diff --git a/AGENTS.md b/AGENTS.md
new file mode 100644
index 0000000..c59aff7
--- /dev/null
+++ b/AGENTS.md
@@ -0,0 +1,117 @@
+# AGENTS.md
+
+This file provides guidance to coding agents when working with code in this repository.
+
+## Project Overview
+
+Ruby gem providing text classification via two algorithms:
+- **Bayes** (`Classifier::Bayes`) - Naive Bayesian classification
+- **LSI** (`Classifier::LSI`) - Latent Semantic Indexing for semantic classification, clustering, and search
+
+## Common Commands
+
+```bash
+# Compile native C extension
+bundle exec rake compile
+
+# Run all tests (compiles first)
+bundle exec rake test
+
+# Run a single test file
+ruby -Ilib test/bayes/bayesian_test.rb
+ruby -Ilib test/lsi/lsi_test.rb
+
+# Run tests with pure Ruby (no native extension)
+NATIVE_VECTOR=true bundle exec rake test
+
+# Run benchmarks
+bundle exec rake benchmark
+bundle exec rake benchmark:compare
+
+# Interactive console
+bundle exec rake console
+
+# Generate documentation
+bundle exec rake doc
+```
+
+## Changelog
+
+`CHANGELOG.md` records every user-visible change. Add an entry in the same pull
+request that makes the change. A reviewer who finds a behavior change without an
+entry asks for one.
+
+Write entries for the person who installs the gem, not for the person who wrote
+the commit:
+
+- Describe the behavior that ships, not the path taken to reach it. A bug that a
+  reviewer found and the author fixed before release never existed for a user.
+- Put the entry under a `## X.Y.Z - YYYY-MM-DD` heading. Use the released
+  version number without a `v` prefix, and an ISO date.
+- Start a breaking change with `**Breaking:**` and say what a caller must now
+  do differently.
+- Name the public constant, method, or command that changed, and show the
+  command or call when an example makes it concrete.
+- Skip changes a user cannot observe: dependency bumps for development, lint
+  configuration, CI, and internal refactors.
+- Order the entries within a release by how much they matter to a user.
+
+Choose the version with semantic versioning: a breaking change is a major, a new
+feature is a minor, and a fix alone is a patch.
+
+## Release Workflow
+
+Update `lib/classifier/version.rb`, `CHANGELOG.md`, and `Gemfile.lock`. Run
+`bundle exec rake` and `bundle exec rubocop`. Then commit and merge to `master`.
+
+Publish by pushing an annotated version tag:
+
+```bash
+git tag -a v2.7.0 -m "Version 2.7.0"
+git push origin v2.7.0
+```
+
+The `Release` workflow runs CI, builds the gem, publishes it through RubyGems
+trusted publishing, and creates the GitHub release. Never run `gem push` from a
+workstation.
+
+## Architecture
+
+### Core Components
+
+**Bayesian Classifier** (`lib/classifier/bayes.rb`)
+- Train with `train(category, text)` or dynamic methods like `train_spam(text)`
+- Classify with `classify(text)` returning the best category
+- Uses log probabilities for numerical stability
+
+**LSI Classifier** (`lib/classifier/lsi.rb`)
+- Uses Singular Value Decomposition (SVD) for semantic analysis
+- Native C extension for 5-50x faster matrix operations; falls back to pure Ruby
+- Key operations: `add_item`, `classify`, `find_related`, `search`
+- `auto_rebuild` option controls automatic index rebuilding after changes
+
+**String Extensions** (`lib/classifier/extensions/word_hash.rb`)
+- `word_hash` / `clean_word_hash` - tokenize text to stemmed word frequencies
+- `CORPUS_SKIP_WORDS` - stopwords filtered during tokenization
+- Uses `fast-stemmer` gem for Porter stemming
+
+**Vector Extensions** (`lib/classifier/extensions/vector.rb`)
+- Pure Ruby SVD implementation (`Matrix#SV_decomp`) - used as fallback
+- Vector normalization and magnitude calculations
+
+### Native C Extension (`ext/classifier/`)
+
+LSI uses a native C extension for fast linear algebra operations:
