ollama_chat 0.0.127 → 0.0.128

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: a72761f0d0b34c0ba21859589410e914dda692d669a39629e4e6bb8f6e2857ea
-  data.tar.gz: 374fa95c62c5510e94c6a7624200a6193a15e9ed842a4b49b5364d43eeecac52
+  metadata.gz: a8c69119a9431fed33aea9fe4f2f23594eea70c570fb5f08b39b88daee8bd31a
+  data.tar.gz: efd37e4ba7c7971369bf4bd3f7ea95dd1a031f21f54f8a1b438a516db5b577c1
 SHA512:
-  metadata.gz: ea024b4607101da7037f0234970cb6e655c6d2b1af46fc1eac1b8dc901aebd5ef95a091d079653bfb58e86d381c4bb56173172f9fb0861e9d8a3424667bfdd32
-  data.tar.gz: dfbbe6cddbb4aac1f047cb39719ed58e8920e22a42775cb9f18a4d3731d0c53cc7c632733241f58f6cd6c132e5979b0a3117bf38e7e0095417310e839fb7421b
+  metadata.gz: cadf4af1255ea0232708e463b4c48f8b7497a556e4e14766f83c2291da8de7c9f77591da2a1d43678502d41bf414ed2e806ca46ef744d09ca85a33212474891a
+  data.tar.gz: 3e286668006435fe9fc00fdafbaa21d1d422ac99b00631838a69b2e334de30a50e3421a5172763827021370d3eb20eacdf16f7ca5f01a60a4baa4634d678b448

data/CHANGES.md

@@ -1,5 +1,94 @@
 # Changes
 
+## 2026-10-07 v0.0.128
+
+### Features
+
+*   Migrate `/clear` to `/conversation clean` chooser:
+    *   Removed `clean` method from `chat.rb` and the `/clear` command from
+        `commands.rb`.
+    *   Added `conversation_clean` to `conversation.rb` with an accumulating
+        `choose_with_state` chooser (options: `tools`, `images`, `thinking`,
+        `messages`, `history`, `links`), a `confirm?` gate, and per-field
+        dispatch.
+    *   Added `what:` keyword argument to `clean_messages` and
+        `clean_messages!` in `message_list.rb` supporting `:tools`, `:images`,
+        and `:thinking` selection.
+    *   Updated `/conversation clean` command handler to delegate to
+        `conversation_clean`.
+*   Add `OLLAMA_EMBEDDING_URL` for dedicated embedding host:
+    *   Add `OC::OLLAMA::EMBEDDING_URL` const_conf setting (optional,
+        `sensitive true`, decoded to `URI`, scheme-checked).
+    *   Add `connect_embedding_ollama` + public `embedding_ollama` accessor in
+        `chat.rb`; fallback to primary chat client when unset.
+    *   Propagate `ollama:` kwarg through `model_present?`,
+        `pull_model_from_remote`, `pull_model_unless_present`, `prepare_model`,
+        `use_model` in `model_handling.rb`.
+    *   Refactor `connect_message` to keyword-arg signature (`base_url:`,
+        `prefix:`) + new `connect_message_done` in `dialog.rb`; remove old
+        positional-arg version.
+    *   Remove `ensure connect_message` from `choose_model`.
+    *   `/reconnect` now calls both `connect_ollama` +
+        `connect_embedding_ollama`; `info_rag` shows embedding URL.
+    *   Fix `source_fetching.rb` Semantic splitter to route chunking
+        calls to `embedding_ollama` instead of the primary chat client.
+    *   Add `url` + `model` to log payloads in `collection_search.rb` and
+        `search_knowledge.rb`.
+    *   Update `README.md` with `OLLAMA_EMBEDDING_URL` entry.
+*   Add `/memory dump` and `/memory restore` command:
+    *   Add `/memory` command in `commands.rb` routing `dump` and `restore`
+        subcommands with an optional trailing `.jsonl` path.
+    *   Add `memory_dump` in `rag_handling.rb` with accumulating multi-select
+        chooser (`[ALL]`/`[DONE]`) writing JSONL via `JSONJSONLIO`.
+    *   Add `memory_restore` in `rag_handling.rb` grouping records by
+        `collection` field and re-embedding with original tags and
+        `source` association.
+    *   Extend `create_memory_collection` to accept either a bare persona stem
+        or a full `memory-*` collection name via `start_with?` guard.
+    *   Update `create_memory_collection` YARD prose to reflect dual-input
+        behavior.
+    *   Swap `miyu_pairing` → `personal_assistant` in `forget.rb`,
+        `memorize.rb`, and `memorize_spec.rb` to decouple docs from active
+        persona.
+
+### Refactoring
+
+*   Refactor all tools to use `tool_config`:
+    *   Replace 20 `config.tools.functions.<name>` call sites across 14 tool
+        files with the `tool_config` accessor from `Tools::Concern`.
+    *   Remove 12 dead `config = chat.config` locals that existed solely to
+        reach the tool's own config.
+    *   Drop `config` parameter from `eval_template` in `execute_grep.rb`,
+        `poll_for_image` in `generate_image.rb`, and `check_path` in
+        `run_tests.rb`.
+    *   Move `tool_config` and `allowed_paths` below `private` in `concern.rb`;
+        add YARD documentation to `tool_config`.
+
+### Documentation
+
+*   Document automatic syntax checking in `patch_file` description:
+    *   Added a `SYNTAX CHECK` note to the `OllamaChat::Tools::PatchFile`
+        description, informing users that the tool automatically runs
+        language-appropriate checks (e.g. `ruby -wc` for `.rb` files) and
+        reports errors in the result message.
+
+### Tests
+
+*   Removed `/clear` describe block from `commands_spec.rb`; replaced
+    `/conversation clean` specs with delegation test.
+*   Added 6 unit examples for `#conversation_clean` in `conversation_spec.rb`
+    covering cancel, deny, field-only, messages+fields, links-only, and
+    history-only paths.
+*   Fix `model_handling_spec.rb` call sites for new `ollama:` kwarg.
+*   Stub `connect_message_done` in `spec_helper.rb`.
+*   Add 5 routing specs in `commands_spec.rb` for `/memory` (with-path,
+    without-path, spaces-in-path).
+*   Add 7 specs in `rag_handling_spec.rb` for `create_memory_collection`,
+    `memory_dump`, and `memory_restore`.
+*   Update 3 `check_path` mocks in `run_tests_spec.rb` to match the new
+    single-argument signature; use `.expose` in `concern_spec.rb` for private
+    method access.
+
 ## 2026-10-06 v0.0.127
 
