diff --git a/README.md b/README.md
index ba9d181..53cffd7 100644
--- a/README.md
+++ b/README.md
@@ -75,6 +75,8 @@ _Actions for managing releases._
 
 #### - [Create](actions/release/create/README.md)
 
+#### - [Summarize changelog](actions/release/summarize-changelog/README.md)
+
 ### Workflow
 
 _Actions for managing workflows._
diff --git a/actions/release/summarize-changelog/README.md b/actions/release/summarize-changelog/README.md
new file mode 100644
index 0000000..7446987
--- /dev/null
+++ b/actions/release/summarize-changelog/README.md
@@ -0,0 +1,62 @@
+# GitHub Action: Release - Summarize Changelog
+
+## Overview
+
+Compile a release changelog from all commits between two refs.
+
+Features:
+
+- Supports conventional commit grouping.
+- Supports configurable Markdown templates.
+- Can inject a summary generated by any LLM provider (`llm-summary` input).
+- Can generate summary through LangChain using multiple providers (`openai`, `anthropic`, `google-genai`).
+- Can generate the summary from the produced `llm-prompt` via a custom command (`llm-summary-command` input).
+- Exposes a provider-agnostic `llm-prompt` output to integrate with any LLM action.
+
+## Usage
+
+```yaml
+- id: changelog
+  uses: hoverkraft-tech/ci-github-publish/actions/release/summarize-changelog@main
+  with:
+    base-ref: v1.2.0
+    head-ref: HEAD
+    conventional-commits: "true"
+    llm-summary: ""
+    llm-provider: "openai"
+    llm-model: "gpt-4o-mini"
+    llm-api-key: ${{ secrets.OPENAI_API_KEY }}
+    llm-base-url: "https://api.openai.com/v1"
+    llm-summary-command: ""
+    markdown-template: |
+      ## Release notes
+      Range: `{{base_ref}}..{{head_ref}}`
+
+      {{summary}}
+
+      {{changes}}
+```
+
+## Inputs
+
+| Input                  | Description                                                                        | Required | Default                     |
+| ---------------------- | ---------------------------------------------------------------------------------- | -------- | --------------------------- |
+| `base-ref`             | Base Git ref (excluded from the range).                                            | true     | -                           |
+| `head-ref`             | Head Git ref (included in the range).                                              | true     | -                           |
+| `conventional-commits` | Group commit messages by conventional commit type.                                 | false    | `true`                      |
+| `llm-summary`          | Optional summary generated by any LLM provider.                                    | false    | `""`                        |
+| `llm-provider`         | LLM provider used with LangChain (`openai`, `anthropic`, `google-genai`).          | false    | `openai`                    |
+| `llm-model`            | Optional model used to generate summary from `llm-prompt`.                         | false    | `""`                        |
+| `llm-api-key`          | Optional API key for the selected LLM provider.                                    | false    | `""`                        |
+| `llm-base-url`         | Optional base URL (used for `openai` provider).                                    | false    | `https://api.openai.com/v1` |
+| `llm-summary-command`  | Optional command that reads `llm-prompt` on stdin and returns a summary on stdout. | false    | `""`                        |
+| `markdown-template`    | Markdown template with placeholders (`base_ref`, `changes`, …)                     | false    | Built-in template           |
+
+## Outputs
+
+| Output         | Description                                         |
+| -------------- | --------------------------------------------------- |
+| `changelog`    | Rendered Markdown changelog.                        |
+| `changes`      | Compiled Markdown list of changes.                  |
+| `llm-prompt`   | Prompt text that can be sent to any LLM provider.   |
+| `commit-count` | Number of commits included in the changelog output. |
diff --git a/actions/release/summarize-changelog/action.yml b/actions/release/summarize-changelog/action.yml
new file mode 100644
index 0000000..736923b
--- /dev/null
+++ b/actions/release/summarize-changelog/action.yml
@@ -0,0 +1,121 @@
+name: "Release - Summarize changelog"
+description: "Compile release changelog entries between two refs, with optional conventional commit grouping and LLM summary injection."
+author: hoverkraft
+branding:
+  icon: file-text
+  color: blue
+
+inputs:
+  base-ref:
+    description: "Base git ref (excluded from the range)."
+    required: true
+  head-ref:
+    description: "Head git ref (included in the range)."
+    required: true
+  conventional-commits:
+    description: "Whether to group commit messages by conventional commit type."
+    required: false
+    default: "true"
+  llm-summary:
+    description: "Optional summary generated by any LLM provider."
+    required: false
+    default: ""
+  llm-model:
+    description: "Optional model used to generate summary from `llm-prompt`."
