diff --git a/.firebaserc b/.firebaserc
index 304a4ac8a7..4c28ddc881 100644
--- a/.firebaserc
+++ b/.firebaserc
@@ -7,9 +7,12 @@
       "hosting": {
         "app": [
           "neuroglancer"
+        ],
+        "docs": [
+          "neuroglancer-docs"
         ]
       }
     }
   },
   "etags": {}
-}
\ No newline at end of file
+}
diff --git a/.github/workflows/build.yml b/.github/workflows/build.yml
index c33f5d19eb..622e215a3e 100644
--- a/.github/workflows/build.yml
+++ b/.github/workflows/build.yml
@@ -97,7 +97,7 @@ jobs:
       # - name: Setup tmate session
       #   uses: mxschmitt/action-tmate@v3
       - name: Install Python packaging/test tools
-        run: python -m pip install --upgrade pip tox nox wheel numpy -r python/requirements-test.txt
+        run: pip install --upgrade pip tox nox wheel numpy -r python/requirements-test.txt
       - uses: ./.github/actions/setup-firefox
       - run: nox -s lint format mypy
       - name: Check for dirty working directory
@@ -174,6 +174,29 @@ jobs:
             dist/*.whl
             dist/*.tar.gz
 
+  docs:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - name: Set up Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: 3.12
+      - name: Setup Graphviz
+        uses: ts-graphviz/setup-graphviz@b1de5da23ed0a6d14e0aeee8ed52fdd87af2363c # v2.0.2
+        with:
+          macos-skip-brew-update: "true"
+      - name: Install nox
+        run: pip install nox
+      - name: Build docs
+        run: nox -s docs
+      - name: Upload docs as artifact
+        uses: actions/upload-artifact@v4
+        with:
+          name: docs
+          path: |
+            dist/docs
+
   publish-package:
     # Only publish package on push to tag or default branch.
     if: ${{ github.event_name == 'push' && (startsWith(github.ref, 'refs/tags/v') || github.ref == 'refs/heads/master') }}
@@ -181,6 +204,7 @@ jobs:
     needs:
       - "client"
       - "python-build-package"
+      - "docs"
     steps:
       - uses: actions/checkout@v4
       - name: Use Node.js
@@ -220,12 +244,26 @@ jobs:
         with:
           name: client
           path: dist/client
-      - name: Publish to Firebase hosting
+      - name: Publish client to Firebase hosting
+        uses: FirebaseExtended/action-hosting-deploy@v0
+        with:
+          firebaseServiceAccount: "${{ secrets.FIREBASE_HOSTING_SERVICE_ACCOUNT_KEY }}"
+          projectId: neuroglancer-demo
+          channelId: live
+          target: app
+      # Download dist/docs after publishing to PyPI, because PyPI publish
+      # action fails if dist/docs directory is present.
+      - uses: actions/download-artifact@v4
+        with:
+          name: docs
+          path: dist/docs
+      - name: Publish docs to Firebase hosting
         uses: FirebaseExtended/action-hosting-deploy@v0
         with:
           firebaseServiceAccount: "${{ secrets.FIREBASE_HOSTING_SERVICE_ACCOUNT_KEY }}"
           projectId: neuroglancer-demo
           channelId: live
+          target: docs
 
   ngauth:
     strategy:
diff --git a/.github/workflows/build_docs_preview.yml b/.github/workflows/build_docs_preview.yml
new file mode 100644
index 0000000000..77523f7f25
--- /dev/null
+++ b/.github/workflows/build_docs_preview.yml
@@ -0,0 +1,29 @@
+name: Build docs preview
+
+on:
+  pull_request:
+
+jobs:
+  upload:
+    runs-on: ubuntu-latest
+
+    steps:
+      - uses: actions/checkout@v4
+      - name: Set up Python ${{ matrix.python-version }}
+        uses: actions/setup-python@v5
+        with:
+          python-version: 3.12
+      - name: Setup Graphviz
+        uses: ts-graphviz/setup-graphviz@b1de5da23ed0a6d14e0aeee8ed52fdd87af2363c # v2.0.2
+        with:
+          macos-skip-brew-update: "true"
+      - name: Install nox
+        run: pip install nox
+      - name: Build documentation
+        run: nox -s docs
+      - name: Upload client as artifact
+        uses: actions/upload-artifact@v4
+        with:
+          name: docs
+          path: |
+            dist/docs/
diff --git a/.github/workflows/deploy_docs_preview.yml b/.github/workflows/deploy_docs_preview.yml
new file mode 100644
index 0000000000..68207e2007
--- /dev/null
+++ b/.github/workflows/deploy_docs_preview.yml
@@ -0,0 +1,67 @@
+name: Deploy docs preview
+
+on:
+  workflow_run:
+    workflows: ["Build docs preview"]
+    types: [completed]
+
+jobs:
+  deploy:
+    runs-on: ubuntu-latest
+    if: ${{ github.event.workflow_run.conclusion == 'success' }}
+    steps:
+      - name: "Create commit status"
+        uses: actions/github-script@v7
+        with:
+          github-token: ${{ secrets.GITHUB_TOKEN }}
+          script: |
+            const commitId = "${{ github.event.workflow_run.head_commit.id }}";
+            await github.rest.repos.createCommitStatus({
+                context: "docs-preview",
+                owner: context.repo.owner,
+                repo: context.repo.repo,
+                sha: commitId,
+                state: "pending",
+                description: `Creating preview`,
+                target_url: `${context.serverUrl}/${context.repo.owner}/${context.repo.repo}/actions/runs/${context.runId}`,
+            });
+      - uses: actions/checkout@v4
+      - uses: actions/download-artifact@v4
+        with:
+          name: docs
+          path: dist/docs
+          github-token: "${{ secrets.GITHUB_TOKEN }}"
+          run-id: "${{ github.event.workflow_run.id }}"
+      - name: Get PR ID
+        # https://github.com/orgs/community/discussions/25220#discussioncomment-7532132
+        id: pr-id
+        run: |
+          PR_ID=$(gh run view -R ${{ github.repository }} ${{ github.event.workflow_run.id }} | grep -oP '#[0-9]+ . ${{ github.event.workflow_run.id }}' | grep -oP '#[0-9]+' | cut -c 2-)
+          echo "pr-id=${PR_ID}" >> $GITHUB_OUTPUT
+        env:
+          GH_TOKEN: "${{ secrets.GITHUB_TOKEN }}"
+      - uses: FirebaseExtended/action-hosting-deploy@v0
+        id: deploy
+        with:
+          repoToken: "${{ secrets.GITHUB_TOKEN }}"
+          firebaseServiceAccount: "${{ secrets.FIREBASE_HOSTING_SERVICE_ACCOUNT_KEY }}"
+          expires: 30d
+          channelId: "pr${{ steps.pr-id.outputs.pr-id }}-docs"
+          projectId: neuroglancer-demo
+          target: docs
+      - name: "Update commit status"
+        uses: actions/github-script@v7
+        with:
+          github-token: ${{ secrets.GITHUB_TOKEN }}
+          script: |
+            const expires = new Date("${{ steps.deploy.outputs.expire_time }}");
+            const commitId = "${{ github.event.workflow_run.head_commit.id }}";
+            await github.rest.repos.createCommitStatus({
+                context: "docs-preview",
+                owner: context.repo.owner,
+                repo: context.repo.repo,
+                sha: commitId,
+                state: "success",
+                target_url: "${{ steps.deploy.outputs.details_url }}",
+                description: `Preview created, expires at: ${expires.toISOString()}`,
+            });
diff --git a/.github/workflows/deploy_preview.yml b/.github/workflows/deploy_preview.yml