 ## Changes

data/README.md

@@ -61,6 +61,7 @@ The following environment variables can be used to configure behavior:
 #### Ollama Core
 - `OLLAMA_URL` - Base URL for Ollama server (default: `http://localhost:11434`)
 - `OLLAMA_HOST` - Base hostname for Ollama server (default: `localhost:11434`)
+- `OLLAMA_EMBEDDING_URL` - Optional base URL for a dedicated Ollama server used exclusively for embedding operations. When set, the embedding model runs on this host while the chat model stays on `OLLAMA_URL`. Falls back to `OLLAMA_URL` when unset.
 - `OLLAMA_SEARXNG_URL` - SearxNG search endpoint URL
 - `OLLAMA_CHAT_TTS_URL` - Base URL for the [audio.cpp](https://github.com/0xshug0/audio.cpp) TTS server (default: `http://localhost:8880`)
 

data/lib/ollama_chat/chat.rb

@@ -2,6 +2,7 @@ require 'tins'
 require 'tins/xt/string_version'
 require 'tins/xt/full'
 require 'json'
+require 'set'
 require 'term/ansicolor'
 require 'reline'
 require 'reverse_markdown'
@@ -116,6 +117,7 @@ class OllamaChat::Chat
     setup_session
     setup_state_selectors(config)
     connect_ollama
+    connect_embedding_ollama
     @documents = setup_documents
     models::Collection.sync(self)
     @cache                = setup_cache
@@ -321,6 +323,16 @@ class OllamaChat::Chat
     voice_handler.new(chat: self, voice: current_voice).speak(text, background:)
   end
 