+- `Classifier::Linalg::Vector` - Vector operations (alloc, normalize, dot product)
+- `Classifier::Linalg::Matrix` - Matrix operations (alloc, transpose, multiply)
+- Jacobi SVD implementation for singular value decomposition
+
+Check current backend: `Classifier::LSI.backend` returns `:native` or `:ruby`
+Force pure Ruby: `NATIVE_VECTOR=true bundle exec rake test`
+
+### Content Nodes (`lib/classifier/lsi/content_node.rb`)
+
+Internal data structure storing:
+- `word_hash` - term frequencies
+- `raw_vector` / `raw_norm` - initial vector representation
+- `lsi_vector` / `lsi_norm` - reduced dimensionality representation after SVD
diff --git a/CHANGELOG.md b/CHANGELOG.md
new file mode 100644
index 0000000..1b5f572
--- /dev/null
+++ b/CHANGELOG.md
@@ -0,0 +1,153 @@
+# Changelog
+
+## 2.7.0 - 2026-08-15
+
+- Add a `keywords` executable for TF-IDF keyword extraction. `keywords fit`
+  builds a vocabulary from files or standard input, `keywords extract` reads a
+  file, `keywords info` prints model statistics, and a bare text argument
+  transforms that text. The command prints `term:score` pairs and maps stems
+  back to whole words, so `keywords "Ruby is elegant"` prints
+  `elegant:0.61 ruby:0.61`. Options set the model path, the top-N terms, quiet
+  mode, `--min-df`, `--max-df`, and `--ngram`. Usage errors exit 2 and other
+  errors exit 1. The gem now installs two executables, `classifier` and
+  `keywords`.
+- Treat each line as a separate document during `keywords fit`. The document
+  count controls the inverse document frequency, so a file with many lines
+  contributes many documents.
+- Add `Classifier::Streaming::MultiIO`. It reads several IO objects or file
+  paths as one sequential stream. It opens and closes each path one at a time,
+  so a corpus larger than the file descriptor limit still fits.
+- Add `String#stem_to_word_hash`. It maps each stemmed root to the most
+  frequent original word in the text.
+- Add the `min_df` and `max_df` readers to `Classifier::TFIDF`.
+- Accept a `MultiIO` in `TFIDF#fit_from_stream` and `Streaming::LineReader`.
+
+## 2.6.0 - 2026-06-25
+
+- Add a `--search` flag to the command line tool to filter local models.
+- Add a model detail view to the command line tool.
+- Read model information in advance when the tool lists local models.
+
+## 2.5.0 - 2026-06-01
+
+- Accept keyword arguments in `train_from_stream`.
+
+## 2.4.0 - 2026-05-19
+
+- Make `min_word_length` configurable, so a caller can keep or drop short
+  words during tokenization.
+- Add a Claude Code plugin with a skill and slash commands.
+
+## 2.3.2 - 2026-01-01
+
+- Force UTF-8 encoding on the HTTP response body, so a remote model loads
+  under any locale.
+
+## 2.3.1 - 2026-01-01
+
+- Force UTF-8 encoding when the locale is not UTF-8. Model data and user input
+  no longer raise an encoding error.
+
+## 2.3.0 - 2025-12-31
+
+- Add a `classifier` executable with a model registry. The tool trains,
+  classifies, and manages saved models from the shell.
+- Fix the examples and the broken links in the README.
+- Add Dependabot for automated dependency updates.
+
+## 2.2.0 - 2025-12-29
+
+- **Breaking:** set `required_ruby_version` to `>= 3.1`. Older Ruby versions
+  are no longer supported.
+- Add a k-Nearest Neighbors classifier.
+- Add a Logistic Regression classifier.
+- Add a TF-IDF vectorizer.
+- Add streaming training and incremental SVD, so a corpus larger than memory
+  can train a model.
+- Add a hash-style API to add items to LSI.
+- Add keyword arguments to `Bayes#train` and `Bayes#untrain`.
+- Accept an array of categories in the classifier constructor.
+- Fix the sentence and paragraph splits in `Summary`.
+- Add property-based tests for the probabilistic invariants.
+
+## 2.1.0 - 2025-12-28
+
+- Replace the optional GSL dependency with a bundled C extension for LSI. The
+  extension has no external dependency and falls back to pure Ruby.