index 545bda17c2..bdde334d02 100644
--- a/.github/workflows/deploy_preview.yml
+++ b/.github/workflows/deploy_preview.yml
@@ -48,6 +48,7 @@ jobs:
           expires: 30d
           channelId: "pr${{ steps.pr-id.outputs.pr-id }}"
           projectId: neuroglancer-demo
+          target: docs
       - name: "Update commit status"
         uses: actions/github-script@v7
         with:
diff --git a/docs/concepts/coordinate_spaces.dot b/docs/concepts/coordinate_spaces.dot
new file mode 100644
index 0000000000..f8a45fceb0
--- /dev/null
+++ b/docs/concepts/coordinate_spaces.dot
@@ -0,0 +1,218 @@
+digraph coordinate_spaces {
+  newrank = true;
+  outputorder = "edgesfirst";
+  compound = true;
+
+  node [
+    style="rounded"
+    shape=rect
+  ]
+
+  edge [
+  ]
+
+  graph [
+    rankdir = LR
+    fillcolor = "var(--md-default-fg-color--lightest)"
+    style="solid,filled"
+  ]
+  {
+    rank=same;
+    projection_view_space cross_section_view_space global_space
+  }
+
+  subgraph cluster_layer {
+    label = <Layer <i>i</i>>
+
+    subgraph cluster_datasource {
+      label = <Data source <i>j</i>>
+
+      { rank=same;  
+
+        datasource_space [
+          label="Coordinate space"
+            href="#data-source-coordinate-space"
+        ]
+
+        datasource_transform [
+        label = "Coordinate transform"
+          href="#data-source-coordinate-transform"
+        ]
+
+      }
+      style="filled,dashed"
+    }
+
+    {
+      rank = same;
+
+      subgraph cluster_layer_space {
+        label = "Coordinate space"
+
+        layer_global_dims [
+          label="Global dimensions"
+        ]
+        layer_local_dims [
+          label="Local dimensions"
+        ]
+        layer_channel_dims [
+           label="Channel dimensions"
+        ]
+      }
+
+      layer_position [
+        label = "Local position"
+      ]
+    }
+
+    datasource_space -> datasource_transform
+    datasource_transform -> layer_local_dims [
+      lhead="cluster_layer_space"
+    ]
+    style="filled,dashed"
+  }
+
+  global_space [
+    label="Global\ncoordinate\nspace"
+  ]
+
+  layer_global_dims -> global_space [
+    label="Merge"
+  ]
+
+  subgraph cluster_globalcamera {
+    label = "Global camera parameters"
+
+    subgraph cluster_globalcamera_cross_section {
+
+      label = "Cross-section"
+
+      global_cross_section_orientation [
+        label = "Orientation"
+      ]                           
+
+      global_cross_section_scale [
+        label = "Scale"
+      ]
+      global_cross_section_depth [
+        label = "Depth bounds"
+      ]
+    }
+
+    global_center_position [
+      label = "Center position"
+    ]
+
+    global_display_dimensions [
+      label = "Display dimensions"
+    ]
+
+    global_relative_display_scales [
+      label = "Relative\ndisplay scales"
+    ]
+
+    subgraph cluster_globalcamera_projection {
+
+      label = "3-d projection"
+
+      global_projection_orientation [
+        label = "Orientation"
+      ]                           
+
+      global_projection_scale [
+        label = "Scale"
+      ]
+
+      global_projection_depth [
+        label = "Depth bounds"
+      ]
+    }
+  }
+
+  subgraph cluster_layergroupviewer {
+    label = <Layer group viewer <i>k</i>>
+
+    layergroupviewer_center_position [
+      label = "Center position"
+    ]
+
+    layergroupviewer_display_dimensions [
+      label = "Display dimensions"
+    ]
+
+    layergroupviewer_relative_display_scales [
+      label = "Relative\ndisplay scales"
+    ]
+
+    subgraph cluster_layergroupviewer_cross_section {
+      label = "Cross-section views"
+
+      layergroupviewer_cross_section_orientation [
+        label = "Orientation"
+      ]                           
+
+      layergroupviewer_cross_section_scale [
+        label = "Scale"
+      ]
+      layergroupviewer_cross_section_depth [
+        label = "Depth bounds"
+      ]
+      cross_section_view_space [
+        label = "View\ncoordinate\nspace"
+      ]
+    }
+
+    subgraph cluster_layergroupviewer_projection {
+      label = "3-d projection view"
+
+      layergroupviewer_projection_orientation [
+        label = "Orientation"
+      ]                           
+
+      layergroupviewer_projection_scale [
+        label = "Scale"
+      ]
+
+      layergroupviewer_projection_depth [
+        label = "Depth bounds"
+      ]
+
+      projection_view_space [
+        label = "View\ncoordinate\nspace"
+      ]
+
+    }
+
+    { layergroupviewer_display_dimensions,
+      layergroupviewer_relative_display_scales,
+      layergroupviewer_center_position,
+      layergroupviewer_cross_section_depth,
+      layergroupviewer_cross_section_scale,
+      layergroupviewer_cross_section_orientation } -> cross_section_view_space
+
+    { layergroupviewer_display_dimensions,
+      layergroupviewer_relative_display_scales,
+      layergroupviewer_center_position,
+      layergroupviewer_projection_depth,
+      layergroupviewer_projection_scale,
+      layergroupviewer_projection_orientation } -> projection_view_space
+    style="filled,dashed"
+  }
+
+  {
+    edge [
+      style=dotted
+    ]
+    global_center_position -> layergroupviewer_center_position
+    global_cross_section_orientation -> layergroupviewer_cross_section_orientation
+    global_cross_section_scale -> layergroupviewer_cross_section_scale
+    global_cross_section_depth -> layergroupviewer_cross_section_depth
+    global_projection_orientation -> layergroupviewer_projection_orientation
+    global_projection_scale -> layergroupviewer_projection_scale
+    global_projection_depth -> layergroupviewer_projection_depth
+    global_display_dimensions -> layergroupviewer_display_dimensions
+    global_relative_display_scales -> layergroupviewer_relative_display_scales
+  }
+
+  global_space -> {cross_section_view_space,projection_view_space}
+}
diff --git a/docs/concepts/coordinate_spaces.rst b/docs/concepts/coordinate_spaces.rst
new file mode 100644
index 0000000000..7b81b31383
--- /dev/null
+++ b/docs/concepts/coordinate_spaces.rst
@@ -0,0 +1,272 @@
+.. _coordinate-spaces:
+
+Coordinate spaces
+=================
+
+A *coordinate space* associates a semantic meaning to *coordinate vectors*
+within that space, vectors of real-valued coordinates specifying a coordinate
+for each dimension of the coordinate space.  At any given time, a coordinate
+space has a fixed number of dimensions, called the *rank* of the coordinate
+space.
+
+Commonly, Neuroglancer is used with 3-dimensional data, and hence coordinate
+spaces of rank 3, but Neuroglancer supports degenerate coordinate spaces of rank
+0, and there is no explicit upper limit on the number of dimensions.
+
+In addition to rank, coordinate spaces have several other properties:
+
+- The coordinate space specifies a unique name for each dimension, such as
+  ``x``, ``y``, or ``z``. Dimension names must match the regular expression
+  :regexp:`[a-zA-Z][a-zA-Z_0-9]*['^]?`: they must consist of an ASCII letter,
+  and followed by zero or more ASCII letters, digits, or underscore characters.