+  # Returns the Ollama client to use for embedding operations.
+  #
+  # Falls back to the primary chat client when no dedicated embedding
+  # URL has been configured.
+  #
+  # @return [Ollama::Client] the embedding or chat Ollama client
+  def embedding_ollama
+    @embedding_ollama || ollama
+  end
+
   # Returns a human-readable string representation of the Chat object,
   # including the class name and the base URL.
   #
@@ -427,47 +439,6 @@ class OllamaChat::Chat
     content
   end
 
-  # The clean method clears various parts of the chat session based on the
-  # specified parameter.
-  #
-  # @param what [ String, nil ] the type of data to clear, defaults to
-  #   'messages' if nil
-  def clean(what)
-    persona_profile = nil
-    case what
-    when 'messages', nil
-      messages.clear
-      feedback("Cleared messages.", type: :info)
-    when 'links'
-      links.clear
-      feedback("Cleared links.", type: :info)
-    when 'history'
-      clear_history
-      feedback("Cleared history.", type: :info)
-    when 'tags'
-      @documents.clear
-      feedback("Cleared all tags.", type: :info)
-    when 'images'
-      messages.clear_images
-      feedback("Cleared all images.", type: :info)
-    when 'all'
-      if confirm?(
-          prompt: '🔔 Are you sure to clear messages and collection? (y/n) ',
-          yes: /\Ay/i
-        )
-      then
-        messages.clear
-        @documents.clear
-        links.clear
-        clear_history
-        feedback("Cleared messages and collection #{bold{collection}}.", type: :info)
-      else
-        feedback("Denied.", type: :denied)
-      end
-    end
-    persona_profile
-  end
-
   # The interact_with_user method manages the interactive loop for user input
   # and chat processing.
   # It handles reading user input, processing commands, managing messages, and
@@ -646,12 +617,13 @@ class OllamaChat::Chat
   # @raise [RuntimeError] if the connected Ollama server API version is less
   #   than 0.9.0
   def connect_ollama
+    connect_message(base_url:)
     @server_version = nil
     @ollama = Ollama::Client.new(
       connect_timeout: config.timeouts.connect_timeout?,
       read_timeout:    config.timeouts.read_timeout?,
       write_timeout:   config.timeouts.write_timeout?,
-      base_url:        base_url,
+      base_url:        ,
       debug:           ,
       user_agent:
     )
@@ -659,6 +631,7 @@ class OllamaChat::Chat
       raise 'require ollama API version 0.9.0 or higher'
     end
     log(:info, "Connection established", data: { base_url: })
+    connect_message_done
     @ollama
   end
 
@@ -673,10 +646,10 @@ class OllamaChat::Chat
     if embedding.on?
       @embedding_model         = config.embedding.model.name
       @embedding_model_options = Ollama::Options[config.embedding.model.options]
-      pull_model_unless_present(@embedding_model)
+      pull_model_unless_present(@embedding_model, ollama: embedding_ollama)
       collection = initial_collection
       @documents = Documentrix::Documents.new(
-        ollama:,
+        ollama:            embedding_ollama,
         model:             @embedding_model,
         model_options:     config.embedding.model.options,
         embedding_length:  config.embedding.model.embedding_length,
@@ -691,6 +664,33 @@ class OllamaChat::Chat
     end
   end
 
+  # Establishes a dedicated Ollama connection for embedding operations.
+  #
+  # When OC::OLLAMA::EMBEDDING_URL is configured, creates a separate
+  # Ollama::Client targeting that host so the embedding model can run
+  # on a different machine (e.g. a GPU box running audio.cpp alongside
+  # the embedding model) while the chat model stays on the primary
+  # Ollama server.
+  #
+  # @return [Ollama::Client, nil] the embedding client, or nil when no
+  #   dedicated embedding URL is configured
+  def connect_embedding_ollama
+    if base_url = OC::OLLAMA::EMBEDDING_URL?
+      connect_message(base_url:, prefix: 'embedding ')
+      @embedding_ollama = Ollama::Client.new(
+        connect_timeout: config.timeouts.connect_timeout?,
+        read_timeout:    config.timeouts.read_timeout?,
+        write_timeout:   config.timeouts.write_timeout?,
+        base_url:        ,
+        debug:           ,
+        user_agent:
+      )
+      log(:info, "Embedding connection established", data: { base_url: })
+      connect_message_done
+      @embedding_ollama
+    end
+  end
+
   # Adds documents from command line arguments to the document collection.
   #
   # Processes a list of document paths or URLs, handling both local files and