+- Add pluggable persistence backends through a storage API.
+- Add `save` and `load` methods for classifier persistence.
+- Add thread safety to the Bayes and LSI classifiers.
+- Expose the LSI tuning parameters, with validation and an introspection API.
+- Cache the expensive computations in the Bayes classifier.
+
+## 2.0.0 - 2025-12-27
+
+- **Breaking:** replace the fixed 0.1 constant in the Bayes classifier with
+  add-one (Laplace) smoothing, where
+  `P(word|category) = (count + 1) / (total + vocabulary_size)`. The smoothing
+  now scales with the vocabulary size and applies to seen and unseen words
+  alike. Classification scores change as a result.
+- Fix an LSI dimension mismatch in the pure Ruby SVD.
+- Fix the numerical stability of the SVD implementation.
+- Replace the separate RBS files with inline annotations.
+- Add a GitHub Actions workflow that publishes the gem on a version tag.
+- Add an LSI benchmark that compares GSL against pure Ruby.
+- Add RuboCop and SimpleCov.
+
+## 1.4.4 - 2024-07-31
+
+- Improve the scaling of the LSI content node.
+
+## 1.4.3 - 2024-07-31
+
+- Require `set` and use the explicit `::Set` namespace.
+- Refactor `prepare_category_name`.
+
+## 1.4.2 - 2024-07-31
+
+- Fix the word count when `remove_category` runs.
+- Add the `mutex_m` dependency and update the `fast-stemmer` version.
+
+## 1.4.1 - 2024-07-31
+
+- Add `remove_category` to the Bayes classifier.
+
+## 1.4.0 - 2024-07-31
+
+- Add `classify_with_confidence` to the LSI classifier.
+- Require `mathn` only for Ruby 2.5 and later, and add `cmath` for Ruby 2.7
+  and later.
+- Silence the warnings about an uninitialized `$GSL`, and correct the rb-gsl
+  URL hint.
+- Package the test files in the gem.
+- Add a Gemfile.lock.
+
+## 1.3.5 - 2018-04-17
+
+- Use Minitest for the test suite.
+- Add the `mathn` dependency, which Ruby 2.5.0 removed.
+- Make the gem installable through Bundler and a git remote.
+- Fix the gemspec and the unit tests.
+
+## 1.3.4 - 2013-12-31
+
+- Use a prior in the Bayes classifier.
+- Change the skip word list from an array to a set.
+- Reduce the number of regular expression matches during tokenization.
+- Stem only the words that the tokenizer keeps.
+- Default the word hash values to 0.
+- Add a gemspec, a README, and Travis CI.
+
+## 1.3.3 - 2010-07-06
+
+- Use fast-stemmer for the Porter stemmer.
+- Check `$GSL` before a call to `Matrix.diag`, so the code uses `GSL::Matrix`.
+
+## 1.3.2 - 2010-07-06
+
+- Fix the reported issue #1.
+
+## 1.3.1 and earlier
+
+Versions 1.0 through 1.3.1 reached RubyGems on 2009-07-25. The git history
+before 2010 is too sparse to attribute each change to one of these versions.
