",
       theme: DEFAULT_SHOWCASE_THEME,