+    required: false
+    default: ""
+  llm-provider:
+    description: "LLM provider used with LangChain (`openai`, `anthropic`, `google-genai`)."
+    required: false
+    default: "openai"
+  llm-api-key:
+    description: "Optional API key for the selected LLM provider."
+    required: false
+    default: ""
+  llm-base-url:
+    description: "Optional base URL (used for `openai` provider)."
+    required: false
+    default: "https://api.openai.com/v1"
+  llm-summary-command:
+    description: "Optional command used to generate the summary from `llm-prompt` (prompt is sent through stdin)."
+    required: false
+    default: ""
+  markdown-template:
+    description: |
+      Markdown template used to build the final changelog.
+      Supported placeholders:
+      - {{base_ref}}
+      - {{head_ref}}
+      - {{commit_count}}
+      - {{summary}}
+      - {{changes}}
+    required: false
+    default: |
+      ## Changelog
+
+      _Changes from `{{base_ref}}` to `{{head_ref}}`._
+
+      {{summary}}
+
+      {{changes}}
+
+outputs:
+  changelog:
+    description: "Rendered markdown changelog."
+    value: ${{ steps.summarize.outputs.changelog }}
+  changes:
+    description: "Compiled markdown list of changes."
+    value: ${{ steps.summarize.outputs.changes }}
+  llm-prompt:
+    description: "Prompt text that can be sent to any LLM provider."
+    value: ${{ steps.summarize.outputs.llm-prompt }}
+  commit-count:
+    description: "Number of commits included in the changelog."
+    value: ${{ steps.summarize.outputs.commit-count }}
+
+runs:
+  using: "composite"
+  steps:
+    - uses: hoverkraft-tech/ci-github-common/actions/checkout@f24ce3360a8abf9bf386a62ab13d0ae5de5f9d13 # 0.31.7
+      with:
+        fetch-depth: "0"
+
+    - shell: bash
+      if: ${{ !inputs.llm-summary && inputs.llm-model != '' }}
+      env:
+        LLM_PROVIDER: ${{ inputs.llm-provider }}
+      run: |
+        provider="${LLM_PROVIDER:-openai}"
+        case "$provider" in
+          openai)
+            npm install --no-save langchain@1.2.25 @langchain/openai@1.2.8
+            ;;
+          anthropic)
+            npm install --no-save langchain@1.2.25 @langchain/anthropic@1.3.18
+            ;;
+          google-genai)
+            npm install --no-save langchain@1.2.25 @langchain/google-genai@2.1.19
+            ;;
+          *)
+            echo "Unsupported llm-provider: $provider. Supported values: openai, anthropic, google-genai"
+            exit 1
+            ;;
+        esac
+
+    - id: summarize
+      uses: actions/github-script@ed597411d8f924073f98dfc5c65a23a2325f34cd # v8.0.0
+      env:
+        BASE_REF: ${{ inputs.base-ref }}
+        HEAD_REF: ${{ inputs.head-ref }}
+        CONVENTIONAL_COMMITS: ${{ inputs.conventional-commits }}
+        LLM_SUMMARY: ${{ inputs.llm-summary }}
+        LLM_MODEL: ${{ inputs.llm-model }}
+        LLM_PROVIDER: ${{ inputs.llm-provider }}
+        LLM_API_KEY: ${{ inputs.llm-api-key }}
+        LLM_BASE_URL: ${{ inputs.llm-base-url }}
+        LLM_SUMMARY_COMMAND: ${{ inputs.llm-summary-command }}
+        MARKDOWN_TEMPLATE: ${{ inputs.markdown-template }}
+      with:
+        script: |
+          const summarize = require(`${process.env.GITHUB_ACTION_PATH}/summarize.js`);
+          await summarize({ core });
diff --git a/actions/release/summarize-changelog/summarize.js b/actions/release/summarize-changelog/summarize.js
new file mode 100644
index 0000000..acc57b5
--- /dev/null
+++ b/actions/release/summarize-changelog/summarize.js
@@ -0,0 +1,145 @@
+const { execSync } = require("node:child_process");
+const { initChatModel } = require("langchain/chat_models/universal");
+
+module.exports = async ({ core }) => {
+  const baseRef = process.env.BASE_REF.trim();
+  const headRef = process.env.HEAD_REF.trim();
+  const useConventionalCommits =
+    process.env.CONVENTIONAL_COMMITS.trim() === "true";
+  const llmSummaryInput = process.env.LLM_SUMMARY.trim();
+  const llmModel = process.env.LLM_MODEL.trim();
