ollama_agent 0.3.0 → 1.0.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: c3518f124795a2efb13b9e88abb2b7c90f935c918da3c7d30458ff2fff245f1b
-  data.tar.gz: 101d5e4bdbfbd5e6196e9744c94f1803c1ca32fea746b275cda8ad30783788d1
+  metadata.gz: 69d5c209f38109fb75cd33524df7cf31af937dbaa5b7b0c493973c6e7cc16c41
+  data.tar.gz: ef92101a3923debec6b193b1c9e56d1817e6154b44874daa0364a564d2336724
 SHA512:
-  metadata.gz: 1aad321bf07bb4ddea9daffeca46724e842408ac3f5d96997d77c3f81b30f10a3b1d28a4f8260bc14cde249cb63069e3fcc5aa1aecc982c0852a979733a2b1b3
-  data.tar.gz: 5d67531648baf16bf30b63e9d3726f14243d19d939f56875fb245254ad9538eded9ea4c1fe3e38cf1ded9d93b77f392f1ac4145ae0e351b654a531802696f230
+  metadata.gz: 6f71a58b11c1a16eb5310313c8e3036323684861bd594c4c1dfc114504ca9f4a4b6cec3cef0309476d9be607aafd09fbaf01d4c5f8cea77b789708b7f8874ffe
+  data.tar.gz: e3f9747d242a68858beca29fc643711057f04d36f0bbd57929001b4feae56d8b6acd45969abf1adc64b57522f571e1f3e254142896c82b398cc21f0f9bced0e1

data/CHANGELOG.md

@@ -1,8 +1,26 @@
 ## [Unreleased]
 
+## [1.0.0] - 2026-04-11
+
+First **stable** release under [Semantic Versioning](https://semver.org/): within the **`1.x`** series, the documented public Ruby API and CLI behavior are intended to remain **backwards compatible** except where called out in the changelog.
+
+### Changed
+
+- **CLI default:** `ollama_agent` with no subcommand runs **`ask`** with the **interactive TUI** when no query is given (same as `ollama_agent ask` with no query). Use **`ask -i` without `--tui`** for the plain line REPL, **`ask "…"`** for a one-shot task, or another subcommand (`self_review`, `sessions`, …) for other workflows.
+
+### Breaking changes (from 0.3.x)
+
+- **`tty-spinner` removed** from the gem; **`OllamaAgent::TUI`** no longer exposes `with_spinner`, `dismiss_progress_spinner`, `pause_progress_for_user_prompt`, or `resume_progress_after_user_prompt`. Code that called those methods or relied on the transitive `tty-spinner` dependency must be updated.
+- **`Agent#chat_assistant_message`** uses the streaming HTTP path **only** when something subscribes to **`on_token`** (not merely **`on_thinking`**). Integrations that registered only `on_thinking` expecting streamed chat must also subscribe to `on_token` or use another supported hook.
+
+### Fixed
+
+- TUI slash-command tab completion no longer raises `FrozenError` when editing a completed frozen candidate (e.g. `/help`).
+
 ## [0.3.0] - 2026-04-06
 
 ### Added
+
 - `ToolRuntime` — JSON plan loop for custom tools (`OllamaJsonPlanner`, registry, executor); see `docs/TOOL_RUNTIME.md`
 - Optional **ruby_mastery** context for `self_review` / `improve` (`OLLAMA_AGENT_RUBY_MASTERY`, `--no-ruby-mastery`)
 - `OllamaAgent::ModelEnv` — shared model name resolution from environment
@@ -11,14 +29,17 @@
 - External agents / argv expansion and related orchestration refinements
 
 ### Changed
+
 - `SearchBackend` finds `rg` / `grep` by scanning `PATH` (avoids relying on a `command` executable on trimmed `PATH`)
 
 ### Fixed
+
 - `SelfImprovement::Improver#run` accepts `max_tokens` and `context_summarize` from the CLI (Ruby 3 keyword compatibility)
 