+  The optional suffix of ``'`` or ``^`` indicates a :ref:`local or channel
+  dimension<dimension-kinds>`, respectively.
+- For each dimension, the coordinate space specify either:
+  
+  - a *physical unit*, which may include both a base unit and a coefficient,
+    such as ``4e-9 m``, or may omit a base unit to indicate a unitless
+    dimension, such as ``1`` or ``1e3``; or
+  - a *coordinate array* specifying a string label associated with some of the
+    coordinates (indicating a discrete dimension).
+- Coordinate spaces may optionally have a list of associated bounding boxes,
+  from which lower and upper bounds for each coordinate may be inferred.
+
+Neuroglancer makes use of a number of interrelated coordinate spaces and
+associated positions, orientations, and other coordinate transformation
+parameters:
+
+.. graphviz:: coordinate_spaces.dot
+   :caption: Coordinate spaces and transforms in Neuroglancer.  The labels above
+             link to the corresponding description below.
+
+The series of coordinate transformations, starting from the coordinate spaces of
+each :ref:`data source<data-source-coordinate-space>`, into a common
+:ref:`global coordinate space<global-coordinate-space>`, and then into the
+coordinate space for each rendered view, is described below.
+
+.. _data-source-coordinate-space:
+
+Data source coordinate space
+----------------------------
+
+The starting point for all coordinate transformations in Neuroglancer is the
+data source itself. Each :ref:`data source<layer-data-source>` added to a
+:ref:`layer` has an inherent associated coordinate space.
+
+..
+  Screenshots not yet supported
+  .. neuroglancer-screenshot:: concepts/data_source_coordinate_space
+
+The dimension names, physical units or coordinate arrays, and bounds are
+determined from the source data automatically; if dimension names or units
+cannot be determined, default values are chosen by the data source
+implementation.  The dimension names and bounds within the source coordinate
+space are fixed, but the :ref:`coordinate
+transform<data-source-coordinate-transform>` controls how the source data
+coordinate space maps to the :ref:`layer and global coordinate
+spaces<global-coordinate-space>`.
+
+.. _data-source-coordinate-transform:
+
+Coordinate transform
+^^^^^^^^^^^^^^^^^^^^
+
+A configurable :wikipedia:`affine<Affine_transformation>` *coordinate
+transform*, represented by an :wikipedia:`affine transformation
+matrix<Transformation_matrix#Affine_transformations>` and a list of output
+dimension names, maps the source coordinate space to the :ref:`layer coordinate
+space<layer-coordinate-space>` and to the :ref:`global coordinate
+space<global-coordinate-space>`.
+
+The data source provides a default value for the coordinate transform,
+normally an identity matrix.
+
+The user can configure the coordinate transform in three ways:
+
+1. The affine transformation matrix scaling and translation coefficients may be
+   changed directly.  Note that the translation coefficients are in the units
+   specified for the output (layer) dimension.
+
+2. The names of the output dimensions of the transform may be changed.
+   Permuting the output dimension names has a similar effect to permuting the
+   rows of the transformation matrix in the same way, but may be more
+   convenient.
+
+3. The source dimension scales/units may be changed, in order to rescale the
+   input.  This is equivalent to applying an appropriate scale transformation to
+   the affine transformation matrix, but in many cases is more convenient.
+
+   .. note::
+
+      Changing the units of an output dimension does *not* rescale the data, it
+      simply changes the unit used to display coordinates.
+
+The output space of the coordinate transform is a subspace of the :ref:`layer
+coordinate space<layer-coordinate-space>`.  If two data sources associated with
+a layer both have a coordinate transform with an output dimension named ``x``,
+both coordinate transforms are referring to the same dimension ``x``.  In
+contrast, the names of the *source dimensions* of the coordinate transform are
+purely descriptive; if two data sources associated with a layer both have a
+source dimension ``x``, there is no direct correspondence between those two
+source dimensions.
+
+.. _layer-coordinate-space:
+
+Layer coordinate space
+----------------------
+
+The output coordinate spaces of the :ref:`coordinate
+transforms<data-source-coordinate-transform>` of each data source in a given
+layer are *merged* into a single coordinate space that is called the *layer
+coordinate space*:
+
+- If a layer has just a single data source (most common case), then the layer
+  coordinate space is simply the output coordinate space of the coordinate
+  transform.
+
+- In general, if a layer has more than one data source, the layer coordinate
+  space consists of the distinct output dimensions of the coordinate transforms
+  of each of the data sources.
+
+.. _dimension-kinds:
+
+Global, local and channel dimensions
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+A layer coordinate space dimension may be one of three *kinds*, determined based
+on the dimension name:
+
+.. _global-dimensions:
+
+Global dimensions
+~~~~~~~~~~~~~~~~~
+
+Global dimensions have names consisting of only ASCII
+alphanumeric and underscore characters without a special suffix, e.g. ``x``.
+
+- The global dimensions of each layer are merged into the :ref:`global
+  coordinate space<global-coordinate-space>`, which specifies the units for
+  global dimensions.
+
+- A global dimension ``x`` in one layer refers to the same dimension as a
+  global dimension ``x`` in another layer.
+
+- Only global dimensions may be used as :ref:`display
+  dimensions<display-dimensions>`.
+
+.. _local-dimensions:
+
+Local dimensions
+~~~~~~~~~~~~~~~~
+
+Local dimensions have names ending in ``'`` (ASCII single quote), e.g. ``c'``.
+
+- The :ref:`local coordinate space<local-coordinate-space>` of each layer
+  consists of the subset of the layer dimensions that are local dimensions,
+  and specifies the units for local dimensions.
+
+- A local dimension ``c'`` in one layer is completely independent from a local
+  dimension with the same name ``c'`` in another layer.
+
+- Local dimensions may not be used as :ref:`display
+  dimensions<display-dimensions>`. Instead, :ref:`data views<data-view>` always
+  display a cross section at a single position along each local dimension; this
+  position is determined by the :ref:`local position<local-position>`.
+
+- A global dimension with a unique name that is not a :ref:`display
+  dimension<display-dimensions>` may be used as an alternative to a local
+  dimension; a local dimension simply avoids the need to assign unique names,
+  and may be more convenient in some cases.
+
+.. _channel-dimensions:
+
+Channel dimensions
+~~~~~~~~~~~~~~~~~~
+
+:ref:`Image layers<image-layer>` additionally support channel dimensions,
+which have names ending ``^`` (ASCII caret), e.g. ``c^``.
+
+- The :ref:`shader<image-layer-shader>` can access the value at every position
+  within the channel dimensions when computing the output pixel color.  For
+  example, if there is a single channel dimension with a range of ``[0, 3)``,
+  the shader can compute the output pixel color based on the data value at
+  each of the 3 positions.
+
+- Like :ref:`local dimensions<local-dimensions>`, a channel dimension ``c^``
+  in one layer is completely independent from a channel dimension with the
+  same name ``c^`` in another layer.
+
+- A dimension can be used as a channel dimensions only if the data source is
+  unchunked along that dimension.
+
+.. _local-coordinate-space:
+
+Local coordinate space
+^^^^^^^^^^^^^^^^^^^^^^
+
+The local coordinate space of a layer consists of the local dimensions of the
+layer coordinate space.
+
+By default, dimensions are ordered based on when they are first added, with
+dimensions added later ordered after dimensions added earlier, but dimensions
+may be explicitly reordered.
