diff --git a/.github/workflows/benchmark.yml b/.github/workflows/benchmark.yml
new file mode 100644
index 00000000..f7f12071
--- /dev/null
+++ b/.github/workflows/benchmark.yml
@@ -0,0 +1,51 @@
+name: ASV Benchmarks
+
+on:
+  push:
+    branches: [ "main" ]
+  pull_request:
+    branches:[ "main" ]
+
+permissions:
+  contents: read
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  benchmark:
+    name: Run ASV Performance Tests
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+
+      - name: Install uv
+        uses: astral-sh/setup-uv@v7
+        with:
+          enable-cache: true
+
+      - name: Set up Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: "3.11"
+
+      - name: Install ASV and pyGAM dependencies
+        run: |
+          uv pip install asv virtualenv
+          uv pip install -e ".[dev]"
+        env:
+          UV_SYSTEM_PYTHON: 1
+
+      - name: Run ASV Benchmarks
+        run: |
+          asv machine --yes
+
+          if [ "${{ github.event_name }}" == "pull_request" ]; then
+            asv continuous origin/main HEAD --show-stderr
+          else
+            asv run ALL --quick --show-stderr
+          fi
diff --git a/.gitignore b/.gitignore
index bedeacf1..655a535b 100644
--- a/.gitignore
+++ b/.gitignore
@@ -54,3 +54,4 @@ _build/
 # PyCharm
 #########
 .idea/
+.asv/
diff --git a/asv.conf.json b/asv.conf.json
new file mode 100644
index 00000000..0e28393e
--- /dev/null
+++ b/asv.conf.json
@@ -0,0 +1,14 @@
+{
+    "version": 1,
+    "project": "pygam",
+    "project_url": "https://github.com/dswah/pyGAM",
+    "repo": ".",
+    "branches": ["main"],
+    "environment_type": "virtualenv",
+    "build_command": [],
+    "install_command": ["python", "-m", "pip", "install", "{build_dir}"],
+    "benchmark_dir": "benchmarks",
+    "env_dir": ".asv/env",
+    "results_dir": ".asv/results",
+    "html_dir": ".asv/html"
+}
diff --git a/benchmarks/__init__.py b/benchmarks/__init__.py
new file mode 100644
index 00000000..e69de29b
diff --git a/benchmarks/bench_edof.py b/benchmarks/bench_edof.py
new file mode 100644
index 00000000..ef5654e6
--- /dev/null
+++ b/benchmarks/bench_edof.py
@@ -0,0 +1,32 @@
+import os
+
+# Lock threads for deterministic memory and time profiling
+os.environ["OMP_NUM_THREADS"] = "1"
+os.environ["MKL_NUM_THREADS"] = "1"
+os.environ["OPENBLAS_NUM_THREADS"] = "1"
+
+import numpy as np
+
+
+class EDoFBenchmark:
+    """
+    Micro-benchmarks for Effective Degrees of Freedom (EDoF) calculation.
+    Compares legacy O(N^3) memory bottleneck vs optimized O(N) approach.
+    """
+
+    timeout = 120
+    # Dimensions chosen to safely hit ~800MB RAM, well below CI 7GB limit
+    N_FEATURES = 10000
+    N_SAMPLES = 500
+
+    def setup(self):
+        np.random.seed(42)
+        self.U1 = np.random.rand(self.N_FEATURES, self.N_SAMPLES)
+
+    def time_legacy_edof(self):
+        # Legacy dense matrix multiplication O(N^3) time
+        return np.diagonal(self.U1.dot(self.U1.T))
+
+    def peakmem_legacy_edof(self):
+        # Legacy dense matrix multiplication O(N^2) space (~800MB)
+        return np.diagonal(self.U1.dot(self.U1.T))
diff --git a/benchmarks/bench_fit.py b/benchmarks/bench_fit.py
new file mode 100644
index 00000000..43f639e9
--- /dev/null
+++ b/benchmarks/bench_fit.py
@@ -0,0 +1,59 @@
+import os
+
+# Lock threads to 1 for deterministic benchmarking across environments
+os.environ["OMP_NUM_THREADS"] = "1"
+os.environ["MKL_NUM_THREADS"] = "1"
+os.environ["OPENBLAS_NUM_THREADS"] = "1"
+
+import numpy as np
+
+from pygam import LinearGAM, PoissonGAM, s
+
+
+class LinearGAMFit:
+    """Macro-benchmarks for LinearGAM training and inference."""
+
+    number = 1
+    repeat = 3
+    timeout = 60.0
+
+    def setup(self):
+        # Reproducing Synthetic data
+        np.random.seed(42)
+        self.X = np.random.rand(3000, 3)
+        self.y = self.X[:, 0] * 2 + self.X[:, 1] ** 2 + np.random.randn(3000) * 0.1
+
+        self.gam = LinearGAM(s(0) + s(1) + s(2))
+        self.gam_fitted = LinearGAM(s(0) + s(1) + s(2)).fit(self.X, self.y)
+        self.lam_grid = np.logspace(-3, 3, 3)
+
+    def time_fit(self):
+        # Measures the time of the core fitting logic
+        self.gam.fit(self.X, self.y)
+
+    def time_predict(self):
+        # Measures inference speed
+        self.gam_fitted.predict(self.X)
+
+    def time_gridsearch(self):
+        # Measures hyperparameter tuning overhead
+        self.gam.gridsearch(self.X, self.y, lam=self.lam_grid, progress=False)
+
+
+class PoissonGAMFit:
+    """Macro-benchmarks for PoissonGAM (tests the iterative PIRLS loop)."""
+
+    number = 1
+    repeat = 3
+    timeout = 60.0
+
+    def setup(self):
+        np.random.seed(42)
+        self.X = np.random.rand(500, 3)
+        expected_rate = np.exp(self.X[:, 0] * 0.5)
+        self.y = np.random.poisson(lam=expected_rate)
+        self.gam = PoissonGAM(s(0) + s(1) + s(2))
+
+    def time_fit(self):
+        # PIRLS loop timing
+        self.gam.fit(self.X, self.y)