 ## [0.2.0] - 2026-03-26
 
 ### Added
+
 - `write_file` tool — create or overwrite files (complements `edit_file` for surgical diffs)
 - `OllamaAgent::Tools.register` — extensible tool registry for library consumers
 - `Streaming::Hooks` — event bus (`on_token`, `on_tool_call`, `on_tool_result`, `on_complete`, `on_error`, `on_retry`)
@@ -32,10 +53,12 @@
 - `docs/ARCHITECTURE.md`, `docs/TOOLS.md`, `docs/SESSIONS.md`
 
 ### Changed
+
 - `READ_ONLY_TOOLS` now excludes both `edit_file` and `write_file`
 - `Agent` now exposes `#hooks` (`Streaming::Hooks`) and `#session_id`
 
 ### New environment variables
+
 - `OLLAMA_AGENT_STREAM`, `OLLAMA_AGENT_MAX_TOKENS`
 - `OLLAMA_AGENT_MAX_RETRIES`, `OLLAMA_AGENT_RETRY_BASE_DELAY`
 - `OLLAMA_AGENT_AUDIT`, `OLLAMA_AGENT_AUDIT_LOG_PATH`

data/README.md

@@ -1,6 +1,6 @@
 # ollama_agent
 
-Version: 0.1.0
+Version: 1.0.0
 
 Ruby gem that runs a **CLI coding agent** against a local [Ollama](https://ollama.com) model. It exposes tools to **list files**, **read files**, **search the tree** (ripgrep or grep), and **apply unified diffs** so the model can make small, reviewable edits.
 
@@ -40,6 +40,16 @@ bundle install
 
 ## Usage
 
+**Default:** run the gem with **no subcommand** to open the **interactive TUI** (same as `ask` with no query):
+
+```bash
+ollama_agent
+# or from this repo:
+bundle exec ruby exe/ollama_agent
+```
+
+Other entry points are **opt-in**: pass a **subcommand** (`self_review`, `sessions`, …) or **`ask` / `orchestrate`** with a **query** for a one-shot task, or flags for a plain line REPL (see below).
+
 From the project you want the agent to modify (set the working directory accordingly):
 
 ```bash
@@ -63,10 +73,11 @@ Long-running models (slow local inference):
 bundle exec ruby exe/ollama_agent ask --timeout 300 "Your task"
 ```
 
-Interactive REPL:
+**Plain line REPL** (no TUI boxes / markdown shell): use **`ask` (or `orchestrate`) with `-i` and without `--tui`**—for example when you omit the query you must opt out of the default TUI this way:
 
 ```bash
 bundle exec ruby exe/ollama_agent ask --interactive
+# same idea: explicit -i, no --tui
 ```
 
 Self-review modes (default project root is the **current working directory** unless you set `--root` or `OLLAMA_AGENT_ROOT`):
@@ -271,7 +282,7 @@ Repository **secrets** (Settings → Secrets and variables → Actions):
 Release steps:
 
 1. Bump `OllamaAgent::VERSION` in `lib/ollama_agent/version.rb` and commit to `main`.
-2. Tag: `git tag v0.1.0` (must match the version string) and `git push origin v0.1.0`.
+2. Tag: `git tag v1.0.0` (must match the version string) and `git push origin v1.0.0`.
 
 ## License
 

data/lib/ollama_agent/agent/agent_config.rb

@@ -7,7 +7,10 @@ module OllamaAgent
       attr_reader :model, :root, :confirm_patches, :http_timeout, :think, :read_only, :patch_policy,
                   :skill_paths, :skills_enabled, :skills_include, :skills_exclude, :external_skills_enabled,
                   :orchestrator, :confirm_delegation, :max_retries, :audit, :session_id, :resume,
-                  :max_tokens, :context_summarize, :stdin, :stdout
+                  :max_tokens, :context_summarize, :stdin, :stdout, :user_prompt,
+                  # v2 platform options
+                  :provider, :provider_name, :budget, :permissions, :policies,
+                  :memory_manager, :trace_logger, :approval_gate
 