+
+.. _local-position:
+
+Local position
+~~~~~~~~~~~~~~
+
+Each layer has an associated *local position*, specifying for each dimension in
+the :ref:`local coordinate space<local-coordinate-space>` the single slice for
+each dimension in the local coordinate space to be displayed in any views of the
+layer.
+
+.. _global-coordinate-space:
+
+Global coordinate space
+-----------------------
+
+The global coordinate space consists of the global dimensions from the
+coordinate spaces of each layer added to the viewer.
+
+By default, dimensions are ordered based on when they are first added to a layer
+coordinate space, with dimensions added later ordered after dimensions added
+earlier, but dimensions may be explicitly reordered.
+
+
+.. _global center position:
+
+Global center position
+^^^^^^^^^^^^^^^^^^^^^^
+
+The viewer has a single default global position, called the global center
+position, which specifies a center coordinate for each dimension in the
+:ref:`global coordinate space<global-coordinate-space>`.
+
+.. _global-mouse-position:
+
+Global mouse position
+^^^^^^^^^^^^^^^^^^^^^
+
+The position within the :ref:`global-coordinate-space` corresponding to the
+current mouse position within a :ref:`data view<data-view>` is called the
+*global mouse position*.
+
+.. _layer group center position:
+
+Layer group center position
+^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+Additionally, each layer group sub-viewer has a separate center position in the
+:ref:`global coordinate space<global-coordinate-space>` which may optionally be
+fixed to, or linked by an offset to, the global center position.
+
+.. _display-dimensions:
+
+Display dimensions
+------------------
+
+:ref:`Data views<data-view>` project data from at most three
+:ref:`global dimensions<global-coordinate-space>`; these projected dimensions
+are called *display dimensions*. From all other global dimensions, only a
+cross section of the data is displayed at any given time.
diff --git a/docs/concepts/data_source_coordinate_space.screenshot.py b/docs/concepts/data_source_coordinate_space.screenshot.py
new file mode 100644
index 0000000000..1e7b04c2b2
--- /dev/null
+++ b/docs/concepts/data_source_coordinate_space.screenshot.py
@@ -0,0 +1,59 @@
+from selenium.webdriver.common.by import By
+
+rec.begin(
+    {
+        "layers": [
+            {
+                "type": "image",
+                "source": "precomputed://gs://neuroglancer-public-data/flyem_fib-25/image",
+                "tab": "source",
+                "name": "image",
+            }
+        ],
+        "selectedLayer": {"layer": "image", "visible": True},
+    }
+)
+
+# rec.highlight_element_with_outline(
+#     elements=(
+#         rec.webdriver.driver.find_elements(
+#             By.CSS_SELECTOR, '.neuroglancer-coordinate-space-transform-source-label') +
+#         rec.webdriver.driver.find_elements(
+#             By.CSS_SELECTOR, '.neuroglancer-coordinate-space-transform-input-scale-container')),
+#     caption='Data source coordinate space',
+#     direction='left',
+# )
+
+# rec.highlight_element_with_outline(
+#     elements=(rec.webdriver.driver.find_elements(By.CSS_SELECTOR,
+#                                                  '.neuroglancer-coordinate-space-transform-coeff') +
+#               rec.webdriver.driver.find_elements(
+#                   By.CSS_SELECTOR, '.neuroglancer-coordinate-space-transform-output-label')),
+#     caption='Coordinate transform',
+#     direction='left')
+
+
+rec.add_label(
+    elements=(
+        rec.webdriver.driver.find_elements(
+            By.CSS_SELECTOR, ".neuroglancer-coordinate-space-transform-source-label"
+        )
+        + rec.webdriver.driver.find_elements(
+            By.CSS_SELECTOR,
+            ".neuroglancer-coordinate-space-transform-input-scale-container",
+        )
+    ),
+    caption="Data source coordinate space",
+)
+
+rec.add_label(
+    elements=(
+        rec.webdriver.driver.find_elements(
+            By.CSS_SELECTOR, ".neuroglancer-coordinate-space-transform-coeff"
+        )
+        + rec.webdriver.driver.find_elements(
+            By.CSS_SELECTOR, ".neuroglancer-coordinate-space-transform-output-label"
+        )
+    ),
+    caption="Coordinate transform",
+)
diff --git a/docs/concepts/data_source_coordinate_transform.screenshot.py b/docs/concepts/data_source_coordinate_transform.screenshot.py
new file mode 100644
index 0000000000..7b0301ae07
--- /dev/null
+++ b/docs/concepts/data_source_coordinate_transform.screenshot.py
@@ -0,0 +1,59 @@
+from selenium.webdriver.common.by import By
+
+rec.begin(
+    {
+        "layers": [
+            {
+                "type": "image",
+                "source": "precomputed://gs://neuroglancer-public-data/flyem_fib-25/image",
+                "tab": "source",
+                "name": "image",
+            }
+        ],
+        "selectedLayer": {"layer": "image", "visible": True},
+    }
+)
+
+
+rec.add_labels(
+    [
+        (
+            rec.webdriver.driver.find_elements(
+                By.CSS_SELECTOR,
+                ".neuroglancer-coordinate-space-transform-input-scale-container",
+            ),
+            "top",
+            "Source dimension units",
+        ),
+        (
+            rec.webdriver.driver.find_elements(
+                By.CSS_SELECTOR, ".neuroglancer-coordinate-space-transform-coeff"
+            ),
+            "bottom",
+            "Affine transform matrix",
+        ),
+        (
+            rec.webdriver.driver.find_elements(
+                By.CSS_SELECTOR, ".neuroglancer-coordinate-space-transform-output-label"
+            ),
+            "left",
+            "Output dimension labels",
+        ),
+        (
+            rec.webdriver.driver.find_elements(
+                By.CSS_SELECTOR,
+                ".neuroglancer-coordinate-space-transform-output-scale-container",
+            ),
+            "bottom",
+            "Output dimension units",
+        ),
+        (
+            rec.webdriver.driver.find_elements(
+                By.CSS_SELECTOR,
+                ".neuroglancer-coordinate-space-transform-output-extend",
+            ),
+            "left",
+            "Extend output space",
+        ),
+    ]
+)
diff --git a/docs/concepts/data_views.rst b/docs/concepts/data_views.rst
new file mode 100644
index 0000000000..48b18cfbda
--- /dev/null
+++ b/docs/concepts/data_views.rst
@@ -0,0 +1,18 @@
+.. _data-view:
+
+Data view
+=========
+
+.. _cross-section-view:
+
+Cross-section view
+------------------
+
+TODO
+
+.. _projection-view:
+
+3-D projection view
+-------------------
+
+TODO
diff --git a/docs/concepts/layers.rst b/docs/concepts/layers.rst
new file mode 100644
index 0000000000..319cf7d9db
--- /dev/null
+++ b/docs/concepts/layers.rst
@@ -0,0 +1,26 @@
+.. _layer:
+
+Layer
+=====
+
+.. _layer-data-source:
+
+Layer data source
+-----------------
+
+TODO
+
+Layer kinds
+-----------
+
+.. _image-layer:
+
+Image layer
+^^^^^^^^^^^
+
+.. _image-layer-shader:
+
+Image layer shader
+~~~~~~~~~~~~~~~~~~
+
+TODO
diff --git a/docs/conf.py b/docs/conf.py
new file mode 100644
index 0000000000..1b16283f0f
--- /dev/null
+++ b/docs/conf.py
@@ -0,0 +1,328 @@
+project = "Neuroglancer"
+copyright = "2021 The Neuroglancer Authors"
+version = ""
+release = ""
+
+import typing
+from typing import NamedTuple, Optional
+import os
+import re
+import sys
+
+import docutils.nodes
+import sphinx.application
+import sphinx.addnodes
+import sphinx.domains.python
+import sphinx.environment
+
+
+os.environ["NEUROGLANCER_BUILDING_DOCS"] = "1"
+
+extensions = [
+    "sphinx.ext.extlinks",
+    "sphinx.ext.autodoc",
+    "sphinx.ext.intersphinx",
+    "sphinx.ext.graphviz",
+    "sphinx.ext.mathjax",
+    "sphinx.ext.napoleon",
+    "sphinx_immaterial",
+    "sphinx_immaterial.apidoc.python.apigen",
+    "sphinx_immaterial.apidoc.json.domain",
+    "sphinx_immaterial.graphviz",
+    "sphinx_immaterial.apidoc.format_signatures",
+]
+
+napoleon_numpy_docstring = False
+napoleon_use_admonition_for_examples = True
+napoleon_use_admonition_for_notes = True
+
+
+html_title = "Neuroglancer"
+
+# Don't include "View page source" links, since they aren't very helpful,
+# especially for generated pages.