diff --git a/CLAUDE.md b/CLAUDE.md
index efec299..43c994c 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -1,77 +1 @@
-# CLAUDE.md
-
-This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
-
-## Project Overview
-
-Ruby gem providing text classification via two algorithms:
-- **Bayes** (`Classifier::Bayes`) - Naive Bayesian classification
-- **LSI** (`Classifier::LSI`) - Latent Semantic Indexing for semantic classification, clustering, and search
-
-## Common Commands
-
-```bash
-# Compile native C extension
-bundle exec rake compile
-
-# Run all tests (compiles first)
-bundle exec rake test
-
-# Run a single test file
-ruby -Ilib test/bayes/bayesian_test.rb
-ruby -Ilib test/lsi/lsi_test.rb
-
-# Run tests with pure Ruby (no native extension)
-NATIVE_VECTOR=true bundle exec rake test
-
-# Run benchmarks
-bundle exec rake benchmark
-bundle exec rake benchmark:compare
-
-# Interactive console
-bundle exec rake console
-
-# Generate documentation
-bundle exec rake doc
-```
-
-## Architecture
-
-### Core Components
-
-**Bayesian Classifier** (`lib/classifier/bayes.rb`)
-- Train with `train(category, text)` or dynamic methods like `train_spam(text)`
-- Classify with `classify(text)` returning the best category
-- Uses log probabilities for numerical stability
-
-**LSI Classifier** (`lib/classifier/lsi.rb`)
-- Uses Singular Value Decomposition (SVD) for semantic analysis
-- Native C extension for 5-50x faster matrix operations; falls back to pure Ruby
-- Key operations: `add_item`, `classify`, `find_related`, `search`
-- `auto_rebuild` option controls automatic index rebuilding after changes
-
-**String Extensions** (`lib/classifier/extensions/word_hash.rb`)
-- `word_hash` / `clean_word_hash` - tokenize text to stemmed word frequencies
-- `CORPUS_SKIP_WORDS` - stopwords filtered during tokenization
-- Uses `fast-stemmer` gem for Porter stemming
-
-**Vector Extensions** (`lib/classifier/extensions/vector.rb`)
-- Pure Ruby SVD implementation (`Matrix#SV_decomp`) - used as fallback
-- Vector normalization and magnitude calculations
-
-### Native C Extension (`ext/classifier/`)
-
-LSI uses a native C extension for fast linear algebra operations:
-- `Classifier::Linalg::Vector` - Vector operations (alloc, normalize, dot product)
-- `Classifier::Linalg::Matrix` - Matrix operations (alloc, transpose, multiply)
-- Jacobi SVD implementation for singular value decomposition
-
-Check current backend: `Classifier::LSI.backend` returns `:native` or `:ruby`
-Force pure Ruby: `NATIVE_VECTOR=true bundle exec rake test`
-
-### Content Nodes (`lib/classifier/lsi/content_node.rb`)
-
-Internal data structure storing:
-- `word_hash` - term frequencies
-- `raw_vector` / `raw_norm` - initial vector representation
-- `lsi_vector` / `lsi_norm` - reduced dimensionality representation after SVD
+@AGENTS.md
diff --git a/Gemfile.lock b/Gemfile.lock
index fb0a83c..2932c42 100644
--- a/Gemfile.lock
+++ b/Gemfile.lock
@@ -1,7 +1,7 @@
 PATH
   remote: .
   specs:
-    classifier (2.6.0)
+    classifier (2.7.0)
       fast-stemmer (~> 1.0)
       matrix
       mutex_m (~> 0.2)
diff --git a/classifier.gemspec b/classifier.gemspec
index 0a0a432..a8b6ee0 100644
--- a/classifier.gemspec
+++ b/classifier.gemspec
@@ -15,10 +15,11 @@ Gem::Specification.new do |s|
     'documentation_uri' => 'https://rubyclassifier.com/docs',
     'source_code_uri' => 'https://github.com/cardmagic/classifier',
     'bug_tracker_uri' => 'https://github.com/cardmagic/classifier/issues',
-    'changelog_uri' => 'https://github.com/cardmagic/classifier/releases'
+    'changelog_uri' => 'https://github.com/cardmagic/classifier/blob/master/CHANGELOG.md'
   }
   s.required_ruby_version = '>= 3.1'
-  s.files = Dir['{lib,sig,exe}/**/*.{rb,rbs}', 'ext/**/*.{c,h,rb}', 'exe/*', 'bin/*', 'LICENSE', '*.md', 'test/*']
+  s.files = Dir['{lib,sig,exe}/**/*.{rb,rbs}', 'ext/**/*.{c,h,rb}', 'exe/*', 'bin/*', 'LICENSE',
+                'README.md', 'CHANGELOG.md', 'test/*']
   s.bindir = 'exe'
   s.executables = %w[classifier keywords]
   s.extensions = ['ext/classifier/extconf.rb']
diff --git a/lib/classifier/version.rb b/lib/classifier/version.rb
index 861b19f..9499416 100644
--- a/lib/classifier/version.rb
+++ b/lib/classifier/version.rb
@@ -1,3 +1,3 @@
 module Classifier
-  VERSION = '2.6.0'.freeze
+  VERSION = '2.7.0'.freeze
 end