       # @param confirm_delegation [Boolean, nil] nil means default true
       # rubocop:disable Metrics/MethodLength, Metrics/ParameterLists, Metrics/AbcSize -- value object mirrors Agent keywords
@@ -19,7 +22,11 @@ module OllamaAgent
                      max_retries: nil, audit: nil,
                      session_id: nil, resume: false,
                      max_tokens: nil, context_summarize: nil,
-                     stdin: $stdin, stdout: $stdout)
+                     stdin: $stdin, stdout: $stdout, user_prompt: nil,
+                     # v2 platform options (all optional — nil keeps existing behaviour)
+                     provider: nil, provider_name: nil, budget: nil,
+                     permissions: nil, policies: nil,
+                     memory_manager: nil, trace_logger: nil, approval_gate: nil)
         @model = model
         @root = root
         @confirm_patches = confirm_patches
@@ -42,6 +49,16 @@ module OllamaAgent
         @context_summarize = context_summarize
         @stdin = stdin
         @stdout = stdout
+        @user_prompt = user_prompt
+        # v2 platform options
+        @provider       = provider
+        @provider_name  = provider_name
+        @budget         = budget
+        @permissions    = permissions
+        @policies       = policies
+        @memory_manager = memory_manager
+        @trace_logger   = trace_logger
+        @approval_gate  = approval_gate
       end
       # rubocop:enable Metrics/MethodLength, Metrics/ParameterLists, Metrics/AbcSize
 

data/lib/ollama_agent/agent/client_wiring.rb

@@ -6,21 +6,16 @@ module OllamaAgent
     module ClientWiring
       private
 
-      # rubocop:disable Metrics/MethodLength
       def build_default_client
-        config = Ollama::Config.new
         @http_timeout_seconds = resolved_http_timeout_seconds
-        config.timeout = @http_timeout_seconds
-        OllamaConnection.apply_env_to_config(config)
-        ollama_client = Ollama::Client.new(config: config)
-        Resilience::RetryMiddleware.new(
-          client: ollama_client,
+        OllamaConnection.retry_wrapped_client(
+          timeout: @http_timeout_seconds,
           max_attempts: resolved_max_retries,
+          base_url: nil,
           hooks: @hooks,
           base_delay: resolved_retry_base_delay
         )
       end
-      # rubocop:enable Metrics/MethodLength
 
       def resolved_max_retries
         return @max_retries unless @max_retries.nil?

data/lib/ollama_agent/agent/session_wiring.rb

@@ -24,14 +24,48 @@ module OllamaAgent
 
       def append_tool_results(messages, tool_calls)
         tool_calls.each do |tool_call|
-          @hooks.emit(:on_tool_call, { name: tool_call.name, args: tool_call.arguments || {}, turn: current_turn })
-          result = execute_tool(tool_call.name, tool_call.arguments || {})
-          @hooks.emit(:on_tool_result, { name: tool_call.name, result: result.to_s, turn: current_turn })
+          name = tool_call.name
+          args = tool_call.arguments || {}
+
+          @hooks.emit(:on_tool_call, { name: name, args: args, turn: current_turn })
+          @loop_detector&.record!(name, args)
+
+          result = platform_guarded_tool_call(name, args)
+
+          @hooks.emit(:on_tool_result, { name: name, result: result.to_s, turn: current_turn })
+          @memory_manager&.record_tool_call(name, args, result)
           messages << tool_message(tool_call, result)
           save_message_to_session(messages.last)
         end
       end
 
+      # Run permission / policy guards before delegating to execute_tool.
+      def platform_guarded_tool_call(name, args)
+        ctx = build_tool_context
+
+        # Permission check
+        if @permissions && !@permissions.allowed?(name)
+          return "Permission denied: tool '#{name}' is not allowed under the current permission profile (#{@permissions.profile})."
+        end
+
+        # Policy check
+        if @policies
+          rejection = @policies.evaluate(name, args, ctx)
+          return rejection if rejection
+        end
+
+        execute_tool(name, args)
+      end
+
+      def build_tool_context
+        {
+          root:             @root,
+          read_only:        @read_only,
+          memory_manager:   @memory_manager,
+          shell_call_count: @shell_call_count || 0
+        }
+      end
+
       def save_message_to_session(msg)
         return unless @session_id
 