+html_show_sourcelink = False
+html_copy_source = False
+
+# Skip unnecessary footer text.
+html_show_sphinx = False
+html_show_copyright = False
+
+# Override default of `utf-8-sig` which can cause problems with autosummary due
+# to the extra Unicode Byte Order Mark that gets inserted.
+source_encoding = "utf-8"
+
+source_suffix = ".rst"
+master_doc = "index"
+language = "en"
+
+html_use_index = False
+
+intersphinx_mapping = {
+    "python": (
+        "https://docs.python.org/3",
+        ("intersphinx_inv/python3.inv", None),
+    ),
+    "numpy": (
+        "https://numpy.org/doc/stable/",
+        ("intersphinx_inv/numpy.inv", None),
+    ),
+    "tensorstore": (
+        "https://google.github.io/tensorstore/",
+        ("intersphinx_inv/tensorstore.inv", None),
+    ),
+    # "sphinx_docs": ("https://www.sphinx-doc.org/en/master", None),
+}
+
+html_theme = "sphinx_immaterial"
+
+html_theme_options = {
+    "icon": {
+        "logo": "material/library",
+        "repo": "fontawesome/brands/github",
+    },
+    "site_url": "https://google.github.io/neuroglancer/",
+    "repo_url": "https://github.com/google/neuroglancer/",
+    "edit_uri": "blob/master/docs",
+    "features": [
+        "navigation.expand",
+        "navigation.tabs",
+        # "toc.integrate",
+        "navigation.sections",
+        # "navigation.instant",
+        # "header.autohide",
+        "navigation.top",
+        "toc.sticky",
+        "toc.follow",
+    ],
+    "palette": [
+        {
+            "media": "(prefers-color-scheme: dark)",
+            "scheme": "slate",
+            "primary": "green",
+            "accent": "light blue",
+            "toggle": {
+                "icon": "material/lightbulb",
+                "name": "Switch to light mode",
+            },
+        },
+        {
+            "media": "(prefers-color-scheme: light)",
+            "scheme": "default",
+            "primary": "green",
+            "accent": "light blue",
+            "toggle": {
+                "icon": "material/lightbulb-outline",
+                "name": "Switch to dark mode",
+            },
+        },
+    ],
+}
+
+default_role = "any"
+
+# Warn about missing references
+nitpicky = True
+
+python_apigen_modules = {
+    "neuroglancer.viewer_state": "python/api/",
+    "neuroglancer.viewer_config_state": "python/api/",
+    "neuroglancer.trackable_state": "python/api/",
+    "neuroglancer.json_wrappers": "python/api/",
+    "neuroglancer": "python/api/",
+}
+
+python_apigen_default_groups = [
+    (r"class:neuroglancer\.viewer_state\..*", "viewer-state"),
+    (r"(class|function):neuroglancer\.viewer_config_state\..*", "viewer-config-state"),
+    (r"class:neuroglancer\.coordinate_space\..*", "coordinate-space"),
+    (r"class:neuroglancer\.viewer_state\..*Tool", "viewer-state-tools"),
+    (r"class:neuroglancer\.viewer_state\..*Layer", "viewer-state-layers"),
+    (r"class:neuroglancer\.viewer\..*", "core"),
+    (r"class:neuroglancer\.server\..*", "server"),
+]
+
+python_apigen_rst_prolog = """
+.. default-role:: py:obj
+
+.. default-literal-role:: python
+
+.. highlight:: python
+
+"""
+
+
+json_schemas = ["json_schema/*.yml"]
+
+rst_prolog = """
+.. role:: python(code)
+   :language: python
+   :class: highlight
+
+.. role:: json(code)
+   :language: json
+   :class: highlight
+
+"""
+
+json_schema_rst_prolog = """
+.. default-role:: json:schema
+
+.. default-literal-role:: json
+
+.. highlight:: json
+"""
+
+
+graphviz_output_format = "svg"
+
+extlinks = {
+    "wikipedia": ("https://en.wikipedia.org/wiki/%s", None),
+}
+
+
+python_module_names_to_strip_from_xrefs = [
+    "neuroglancer.viewer_state",
+    "neuroglancer.trackable_state",
+    "neuroglancer.viewer_config_state",
+    "neuroglancer",
+    "collections.abc",
+    "numbers",
+    "numpy.typing",
+    "numpy",
+]
+
+object_description_options = [
+    ("py:.*", dict(black_format_style={})),
+]
+
+
+# Monkey patch numpy.typing.NDArray
+def _monkey_patch_numpy_typing_ndarray():
+    import numpy.typing
+
+    T = typing.TypeVar("T")
+
+    class NDArray(typing.Generic[T]):
+        pass
+
+    NDArray.__module__ = "numpy.typing"
+
+    numpy.typing.NDArray = NDArray
+
+
+_monkey_patch_numpy_typing_ndarray()
+
+
+# Monkey patch Sphinx to generate custom cross references for specific type
+# annotations.
+#
+# The Sphinx Python domain generates a `py:class` cross reference for type
+# annotations.  However, in some cases in the TensorStore documentation, type
+# annotations are used to refer to targets that are not actual Python classes,
+# such as `DownsampleMethod`, `DimSelectionLike`, or `NumpyIndexingSpec`.
+# Additionally, some types like `numpy.typing.ArrayLike` are `py:data` objects
+# and can't be referenced as `py:class`.