data/lib/ollama_chat/collection_search.rb

@@ -31,6 +31,7 @@ module OllamaChat::CollectionSearch
     log(
       :info, 'Embedding search request',
       data: {
+        url:        embedding_ollama.base_url,
         model:      config.embedding.model.to_h,
         collection: collection,
         query:      effective_query,
@@ -48,6 +49,7 @@ module OllamaChat::CollectionSearch
     log(
       :info, 'Embedding search response',
       data: {
+        url:          embedding_ollama.base_url,
         collection:   collection,
         query:        effective_query,
         result_count: records.size,

data/lib/ollama_chat/commands.rb

@@ -324,24 +324,6 @@ module OllamaChat::Commands
     :next
   end
 
-  command(
-    name: :clear,
-    regexp: %r(^/clear(?:\s+(messages|images|links|history|tags|all))?$),
-    complete: [ 'clear', %w[ messages images links history tags all ] ],
-    optional: true,
-    help: <<~EOT
-      🧹 Clear chat state (messages, images, links,
-         history, tags, all)
-    EOT
-  ) do |subcommand|
-    if result = clean(subcommand)
-      disable_content_parsing
-      result
-    else
-      :next
-    end
-  end
-
   command(
     name: :links,
     regexp: %r(^/links(?:\s+(clear))?$),
@@ -397,7 +379,7 @@ module OllamaChat::Commands
     help: <<~EOT
       💾 Manage conversation content:
          - save/load: Export/import as .json or .jsonl
-         - clean: Remove tool content, images, thinking
+         - clean: Interactively clean conversation content
          - compact: Summarize old messages, keep recent
          - compact summary: Show last compaction summary
          - summarize: Per-message narrative (-s sentence)
@@ -416,17 +398,7 @@ module OllamaChat::Commands
       load_conversation(path)
       repair_group_uuids
     when 'clean'
-      if confirm?(
-          prompt: '🔔 Clean tool content, images, and thinking from conversation? (y/n) ',
-          yes: /\Ay/i
-        )
-      then
-        messages.clean_messages!
-        session_sync
-        feedback("Conversation cleaned.", type: :info)
-      else
-        feedback("Denied.", type: :denied)
-      end
+      conversation_clean
     when 'compact summary'
       show_compaction_summary
     when 'compact'
@@ -699,6 +671,25 @@ module OllamaChat::Commands
     :next
   end
 
+  command(
+    name: :memory,
+    regexp: %r{^/memory(?:\s+(dump|restore))(?:\s+([^-].*\.jsonl?))?$},
+    complete: [ 'memory', %w[ dump restore ] ],
+    options: '[FILENAME]',
+    help: <<~EOT
+      💾 Manage persona memory (dump/restore)
+         optional path must end in .json or .jsonl
+    EOT
+  ) do |subcommand, path|
+    case subcommand
+    when 'dump'
+      memory_dump(path)
+    when 'restore'
+      memory_restore(path)
+    end
+    :next
+  end
+
   category :Persona
 
   command(
@@ -1042,9 +1033,8 @@ module OllamaChat::Commands
       🔌 Reconnect to Ollama server
     EOT
   ) do
-    feedback green { "Reconnecting to ollama #{base_url.to_s.inspect}…" }, type: :info
     connect_ollama
-    feedback green { "Done." }, type: :success
+    connect_embedding_ollama
     :next
   end
 