data/lib/ollama_agent/agent.rb

@@ -16,10 +16,14 @@ require_relative "context/manager"
 require_relative "session/store"
 require_relative "env_config"
 require_relative "model_env"
+require_relative "ollama_cloud_catalog"
 require_relative "agent/agent_config"
 require_relative "agent/client_wiring"
 require_relative "agent/prompt_wiring"
 require_relative "agent/session_wiring"
+require_relative "core/budget"
+require_relative "core/loop_detector"
+require_relative "core/trace_logger"
 
 module OllamaAgent
   # Runs a tool-calling loop against Ollama: read files, search, apply unified diffs.
@@ -34,7 +38,7 @@ module OllamaAgent
     MAX_TURNS = 64
     DEFAULT_HTTP_TIMEOUT = 120
 
-    attr_reader :client, :root, :hooks
+    attr_reader :client, :root, :hooks, :model
 
     # @param config [AgentConfig, nil] when set, keyword options are ignored (use {Runner} or build {AgentConfig}).
     # rubocop:disable Metrics/ParameterLists
@@ -48,7 +52,10 @@ module OllamaAgent
                    max_retries: nil, audit: nil,
                    session_id: nil, resume: false,
                    max_tokens: nil, context_summarize: nil,
-                   stdin: $stdin, stdout: $stdout)
+                   stdin: $stdin, stdout: $stdout,
+                   provider: nil, provider_name: nil, budget: nil,
+                   permissions: nil, policies: nil,
+                   memory_manager: nil, trace_logger: nil, approval_gate: nil, user_prompt: nil)
       cfg = config || AgentConfig.new(
         model: model, root: root, confirm_patches: confirm_patches, http_timeout: http_timeout, think: think,
         read_only: read_only, patch_policy: patch_policy,
@@ -56,10 +63,14 @@ module OllamaAgent
         skills_exclude: skills_exclude, external_skills_enabled: external_skills_enabled,
         orchestrator: orchestrator, confirm_delegation: confirm_delegation,
         max_retries: max_retries, audit: audit, session_id: session_id, resume: resume,
-        max_tokens: max_tokens, context_summarize: context_summarize, stdin: stdin, stdout: stdout
+        max_tokens: max_tokens, context_summarize: context_summarize, stdin: stdin, stdout: stdout,
+        provider: provider, provider_name: provider_name, budget: budget,
+        permissions: permissions, policies: policies,
+        memory_manager: memory_manager, trace_logger: trace_logger, approval_gate: approval_gate,
+        user_prompt: user_prompt
       )
       apply_agent_config(cfg)
-      @user_prompt = UserPrompt.new(stdin: cfg.stdin, stdout: cfg.stdout)
+      @user_prompt = cfg.user_prompt || UserPrompt.new(stdin: cfg.stdin, stdout: cfg.stdout)
       @context_manager = Context::Manager.new(max_tokens: @max_tokens, context_summarize: @context_summarize)
       @hooks = Streaming::Hooks.new
       attach_audit_logger if resolved_audit_enabled
@@ -74,6 +85,38 @@ module OllamaAgent
       execute_agent_turns(messages)
     end
 
+    # Switch the chat model for subsequent {#run} calls (same session, same client).
+    # Accepts Ollama local tags (e.g. +llama3.2+) or cloud catalog names (e.g. +glm-5.1+).
+    #
+    # @param name [String]
+    # @return [String] the normalized model id
+    # @raise [OllamaAgent::Error] when +name+ is blank
+    def assign_chat_model!(name)
+      n = name.to_s.strip
+      raise Error, "Model name cannot be empty" if n.empty?
+
+      @model = n
+      n
+    end
+
+    # Names from the local Ollama daemon (+/api/tags+ on your +base_url+). Not used by the REPL +/models+ command.
+    #
+    # @return [Array<String>]
+    def list_local_model_names
+      return [] unless @client.respond_to?(:list_model_names)
+
+      @client.list_model_names
+    rescue StandardError
+      []
+    end
+
+    # Cloud catalog from +https://ollama.com/api/tags+ (see {OllamaCloudCatalog}). REPL +/models+ uses this only.
+    #
+    # @return [Array<String>]
+    def list_cloud_model_names
+      OllamaCloudCatalog.list_model_names
+    end
+
     private
 