+class TypeXrefTarget(NamedTuple):
+    domain: str
+    reftype: str
+    target: str
+    title: str
+
+
+python_type_to_xref_mappings = {
+    f"numpy.{name}": TypeXrefTarget("py", "obj", f"numpy.{name}", name)
+    for name in [
+        "int64",
+        "uint64",
+        "float32",
+        "float64",
+    ]
+}
+
+python_type_to_xref_mappings["numpy.typing.NDArray"] = TypeXrefTarget(
+    "py", "obj", "numpy.typing.NDArray", "NDArray"
+)
+
+python_type_to_xref_mappings["types.EllipsisType"] = TypeXrefTarget(
+    "py", "obj", "types.EllipsisType", "EllipsisType"
+)
+
+python_strip_property_prefix = True
+
+
+_orig_python_type_to_xref = sphinx.domains.python.type_to_xref
+
+
+def _python_type_to_xref(
+    target: str,
+    env: sphinx.environment.BuildEnvironment,
+    *,
+    suppress_prefix: bool = False,
+) -> sphinx.addnodes.pending_xref:
+    xref_info = python_type_to_xref_mappings.get(target)
+    if xref_info is not None:
+        return sphinx.addnodes.pending_xref(
+            "",
+            docutils.nodes.Text(xref_info.title),
+            refdomain=xref_info.domain,
+            reftype=xref_info.reftype,
+            reftarget=xref_info.target,
+            refspecific=False,
+            refexplicit=True,
+            refwarn=True,
+        )
+    return _orig_python_type_to_xref(target, env, suppress_prefix=suppress_prefix)
+
+
+sphinx.domains.python.type_to_xref = _python_type_to_xref
+sphinx.domains.python._annotations.type_to_xref = _python_type_to_xref
+
+
+python_type_aliases = {
+    "concurrent.futures._base.Future": "concurrent.futures.Future",
+    "ellipsis": "types.EllipsisType",
+}
+
+
+def _should_document_python_class_base(base: type) -> bool:
+    if base.__name__.startswith("_"):
+        return False
+    if base.__module__ in ("neuroglancer.json_wrappers", "neuroglancer.viewer_base"):
+        return False
+    return True
+
+
+PYTHON_MEMBER_SKIP_PATTERNS = re.compile(r"supports_(readonly|validation)")
+
+
+def _autodoc_skip_member(
+    app: sphinx.application.Sphinx,
+    what: str,
+    name: str,
+    obj,
+    skip: bool,
+    options,
+):
+    if PYTHON_MEMBER_SKIP_PATTERNS.fullmatch(name) is not None:
+        return True
+    return None
+
+
+def _python_apigen_skip_base(
+    app: sphinx.application.Sphinx, subclass: type, base: type
+):
+    if base.__module__ == "collections.abc":
+        return True
+    return None
+
+
+def setup(app: sphinx.application.Sphinx):
+    # Ignore certain base classes.
+    def _autodoc_process_bases(app, name, obj, options, bases):
+        bases[:] = [base for base in bases if _should_document_python_class_base(base)]
+
+    app.connect("autodoc-process-bases", _autodoc_process_bases)
+
+    app.connect("autodoc-skip-member", _autodoc_skip_member)
+    app.connect("python-apigen-skip-base", _python_apigen_skip_base)
diff --git a/docs/index.rst b/docs/index.rst
new file mode 100644
index 0000000000..bdd78a65f7
--- /dev/null
+++ b/docs/index.rst
@@ -0,0 +1,34 @@
+
+Neuroglancer
+============
+
+.. toctree::
+   :hidden:
+   :caption: User Guide
+
+   user-guide/navigation
+
+.. toctree::
+   :hidden:
+   :caption: Concepts
+
+   concepts/coordinate_spaces
+   concepts/data_views
+   concepts/layers
+
+.. toctree::
+   :hidden:
+   :caption: JSON API
+
+   json/api/index
+
+.. toctree::
+   :hidden:
+   :caption: Python API
+
+   python/api/index
+
+Neuroglancer is a WebGL-based viewer for volumetric data.  It is capable of
+displaying arbitrary (non axis-aligned) cross-sectional views of volumetric
+data, as well as 3-D meshes and line-segment based models (skeletons).
+
diff --git a/docs/intersphinx_inv/numpy.inv b/docs/intersphinx_inv/numpy.inv
new file mode 100644
index 0000000000..9a95a49eeb
Binary files /dev/null and b/docs/intersphinx_inv/numpy.inv differ
diff --git a/docs/intersphinx_inv/python3.inv b/docs/intersphinx_inv/python3.inv
new file mode 100644
index 0000000000..adf47f5d36
Binary files /dev/null and b/docs/intersphinx_inv/python3.inv differ
diff --git a/docs/intersphinx_inv/tensorstore.inv b/docs/intersphinx_inv/tensorstore.inv
new file mode 100644
index 0000000000..33fcd83b0c
Binary files /dev/null and b/docs/intersphinx_inv/tensorstore.inv differ
diff --git a/docs/json/api/index.rst b/docs/json/api/index.rst
new file mode 100644
index 0000000000..0efe01ff0d
--- /dev/null
+++ b/docs/json/api/index.rst
@@ -0,0 +1,33 @@
+API Reference
+=============
+
+.. json:schema:: ViewerState
+
+.. json:schema:: Layer
+
+.. json:schema:: Position
+
+.. json:schema:: Orientation
+
+.. json:schema:: DepthRange
+
+.. json:schema:: DisplayDimensions
+
+.. json:schema:: DisplayScale
+
+.. json:schema:: RelativeDisplayScales
+
+.. json:schema:: CoordinateSpace
+
+Layout
+------
+
+.. json:schema:: Layout
+
+.. json:schema:: DataPanelLayoutType
+
+.. json:schema:: StackLayout
+
+.. json:schema:: DataPanelLayout
+
+.. json:schema:: LayerGroupViewer
diff --git a/docs/json_schema/layer.yml b/docs/json_schema/layer.yml
new file mode 100644
index 0000000000..aaac47cbd1
--- /dev/null
+++ b/docs/json_schema/layer.yml
@@ -0,0 +1,14 @@
+$schema: http://json-schema.org/draft-07/schema#
+$id: Layer
+title: "Layer within a Neuroglancer instance."
+type: object
+properties:
+  type:
+    type: string
+    title: "Specifies the layer type."
+  name:
+    type: string
+    title: "Specifies the layer name to show in the UI."
+  visible:
+    type: boolean
+    title: "Indicates whether the layer is displayed in 2d and 3d projections."
diff --git a/docs/json_schema/viewer_state.yml b/docs/json_schema/viewer_state.yml
new file mode 100644
index 0000000000..5caed60874
--- /dev/null
+++ b/docs/json_schema/viewer_state.yml
@@ -0,0 +1,200 @@
+$schema: http://json-schema.org/draft-07/schema#
+$id: ViewerState
+title: "Complete state of a Neuroglancer instance."
+type: object
+properties:
+  dimensions:
+    $ref: CoordinateSpace
+    title: "Global coordinate space."
+  relativeDisplayScales:
+    $ref: RelativeDisplayScales
+    title: "Default additional relative display scale factors for each `global dimension<.dimensions>`."
+  displayDimensions:
+    $ref: DisplayDimensions
+    title: "Default display dimensions for 2d and 3d projections."
+  position:
+    $ref: Position
+    title: "Global position within each `global dimension<.dimensions>`."
+  crossSectionOrientation:
+    $ref: Orientation
+    title: "Default orientation within the `.displayDimensions`."
+  crossSectionScale:
+    $ref: DisplayScale
+    title: "Default display scale (zoom level) for cross-section views."
+  crossSectionDepth:
+    $ref: DepthRange
+    title: "Default depth-of-field for cross-section views."
+  projectionOrientation:
+    $ref: Orientation
+    title: "Default orientation within the `.displayDimensions`."
+  projectionScale:
+    $ref: DisplayScale
+    title: "Default display scale (zoom level) for projection views."
+  projectionDepth:
+    $ref: DepthRange
+    title: "Default depth-of-field for projection views."
+  layers:
+    type: array
+    items:
+      $ref: Layer
+  showAxisLines:
+    type: boolean
+    title: "Indicates whether to show the red/green/blue axis lines."
+    default: true
+  wireFrame:
+    type: boolean
+    title: "Indicates whether to enable wireframe rendering mode (for debugging)."
+    default: false
+  showScaleBar:
+    type: boolean
+    title: "Indicates whether to show scale bars."