data/lib/ollama_chat/conversation.rb

@@ -63,4 +63,52 @@ module OllamaChat::Conversation
                "#{filename.to_s.inspect} failed.", type: :warn)
     end
   end
+
+  # Selectively cleans parts of the current conversation.
+  #
+  # Presents an accumulating chooser of cleanable targets:
+  # tool content, images, thinking, messages, history, links.
+  # The user picks one or more (or [ALL]), confirms, and the
+  # selected items are cleared.
+  def conversation_clean
+    options = %w[ tools images thinking messages history links ]
+
+    selected = Set.new
+    choose_with_state do
+      loop do
+        remaining = options - selected.to_a
+        entries   = (remaining.empty? ? [] : ['[ALL]'] + remaining) + ['[DONE]']
+        choice    = choose_entry(entries, prompt: 'What to clean from conversation? %s')
+        case choice
+        when nil, '[DONE]'
+          break
+        when '[ALL]'
+          selected.merge(remaining)
+          break
+        else
+          selected.add(choice)
+        end
+      end
+    end
+
+    return feedback('Cancelled, nothing selected.', type: :cancel) if selected.empty?
+
+    what = selected.map { |s| bold{s} }.to_a
+
+    unless confirm?(
+      prompt: "🔔 Clean #{what * ', '} from conversation? (y/n) ",
+      yes: /\Ay/i
+    )
+      return feedback('Denied.', type: :denied)
+    end
+
+    field_options = %w[ tools images thinking ].select { selected.include?(_1) }
+    messages.clear if selected.include?('messages')
+    messages.clean_messages!(what: field_options.map(&:to_sym)) unless field_options.empty?
+    clear_history if selected.include?('history')
+    links.clear if selected.include?('links')
+
+    session_sync
+    feedback("Cleaned #{what * ', '} from conversation.", type: :info)
+  end
 end

data/lib/ollama_chat/dialog.rb

@@ -118,14 +118,21 @@ module OllamaChat::Dialog
     files
   end
 
-  # The connect_message method displays a connection status message.
+  # Displays an info-level feedback message indicating that a connection
+  # to an Ollama server is being established.
   #
-  # @param model [String] the model name to connect to
-  # @param base_url [String] the base URL of the connection
-  def connect_message(model, base_url)
-    msg = "Connecting to #{model}@#{base_url} now…"
-    log(:info, msg, data: { model:, base_url: })
-    feedback green { msg }, type: :info
+  # @param base_url [URI, String] the target Ollama server URL
+  # @param prefix [String] an optional label prepended to "ollama" to
+  #   distinguish the connection type, e.g. `"embedding "` for a dedicated
+  #   embedding host (default: `""`)
+  def connect_message(base_url:, prefix: '')
+    feedback green { "Connecting to #{prefix}ollama #{base_url.to_s.inspect}…" }, type: :info
+  end
+
+  # Displays a success-level feedback message confirming that a
+  # connection was established. Counterpart to {#connect_message}.
+  def connect_message_done
+    feedback green { "Done." }, type: :success
   end
 
   # The change_voice method allows the user to select a voice from a list of

data/lib/ollama_chat/information.rb

@@ -191,6 +191,9 @@ module OllamaChat::Information
   # @param output [IO] the output stream to write the information to, defaults to STDOUT
   def info_rag(output: STDOUT)
     if embedding.on?
+      if base_url = OC::OLLAMA::EMBEDDING_URL?
+        output.puts "🔌 Connected to embedding ollama server on: #{base_url.to_s.inspect}"
+      end
       output.puts "🗄️ Current RAG model is #{bold{@embedding_model}}"
       if @embedding_model_options.present?
         output.puts "  Options: #{JSON.pretty_generate(@embedding_model_options).gsub(/(?<!\A)^/, '  ')}"

data/lib/ollama_chat/message_list.rb

@@ -407,11 +407,13 @@ class OllamaChat::MessageList
   # @param messages [Array<OllamaChat::Message>] the list of messages to clean
   # @return [Array<OllamaChat::Message>] a new array containing duplicated
   #   messages with stripped content
-  def clean_messages(messages: @messages)
+  def clean_messages(messages: @messages, what: %i[ tools images ])
+    what = Array(what).map(&:to_sym)
     messages.map do |message|
       message = message.dup
-      message.content = '' if message.tool?
-      message.images = nil
+      message.content = '' if what.include?(:tools) && message.tool?
+      message.images = nil if what.include?(:images)
+      message.thinking = nil if what.include?(:thinking)
       message
     end
   end
@@ -422,8 +424,8 @@ class OllamaChat::MessageList
   # @param messages [Array<OllamaChat::Message>] the messages to clean.
   #   Defaults to all current messages.
   # @return [OllamaChat::MessageList] self to allow for method chaining.
-  def clean_messages!(messages: @messages)
-    @messages = clean_messages(messages:)
+  def clean_messages!(messages: @messages, what: %i[ tools images ])
+    @messages = clean_messages(messages:, what:)
     self
   end
 