     # rubocop:disable Metrics/MethodLength, Metrics/AbcSize -- maps AgentConfig to ivars + resolved max turns
@@ -100,20 +143,50 @@ module OllamaAgent
       @context_summarize = cfg.context_summarize
       strict = EnvConfig.strict_env?
       @max_turns = EnvConfig.fetch_int("OLLAMA_AGENT_MAX_TURNS", MAX_TURNS, strict: strict)
+      # v2 platform subsystems (all optional; nil keeps legacy behaviour)
+      @budget        = cfg.budget        || Core::Budget.new(max_steps: @max_turns, max_tokens: @max_tokens)
+      @loop_detector = Core::LoopDetector.new
+      @trace_logger  = cfg.trace_logger
+      @memory_manager = cfg.memory_manager
+      @permissions   = cfg.permissions
+      @policies      = cfg.policies
+      @approval_gate = cfg.approval_gate
+      @provider      = cfg.provider
+      @provider_name = cfg.provider_name
     end
     # rubocop:enable Metrics/MethodLength, Metrics/AbcSize
 
     # rubocop:disable Metrics/MethodLength -- turn loop with early break
     def execute_agent_turns(messages)
       @current_turn = 0
+      @budget.reset!
+      @loop_detector.reset!
+      @trace_logger&.start_run(query: messages.last&.fetch(:content, nil))
+
       @max_turns.times do
         @current_turn += 1
-        trimmed = trimmed_messages_for_chat(messages)
-        message = chat_assistant_message(trimmed)
+        @budget.record_step!
+
+        if @budget.exceeded?
+          reason = @budget.exceeded_reason
+          warn "ollama_agent: budget exceeded — #{reason}"
+          @trace_logger&.budget_exceeded(reason: reason)
+          break
+        end
+
+        trimmed    = trimmed_messages_for_chat(messages)
+        message    = chat_assistant_message(trimmed)
         tool_calls = tool_calls_from(message)
         persist_assistant_turn(messages, message)
         break if tool_calls.empty?
 
+        if @loop_detector.loop_detected?
+          summary = @loop_detector.loop_summary
+          warn "ollama_agent: #{summary}"
+          @trace_logger&.loop_detected(summary: summary)
+          break
+        end
+
         append_tool_results(messages, tool_calls)
       end
       emit_turn_complete(messages)
@@ -206,6 +279,9 @@ module OllamaAgent
     end
 
     def announce_assistant_content(message)
+      @hooks.emit(:on_assistant_message, { message: message })
+      return if @hooks.subscribed?(:on_assistant_message)
+
       Console.puts_assistant_message(message)
     end
 