+    default: true
+  showDefaultAnnotations:
+    type: boolean
+    title: "Indicates whether to show bounding boxes of data sources."
+    default: true
+  showSlices:
+    type: boolean
+    title: "Indicates whether to show cross sections in the 3-d view of `~DataPanelLayoutType.4panel` layouts."
+    default: true
+  gpuMemoryLimit:
+    type: integer
+    title: "GPU memory limit, in bytes."
+  systemMemoryLimit:
+    type: integer
+    title: "System memory limit, in bytes."
+  concurrentDownloads:
+    type: integer
+    title: "Maximum number of concurrent downloads."
+  prefetch:
+    type: boolean
+    title: "Indicates whether to use adaptive prefetching."
+    default: true
+  title:
+    type: string
+    title: "Additional text to include in the page title."
+  layout:
+    $ref: Layout
+    title: "Data panel and layer group layout."
+definitions:
+  CoordinateSpace:
+    $id: CoordinateSpace
+    title: "Specifies a coordinate space."
+    type: object
+  DisplayDimensions:
+    $id: DisplayDimensions
+    type: array
+    items:
+      type: string
+    minItems: 0
+    maxItems: 3
+    uniqueItems: true
+    title: "Specifies the display dimensions for 2d and 3d projections."
+    description: |
+  Orientation:
+    $id: Orientation
+    title: "Specifies a 3-d orientation as a unit quaternion."
+    description: |
+      For the 3-d projection view and for the `DataPanelLayoutType.xy` cross-section
+      view, with the default orientation of ``[0, 0, 0, 1]`` the first display
+      dimension (red axis) points right, the second display dimension (green
+      axis) points down, and the third display dimension (blue axis) points away
+      from the camera.
+    type: array
+    items:
+      - type: number
+      - type: number
+      - type: number
+      - type: number
+    default: [0, 0, 0, 1]
+  DisplayScale:
+    $id: DisplayScale
+    title: "Specifies the scale (zoom level) of a cross-section or 3d projection view."
+    description: |
+      For cross-section views, the scale is specified in canonical voxels per
+      screen pixel, and defaults to ``1``.  For 3d projection views, the scale is specified in
+      canonical voxels per viewport height.
+    type: number
+    exclusiveMinimum: 0
+  RelativeDisplayScales:
+    $id: RelativeDisplayScales
+    title: "Specifies additional relative display scale factors for each `global dimension<ViewerState.dimensions>`."
+    description: |
+      The length must be equal to the number of `global
+      dimensions<ViewerState.dimensions>`.  Defaults to a vector of all ones.
+    type: array
+    items:
+      type: number
+  Position:
+    $id: Position
+    title: "Specifies the position within a `CoordinateSpace`."
+    description: |
+      The length must be equal to the number of dimensions in the coordinate
+      space.
+    type: array
+    items:
+      type: number
+  DepthRange:
+    $id: DepthRange
+    title: "Specifies the depth-of-field for cross section or 3d projection views."
+    type: number
+    exclusiveMinimum: 0
+  Layout:
+    $id: Layout
+    title: "Specifies a hierarchical grid of layer groups and data views."
+    oneOf:
+      - $ref: DataPanelLayoutType
+      - $ref: DataPanelLayout
+      - $ref: LayerGroupViewer
+      - $ref: StackLayout
+  DataPanelLayoutType:
+    $id: DataPanelLayoutType
+    title: "Specifies a layout of 3-d projection and 2-d cross-section views for a layer group."
+    oneOf:
+      - const: "4panel"
+        title: "2x2 grid layout with `.xy`, `.xz`, `.3d`, and `.yz` panels."
+        description: |
+          The top left is an `.xy` cross-section view, the top right is an `.xz`
+          cross-section view, the bottom left is a `.3d` projection view, and the
+          bottom right is a `.yz` cross-section view.
+      - const: "xy"
+        title: "Single cross-section view in the default orientation."
+        description: |
+          with the first display dimension (red)
+          pointing right, the second display dimension (green) pointing down, and
+          the third display dimension (blue) pointing away from the camera.
+      - const: "xz"
+        title: |
+          Single cross-section view with the first display dimension (red)
+          pointing right, the third display dimension (blue) pointing down, and
+          the second display dimension (green) pointing towards the camera.
+      - const: "yz"
+        title: |
+          Single cross-section view with the second display dimension (green)
+          pointing down, the third display dimension (blue) pointing left, and
+          the first display dimension (red) pointing away from the camera.
+      - const: "3d"
+        title: "Single 3-d projection view."
+      - const: "xy-3d"
+      - const: "xz-3d"
+      - const: "yz-3d"
+  DataPanelLayout:
+    $id: DataPanelLayout
+    title: "Describes the :ref:`data views<data-view>` to display."
+    type: object
+    properties:
+      type:
+        $ref: DataPanelLayoutType
+        title: Indicates the layout type.
+      orthographicProjection:
+        type: boolean
+        title: |
+          Indicates whether the :ref:`projection views<projection-view>`, if
+          present, uses orthographic rather than perspective projection.
+        default: false
+  LayerGroupViewer:
+    $id: LayerGroupViewer
+    title: "Specifies a `DataPanelLayout` for a subset of the layers."
+  StackLayout:
+    $id: StackLayout
+    title: "Specifies a row or column of sub-layouts."
diff --git a/docs/pyproject.toml b/docs/pyproject.toml
new file mode 100644
index 0000000000..c9184b31a1
--- /dev/null
+++ b/docs/pyproject.toml
@@ -0,0 +1,2 @@
+[tool.ruff]
+exclude = ["*"]
diff --git a/docs/python/api/.gitignore b/docs/python/api/.gitignore
new file mode 100644
index 0000000000..e15a7a67e2
--- /dev/null
+++ b/docs/python/api/.gitignore
@@ -0,0 +1,2 @@
+-index.rst
+*.rst
diff --git a/docs/python/api/index.rst b/docs/python/api/index.rst
new file mode 100644
index 0000000000..074fbbec8b
--- /dev/null
+++ b/docs/python/api/index.rst
@@ -0,0 +1,78 @@
+API Reference
+=============
+
+Core
+----
+
+.. python-apigen-group:: core
+
+Viewer state
+------------
+
+.. python-apigen-group:: viewer-state
+
+Segment sets
+^^^^^^^^^^^^
+
+.. python-apigen-group:: viewer-state-segments
+
+URL representation
+^^^^^^^^^^^^^^^^^^
+
+.. python-apigen-group:: viewer-state-url
+
+Coordinate space
+^^^^^^^^^^^^^^^^
+
+.. python-apigen-group:: coordinate-space
+
+Layers
+^^^^^^
+
+.. python-apigen-group:: viewer-state-layers
+
+Tools
+^^^^^
+
+.. python-apigen-group:: viewer-state-tools
+
+Viewer Config state
+-------------------
+
+.. python-apigen-group:: viewer-config-state
+
+
+Exposing local data as data source
+----------------------------------
+
+.. python-apigen-group:: serve-data
+
+Credentials
+-----------
+
+.. python-apigen-group:: credentials
+
+Capturing Screenshots
+---------------------
+
+.. python-apigen-group:: screenshots
+
+Server
+------
+
+.. python-apigen-group:: server
+
+
+Utilities
+---------
+
+JSON Containers
+^^^^^^^^^^^^^^^
+
+.. python-apigen-group:: json-containers
+
+
+Trackable state
+^^^^^^^^^^^^^^^
+
+.. python-apigen-group:: trackable-state
diff --git a/docs/requirements.txt b/docs/requirements.txt
new file mode 100644
index 0000000000..1cb6b52fc3
--- /dev/null
+++ b/docs/requirements.txt
@@ -0,0 +1,7 @@
+sphinx-immaterial[black,json]>=0.12.0
+sphinx>=7.4.0
+pyyaml
+jsonschema
+numpy
+tensorstore
+-r ../python/requirements.txt
diff --git a/docs/update_intersphinx_inventories.py b/docs/update_intersphinx_inventories.py
new file mode 100644
index 0000000000..8ab6fd7a63
--- /dev/null
+++ b/docs/update_intersphinx_inventories.py
@@ -0,0 +1,42 @@
+# Copyright 2020 The TensorStore Authors
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""Downloads the intersphinx inventories referenced by `conf.py`.