+  const llmProvider = process.env.LLM_PROVIDER.trim() || "openai";
+  const llmApiKey = process.env.LLM_API_KEY.trim();
+  const llmBaseUrl = process.env.LLM_BASE_URL.trim();
+  const llmSummaryCommand = process.env.LLM_SUMMARY_COMMAND.trim();
+  const markdownTemplate = process.env.MARKDOWN_TEMPLATE;
+
+  if (!baseRef || !headRef) {
+    core.setFailed("Both base-ref and head-ref inputs are required.");
+    return;
+  }
+
+  const rawLog = execSync(
+    `git log --no-merges --pretty=format:%s ${baseRef}..${headRef}`,
+    {
+      encoding: "utf8",
+    },
+  ).trim();
+
+  const commits = rawLog
+    ? rawLog
+        .split("\n")
+        .map((line) => line.trim())
+        .filter(Boolean)
+    : [];
+  const groupedCommits = commits.reduce((acc, subject) => {
+    const match = subject.match(
+      /^(?<type>[a-z]+)(?:\([^)]+\))?(?:!)?:\s+(?<description>.+)$/i,
+    );
+    const rawType = match?.groups?.type?.toLowerCase();
+    const description = match?.groups?.description || subject;
+
+    let section = "Other changes";
+    if (useConventionalCommits && rawType) {
+      const typeToTitle = {
+        feat: "Features",
+        fix: "Bug fixes",
+        perf: "Performance",
+        refactor: "Refactors",
+        docs: "Documentation",
+        test: "Tests",
+        build: "Build",
+        ci: "CI",
+        chore: "Chores",
+        style: "Style",
+        revert: "Reverts",
+      };
+      section = typeToTitle[rawType] || "Other changes";
+    } else if (!useConventionalCommits) {
+      section = "Changes";
+    }
+
+    if (!acc.has(section)) {
+      acc.set(section, []);
+    }
+
+    acc.get(section).push(useConventionalCommits ? description : subject);
+    return acc;
+  }, new Map());
+
+  const changes = [...groupedCommits.entries()]
+    .map(
+      ([section, entries]) =>
+        `### ${section}\n${entries.map((entry) => `- ${entry}`).join("\n")}`,
+    )
+    .join("\n\n")
+    .trim();
+
+  const llmPrompt = [
+    "Summarize the following release changes as markdown:",
+    `Base ref: ${baseRef}`,
+    `Head ref: ${headRef}`,
+    "",
+    changes || "- No user-facing changes found.",
+  ].join("\n");
+
+  let llmSummary = llmSummaryInput;
+  if (!llmSummary && llmModel) {
+    if (!llmApiKey) {
+      core.setFailed("llm-api-key is required when llm-model is provided.");
+      return;
+    }
+
+    const llmConfig = {
+      model: llmModel,
+      modelProvider: llmProvider,
+    };
+    if (llmProvider === "openai") {
+      llmConfig.apiKey = llmApiKey;
+      if (llmBaseUrl) {
+        llmConfig.configuration = { baseURL: llmBaseUrl };
+      }
+    } else if (llmProvider === "anthropic") {
+      llmConfig.apiKey = llmApiKey;
+    } else if (llmProvider === "google-genai") {
+      llmConfig.apiKey = llmApiKey;
+    } else {
+      core.setFailed(
+        "Unsupported llm-provider. Supported values: openai, anthropic, google-genai.",
+      );
+      return;
+    }
+
+    const llm = await initChatModel(undefined, llmConfig);
+    const response = await llm.invoke([
+      {
+        role: "system",
+        content: "You generate concise release summaries in Markdown.",
+      },
+      { role: "user", content: llmPrompt },
+    ]);
+    llmSummary =
+      typeof response?.content === "string" ? response.content.trim() : "";
+  }
+
+  if (!llmSummary && llmSummaryCommand) {
+    llmSummary = execSync(llmSummaryCommand, {
+      encoding: "utf8",
+      input: llmPrompt,
+    }).trim();
+  }
+
+  const summary = llmSummary ? `### Summary\n${llmSummary}` : "";
+  const changelog = markdownTemplate
+    .replace(/\{\{base_ref\}\}/g, baseRef)
+    .replace(/\{\{head_ref\}\}/g, headRef)
+    .replace(/\{\{commit_count\}\}/g, `${commits.length}`)
+    .replace(/\{\{summary\}\}/g, summary)
+    .replace(/\{\{changes\}\}/g, changes)
+    .trim();
+
+  core.setOutput("commit-count", `${commits.length}`);
+  core.setOutput("changes", changes);
+  core.setOutput("llm-prompt", llmPrompt);
+  core.setOutput("changelog", changelog);
+};