data/lib/ollama_chat/model_handling.rb

@@ -8,13 +8,13 @@
 # model retrieval when necessary.
 #
 # @example Checking if a model is present
-#   chat.model_present?('llama3.1')
+#   chat.model_present?('llama3.1', ollama:)
 #
 # @example Pulling a model from a remote server
-#   chat.pull_model_from_remote('mistral')
+#   chat.pull_model_from_remote('mistral', ollama:)
 #
 # @example Ensuring a model is available locally
-#   chat.pull_model_unless_present('phi3', {})
+#   chat.pull_model_unless_present('phi3', ollama:)
 module OllamaChat::ModelHandling
 
   # A simple data structure representing metadata about a model.
@@ -411,10 +411,11 @@ module OllamaChat::ModelHandling
   # available.
   #
   # @param model [ String ] the name of the Ollama model
+  # @param ollama [Ollama::Client] the client to query for model presence
   #
   # @return [ ModelMetadata, NilClass ] if the model is present,
   #   nil otherwise
-  def model_present?(model)
+  def model_present?(model, ollama:)
     ollama.show(model:) do |md|
       return ModelMetadata.new(
         name:         model,
@@ -432,7 +433,8 @@ module OllamaChat::ModelHandling
   # remote server if it is not found locally.
   #
   # @param model [ String ] the name of the model to be pulled
-  def pull_model_from_remote(model)
+  # @param ollama [Ollama::Client] the client to pull the model from
+  def pull_model_from_remote(model, ollama:)
     feedback("Model #{bold{model}} not found locally, attempting to pull it from remote now…", type: :info)
     ollama.pull(model:)
   end
@@ -444,16 +446,17 @@ module OllamaChat::ModelHandling
   # an UnknownModelError indicating the missing model name.
   #
   # @param model [String] the name of the model to ensure is present
+  # @param ollama [Ollama::Client] the client to query and pull from
   #
   # @return [ModelMetadata] the metadata for the available model
   # @raise [OllamaChat::UnknownModelError] if the model cannot be found after
   #   attempting to pull it from remote
-  def pull_model_unless_present(model)
-    if model_metadata = model_present?(model)
+  def pull_model_unless_present(model, ollama:)
+    if model_metadata = model_present?(model, ollama:)
       return model_metadata
     else
-      pull_model_from_remote(model)
-      if model_metadata = model_present?(model)
+      pull_model_from_remote(model, ollama:)
+      if model_metadata = model_present?(model, ollama:)
         return model_metadata
       end
       raise OllamaChat::UnknownModelError, "unknown model named #{@model.inspect}"
@@ -487,7 +490,7 @@ module OllamaChat::ModelHandling
   # @return [OllamaChat::ModelHandling::ModelMetadata] the metadata for the
   #   prepared model
   def prepare_model(model)
-    @model_metadata = pull_model_unless_present(model)
+    @model_metadata = pull_model_unless_present(model, ollama:)
     if think? && !@model_metadata.can?('thinking')
       think_mode.selected = 'disabled'
     end
@@ -584,7 +587,7 @@ module OllamaChat::ModelHandling
       @model = choose_model(model, config.model.name)
     end
 
-    if @model_metadata = model_present?(@model)
+    if @model_metadata = model_present?(@model, ollama:)
       session.update(current_model: @model)
     else
       session.update(current_model: nil)
@@ -655,7 +658,5 @@ module OllamaChat::ModelHandling
       else
         cli_model || current_model
       end
-  ensure
-    connect_message(model, ollama.base_url)
   end
 end

data/lib/ollama_chat/oc.rb

@@ -112,6 +112,13 @@ module OC
       check { value.scheme =~ /\Ahttps?\z/ }
     end
 
+    EMBEDDING_URL = set do
+      description 'Ollama base URL to connect to for embedding'
+      sensitive   true
+      decode { URI.parse(_1) if _1.present? }
+      check { value.nil? || value.scheme =~ /\Ahttps?\z/ }
+    end
+
     SEARXNG_URL = set do
       description 'URL for the SearXNG service for searches'
       default     'http://localhost:8088/search?q=%{query}&language=en&format=json'