+
+This allows the documentation build to be hermetic.
+"""
+
+import os
+import requests
+
+if __name__ == "__main__":
+    docs_dir = os.path.dirname(os.path.realpath(__file__))
+    conf_path = os.path.join(docs_dir, "conf.py")
+    conf_module = {
+        "__file__": conf_path,
+    }
+    with open(conf_path, "r") as f:
+        exec(f.read(), conf_module)
+    intersphinx_mapping = conf_module["intersphinx_mapping"]
+    for _, (url, (local_path, _)) in intersphinx_mapping.items():
+        if not url.endswith("/"):
+            url += "/"
+        full_url = url + "objects.inv"
+        full_path = os.path.join(docs_dir, local_path)
+        print("Fetching: %r" % (full_url,))
+        resp = requests.get(full_url)
+        resp.raise_for_status()
+        os.makedirs(os.path.dirname(full_path), exist_ok=True)
+        with open(full_path, "wb") as f:
+            f.write(resp.content)
+        print("Fetched %s -> %s" % (full_url, full_path))
diff --git a/docs/user-guide/navigation.rst b/docs/user-guide/navigation.rst
new file mode 100644
index 0000000000..7ef701e8e2
--- /dev/null
+++ b/docs/user-guide/navigation.rst
@@ -0,0 +1,43 @@
+Navigation
+==========
+
+The position, orientation, and zoom level of :ref:`data views<data-view>` can
+be controlled using :ref:`mouse<mouse-navigation>` and
+:ref:`keyboard<keyboard-navigation>` bindings as well as UI controls.
+
+.. _mouse-navigation:
+
+Mouse navigation
+----------------
+
+Translating cross-section views in-plane
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+You can translate a :ref:`cross-section view<cross-section-view>` in-plane by
+left-clicking within the view and dragging the mouse while holding down the left
+button. As you drag, the `global center position` changes but the :ref:`global mouse
+position<global-mouse-position>` remains fixed.
+
+..
+  Videos not yet supported
+  .. neuroglancer-video:: user-guide/navigation_mouse_cross_section_translate
+
+Adjusting the zoom level of cross-section views
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+You can adjust the zoom level of a :ref:`cross-section view<cross-section-view>`
+by holding down :kbd:`Control` and moving the mouse wheel up or down. As you
+move the mouse wheel, the `global center position` may change but the
+:ref:`global mouse position<global-mouse-position>` remains fixed.
+
+..
+  Videos not yet supported
+  .. neuroglancer-video:: user-guide/navigation_mouse_cross_section_zoom
+
+
+.. _keyboard-navigation:
+
+Keyboard navigation
+-------------------
+
+TODO
diff --git a/docs/user-guide/navigation_mouse_cross_section_translate.video.py b/docs/user-guide/navigation_mouse_cross_section_translate.video.py
new file mode 100644
index 0000000000..afd6c3508b
--- /dev/null
+++ b/docs/user-guide/navigation_mouse_cross_section_translate.video.py
@@ -0,0 +1,28 @@
+rec.begin(
+    {
+        "dimensions": {"x": [8e-9, "m"], "y": [8e-9, "m"], "z": [8e-9, "m"]},
+        "position": [3242.833251953125, 3334.499755859375, 4045.5],
+        "crossSectionScale": 1,
+        "projectionOrientation": [
+            0.23160003125667572,
+            0.20444951951503754,
+            -0.017615893855690956,
+            0.9509214162826538,
+        ],
+        "projectionScale": 512,
+        "layers": [
+            {
+                "type": "image",
+                "source": "precomputed://gs://neuroglancer-public-data/flyem_fib-25/image",
+                "tab": "source",
+                "name": "image",
+            }
+        ],
+        "selectedLayer": {"layer": "image"},
+        "layout": "4panel",
+    }
+)
+panel = rec.get_data_panels()[0]
+rec.move_to_element_smoothly(panel, 0.1, 0.1)
+with rec.mouse_buttons_held(1):
+    rec.move_to_element_smoothly(panel, 0.8, 0.9, speed=0.1)
diff --git a/docs/user-guide/navigation_mouse_cross_section_zoom.video.py b/docs/user-guide/navigation_mouse_cross_section_zoom.video.py
new file mode 100644
index 0000000000..9d176578ba
--- /dev/null
+++ b/docs/user-guide/navigation_mouse_cross_section_zoom.video.py
@@ -0,0 +1,30 @@
+rec.begin(
+    {
+        "dimensions": {"x": [8e-9, "m"], "y": [8e-9, "m"], "z": [8e-9, "m"]},
+        "position": [3242.833251953125, 3334.499755859375, 4045.5],
+        "crossSectionScale": 1,
+        "projectionOrientation": [
+            0.23160003125667572,
+            0.20444951951503754,
+            -0.017615893855690956,
+            0.9509214162826538,
+        ],
+        "projectionScale": 512,
+        "layers": [
+            {
+                "type": "image",
+                "source": "precomputed://gs://neuroglancer-public-data/flyem_fib-25/image",
+                "tab": "source",
+                "name": "image",
+            }
+        ],
+        "selectedLayer": {"layer": "image"},
+        "layout": "4panel",
+    }
+)
+panel = rec.get_data_panels()[0]
+rec.move_to_element_smoothly(panel, 0.8, 0.7)
+time.sleep(0.5)
+with rec.keys_held("leftctrl"):
+    rec.mouse_wheel_smoothly(-0.3)
+    rec.mouse_wheel_smoothly(0.3)
diff --git a/firebase.json b/firebase.json
index ae0bf3ae7e..1e4aed1f60 100644
--- a/firebase.json
+++ b/firebase.json
@@ -2,5 +2,9 @@
   "hosting": {
     "target": "app",
     "public": "dist/client"
+  },
+  "hosting": {
+    "target": "docs",
+    "public": "dist/docs"
   }
 }
diff --git a/noxfile.py b/noxfile.py
index f033de152a..147d604f91 100644
--- a/noxfile.py
+++ b/noxfile.py
@@ -1,3 +1,5 @@
+import os
+
 import nox
 
 nox.options.reuse_existing_virtualenvs = True
@@ -20,3 +22,22 @@ def format(session):
 def mypy(session):
     session.install("-r", "python/requirements-mypy.txt")
     session.run("mypy", ".")
+
+
+@nox.session
+def docs(session):
+    session.install("-r", "docs/requirements.txt")
+    session.run(
+        "sphinx-build",
+        "docs",
+        "dist/docs",
+        "-E",
+        "-j",
+        "auto",
+        "-T",
+        "-W",
+        "--keep-going",
+        env={
+            "PYTHONPATH": os.path.join(os.path.dirname(__file__), "python"),
+        },
+    )