From 9f03730a9ffa2ffca381dfbeb32ad26732c7d93f Mon Sep 17 00:00:00 2001
From: bradleyshep <148254416+bradleyshep@users.noreply.github.com>
Date: Tue, 2 Jun 2026 13:30:58 -0400
Subject: [PATCH 1/2] updates

---
 .github/workflows/llm-benchmark-periodic.yml         | 6 +-----
 .github/workflows/llm-benchmark-validate-goldens.yml | 6 +-----
 2 files changed, 2 insertions(+), 10 deletions(-)

diff --git a/.github/workflows/llm-benchmark-periodic.yml b/.github/workflows/llm-benchmark-periodic.yml
index f8b4c1fb179..24c19b0767f 100644
--- a/.github/workflows/llm-benchmark-periodic.yml
+++ b/.github/workflows/llm-benchmark-periodic.yml
@@ -29,11 +29,7 @@ concurrency:
 
 jobs:
   run-benchmarks:
-    runs-on: spacetimedb-new-runner
-    container:
-      image: localhost:5000/spacetimedb-ci:latest
-      options: >-
-        --privileged
+    runs-on: spacetimedb-new-runner-2
     timeout-minutes: 180
 
     steps:
diff --git a/.github/workflows/llm-benchmark-validate-goldens.yml b/.github/workflows/llm-benchmark-validate-goldens.yml
index f9d634c50ea..8f4a17c1484 100644
--- a/.github/workflows/llm-benchmark-validate-goldens.yml
+++ b/.github/workflows/llm-benchmark-validate-goldens.yml
@@ -15,11 +15,7 @@ concurrency:
 
 jobs:
   validate-goldens:
-    runs-on: spacetimedb-new-runner
-    container:
-      image: localhost:5000/spacetimedb-ci:latest
-      options: >-
-        --privileged
+    runs-on: spacetimedb-new-runner-2
     timeout-minutes: 60
 
     strategy:

From 157c2c90e84fa0411c5a3d157e47706e8305579e Mon Sep 17 00:00:00 2001
From: bradleyshep <148254416+bradleyshep@users.noreply.github.com>
Date: Tue, 2 Jun 2026 13:59:02 -0400
Subject: [PATCH 2/2] build local

---
 .github/workflows/llm-benchmark-periodic.yml         | 12 +++++++-----
 .github/workflows/llm-benchmark-validate-goldens.yml | 12 +++++++-----
 2 files changed, 14 insertions(+), 10 deletions(-)

diff --git a/.github/workflows/llm-benchmark-periodic.yml b/.github/workflows/llm-benchmark-periodic.yml
index 24c19b0767f..40ad2c75fe4 100644
--- a/.github/workflows/llm-benchmark-periodic.yml
+++ b/.github/workflows/llm-benchmark-periodic.yml
@@ -33,11 +33,6 @@ jobs:
     timeout-minutes: 180
 
     steps:
-      - name: Install spacetime CLI
-        run: |
-          curl -sSf https://install.spacetimedb.com | sh -s -- -y
-          echo "$HOME/.local/bin" >> $GITHUB_PATH
-
       - name: Checkout master
         uses: actions/checkout@v4
         with:
@@ -71,6 +66,13 @@ jobs:
       - name: Build llm-benchmark tool
         run: cargo install --path tools/xtask-llm-benchmark --locked
 
+      - name: Build SpacetimeDB server for benchmark harness
+        run: |
+          cargo ci smoketests prepare
+          mkdir -p "$HOME/.local/bin"
+          ln -sf "$GITHUB_WORKSPACE/target/release/spacetimedb-cli" "$HOME/.local/bin/spacetime"
+          echo "$HOME/.local/bin" >> "$GITHUB_PATH"
+
       - name: Run benchmarks
         env:
           OPENROUTER_API_KEY: ${{ secrets.OPENROUTER_API_KEY }}
diff --git a/.github/workflows/llm-benchmark-validate-goldens.yml b/.github/workflows/llm-benchmark-validate-goldens.yml
index 8f4a17c1484..17384a654e3 100644
--- a/.github/workflows/llm-benchmark-validate-goldens.yml
+++ b/.github/workflows/llm-benchmark-validate-goldens.yml
@@ -24,11 +24,6 @@ jobs:
         lang: [rust, csharp, typescript]
 
     steps:
-      - name: Install spacetime CLI
-        run: |
-          curl -sSf https://install.spacetimedb.com | sh -s -- -y
-          echo "$HOME/.local/bin" >> $GITHUB_PATH
-
       - name: Checkout master
         uses: actions/checkout@v4
         with:
@@ -66,6 +61,13 @@ jobs:
       - name: Build llm-benchmark tool
         run: cargo install --path tools/xtask-llm-benchmark --locked
 
+      - name: Build SpacetimeDB server for benchmark harness
+        run: |
+          cargo ci smoketests prepare
+          mkdir -p "$HOME/.local/bin"
+          ln -sf "$GITHUB_WORKSPACE/target/release/spacetimedb-cli" "$HOME/.local/bin/spacetime"
+          echo "$HOME/.local/bin" >> "$GITHUB_PATH"
+
       - name: Validate golden answers (${{ matrix.lang }})
         env:
           MSBUILDDISABLENODEREUSE: "1"