data/lib/ollama_agent/cli/repl.rb

@@ -0,0 +1,159 @@
+# frozen_string_literal: true
+
+require "fileutils"
+
+require_relative "repl_shared"
+
+module OllamaAgent
+  class CLI
+    # Interactive REPL for the agent.
+    #
+    # Features:
+    #   - Slash commands (/help, /model, /models, /session, /memory, /status, /clear, /config, /provider, /index)
+    #   - Readline history with persistent file
+    #   - Multi-line input (end with blank line)
+    #   - Session resume inside REPL
+    #   - Token budget and loop-detector status display
+    #   - Graceful Ctrl-C and Ctrl-D handling
+    # rubocop:disable Metrics/ClassLength -- readline wiring + banner + history
+    class Repl
+      include ReplShared
+
+      HISTORY_FILE = File.join(Dir.home, ".config", "ollama_agent", "repl_history")
+      MAX_HISTORY  = 500
+
+      PROMPT = "\e[32mollama\e[0m \e[90m›\e[0m "
+
+      def initialize(agent:, memory: nil, budget: nil, stdout: $stdout, stderr: $stderr)
+        @agent   = agent
+        @memory  = memory
+        @budget  = budget
+        @stdout  = stdout
+        @stderr  = stderr
+        @running = false
+      end
+
+      # rubocop:disable Metrics/MethodLength -- readline loop + history ensure
+      def start
+        @running = true
+        setup_readline
+        print_banner
+
+        loop do
+          input = read_line
+          break if input.nil?
+
+          line = input.chomp.strip
+          next if line.empty?
+
+          break if %w[/exit exit].include?(line)
+
+          if line.start_with?("/")
+            handle_slash(line)
+          else
+            run_query(line)
+          end
+        end
+
+        @stdout.puts "\n\e[90mGoodbye.\e[0m"
+      ensure
+        save_history
+      end
+      # rubocop:enable Metrics/MethodLength
+
+      private
+
+      def setup_readline
+        return unless readline_available?
+
+        load_history
+        Readline.completion_proc = slash_completer
+        configure_readline_slash_completion
+      end
+
+      # Shows command hints: type `/` then Tab (lists matches) or Tab again to cycle.
+      # With GNU Readline, `autocompletion` can show inline candidates while typing.
+      def configure_readline_slash_completion
+        Readline.completion_append_character = "" if Readline.respond_to?(:completion_append_character=)
+        Readline.autocompletion = true if Readline.respond_to?(:autocompletion=)
+      end
+
+      def slash_completer
+        proc do |input|
+          slash_completer_candidates.select { |cmd| cmd.start_with?(input.to_s) }
+        end
+      end
+
+      def read_line
+        if readline_available?
+          Readline.readline(PROMPT, true)
+        else
+          @stdout.print PROMPT
+          @stdout.flush
+          $stdin.gets
+        end
+      rescue Interrupt
+        @stdout.puts ""
+        ""
+      end
+
+      def readline_available?
+        return @readline_available unless @readline_available.nil?
+
+        @readline_available = (require "readline") && true
+      rescue LoadError
+        @readline_available = false
+      end
+
+      def run_query(query)
+        @agent.run(query)
+      rescue OllamaAgent::Error => e
+        @stderr.puts "\e[31mError: #{e.message}\e[0m"
+      rescue StandardError => e
+        @stderr.puts "\e[31m#{e.class}: #{e.message}\e[0m"
+        @stderr.puts e.backtrace.first(5).join("\n") if ENV["OLLAMA_AGENT_DEBUG"] == "1"
+      end
+
+      # rubocop:disable Metrics/MethodLength -- ASCII banner lines
+      def print_banner
+        @stdout.puts "\e[1m\e[34m"
+        @stdout.puts "  ██████╗ ██╗     ██╗      █████╗ ███╗   ███╗ █████╗ "
+        @stdout.puts "  ██╔═══██╗██║     ██║     ██╔══██╗████╗ ████║██╔══██╗"
+        @stdout.puts "  ██║   ██║██║     ██║     ███████║██╔████╔██║███████║"
+        @stdout.puts "  ██║   ██║██║     ██║     ██╔══██║██║╚██╔╝██║██╔══██║"
+        @stdout.puts "  ╚██████╔╝███████╗███████╗██║  ██║██║ ╚═╝ ██║██║  ██║"
+        @stdout.puts "   ╚═════╝ ╚══════╝╚══════╝╚═╝  ╚═╝╚═╝     ╚═╝╚═╝  ╚═╝"
+        @stdout.puts "\e[0m"
+        @stdout.puts "  Universal AI operator runtime  •  type \e[33m/help\e[0m for commands"
+        print_slash_completion_hint if readline_available?
+        @stdout.puts ""
+      end
+      # rubocop:enable Metrics/MethodLength
+
+      def print_slash_completion_hint
+        @stdout.puts "  \e[33m/\e[0m then \e[33mTab\e[0m — slash-command hints; Tab again to list all matches"
+      end
+
+      def load_history
+        return unless File.exist?(HISTORY_FILE)
+
+        File.readlines(HISTORY_FILE, chomp: true).last(MAX_HISTORY).each do |line|
+          Readline::HISTORY.push(line)
+        end
+      rescue StandardError
+        nil
+      end
+
+      def save_history
+        return unless readline_available?
+
+        dir = File.dirname(HISTORY_FILE)
+        FileUtils.mkdir_p(dir)
+        File.write(HISTORY_FILE, Readline::HISTORY.to_a.last(MAX_HISTORY).join("\n"))
+      rescue StandardError
+        nil
+      end
+    end
+    # rubocop:enable Metrics/ClassLength
+  end
+end