turnkit 0.2.4 → 0.2.5

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 75121664c1e081304931fbf125db92a9abc8b9062f920c7e33f7759b52ce51ec
-  data.tar.gz: ccabe905d199d955d281c936a019995a3bd9bc29c0fc009160ea924de4605835
+  metadata.gz: 271ce272a71a97aa2991a580f36205e4cef8e19466e2e480b0ac6f0f0225d51f
+  data.tar.gz: b9a0503f499d3eb850e7eece6f508b6fbc206d6398263f6005520b7ef716493b
 SHA512:
-  metadata.gz: ff0fa50aabb4c4b4fd9ea6f3ae78b62a4b020522a083f96605028dca2f4ca50a4fb6a9b98b36070e070d38a36b205ebf343823b520f5b0e5b4fe7a06b643cdce
-  data.tar.gz: beec35d2fc1f51cc6fe674d12d72e0ec1b44722bdcfab28019e9ab2d2ae313c684125989647e6d5d389f80b2df5f98dd33aa3c154e0af7da0885d2b8bec0221c
+  metadata.gz: f8772f25a95c44b2ba3d1a17a3e89d0ba142d862e798cee6daef9c54e04deaa3d8dee77deae48b5a77f7b6051b467a14c355aabf5115b1ce89832a27c87eb1b6
+  data.tar.gz: 9b12cccaa55c8d791168eca90655e3b9db89409b69fe59f8b45d23bef71aeec296c538696af44e484da9884dfde4ace67bbfd81d4a6647783f1f7f299ef0e485

data/CHANGELOG.md

@@ -1,5 +1,9 @@
 # Changelog
 
+## 0.2.5 - 2026-06-06
+
+- Add per-agent and per-turn provider thinking configuration.
+
 ## 0.2.4 - 2026-06-06
 
 - Add Anthropic prompt cache support for stable system prompt sections.

data/README.md

@@ -22,12 +22,21 @@ bundle install
 
 ## Quick Start
 
-Set a provider key:
+Set a provider key. TurnKit uses RubyLLM under the hood and defaults to Anthropic Claude:
 
 ```sh
 export ANTHROPIC_API_KEY=...
 ```
 
+| Provider | Env var | Example model |
+| --- | --- | --- |
+| Anthropic | `ANTHROPIC_API_KEY` | `claude-sonnet-4-5` |
+| OpenAI | `OPENAI_API_KEY` | `gpt-4.1-mini` |
+| Gemini | `GEMINI_API_KEY` | `gemini-2.5-flash` |
+
+> [!WARNING]
+> TurnKit defaults to `claude-sonnet-4-5`. If `ANTHROPIC_API_KEY` is unset or blank, set `TurnKit.default_model` to a provider you have configured.
+
 Create an agent:
 
 ```ruby
@@ -68,6 +77,52 @@ Set an OpenAI model:
 TurnKit.default_model = "gpt-4.1-mini"
 ```
 
+Use Gemini:
+
+```sh
+export GEMINI_API_KEY=...
+```
+
+Set a Gemini model:
+
+```ruby
+TurnKit.default_model = "gemini-2.5-flash"
+```
+
+### Thinking
+
+Enable provider reasoning or extended thinking per agent:
+
+```ruby
+agent = TurnKit::Agent.new(
+  name: "reasoner",
+  model: "claude-sonnet-4-5",
+  thinking: { budget: 4_000 }
+)
+```
+
+Use effort-based thinking for providers that support it:
+
+```ruby
+agent = TurnKit::Agent.new(
+  name: "reasoner",
+  model: "gemini-2.5-flash",
+  thinking: { effort: :high }
+)
+```
+
+Override or disable thinking for one turn:
+
+```ruby
+conversation = agent.conversation
+conversation.ask("Solve this carefully.", thinking: { budget: 8_000 })
+conversation.ask("Answer quickly.", thinking: nil)
+```
+
+TurnKit passes `thinking` to RubyLLM as `{ effort:, budget: }`. Anthropic requires `budget`; Gemini and OpenRouter can use `effort`, `budget`, or both depending on the model.
+
+When the provider reports reasoning usage, TurnKit records it as `thinking_tokens` and includes it in usage totals and cost calculation.
+
 ### Conversations
 
 Create a conversation:
@@ -130,6 +185,76 @@ turn = agent.conversation.ask("Save a short status report.")
 puts turn.output_text
 ```
 
+#### Defining application tools
+
+Tools are classes, not instances. Namespaced tools work fine, and the default tool name comes from the class name: `Assistant::Tools::WebSearch` becomes `web_search`.
+
+```ruby
+module Assistant
+  module Tools
+    class WebSearch < TurnKit::Tool
+      description "Search the web for current information."
+      usage_hint "Use when current external information is needed."
+
+      parameter :objective, :string, required: true
+      parameter :search_queries, :array, required: false
+
+      def call(objective:, search_queries: nil, context:)
+        ParallelClient.new.web_search(
+          objective: objective,
+          search_queries: search_queries
+        )
+      end
+    end
+  end
+end
+```
+
+Register tool classes on the agent:
+
+```ruby
+agent = TurnKit::Agent.new(
+  name: "researcher",
+  tools: [
+    Assistant::Tools::WebSearch,
+    Assistant::Tools::ReadWebPage
+  ]
+)
+```
+
+#### Tool context
+
+Every tool receives a `context:` object. Use it for logging, correlation, persistence, and domain scoping:
+
+```ruby
+def call(query:, context:)
+  context.turn       # The TurnKit::Turn being run
+  context.execution  # The TurnKit::ToolExecution for this tool call
+
+  { query: query }
+end
+```
+
+If your application already uses a `context:` keyword for something else, use `turnkit_context:` instead:
+
+```ruby
+def call(query:, turnkit_context:)
+  { turn_id: turnkit_context.turn.id, query: query }
+end
+```
+
+#### Tool return values
+
+Prefer returning a `Hash`. TurnKit serializes the normalized value as the tool result:
+
+| Return value | Stored tool result |
+| --- | --- |
+| `Hash` | Keys are stringified. |
+| `Array` | Wrapped as `{ "items" => [...] }`. |
+| Scalar | Wrapped as `{ "result" => value.to_s }`. |
+
+Avoid returning arbitrary objects unless you convert them to a plain Hash or Array first.
+
 ### Skills
 
 Load a skill:
@@ -260,7 +385,7 @@ Create a client:
 
 ```ruby
 class MyClient < TurnKit::Client
-  def chat(model:, messages:, tools:, instructions:, temperature: nil, metadata: nil)
+  def chat(model:, messages:, tools:, instructions:, temperature: nil, thinking: nil, metadata: nil)
     TurnKit::Result.new(
       text: "provider response",
       model: model,
@@ -295,6 +420,17 @@ Install Rails persistence:
 bin/rails generate turnkit:install
 ```
 
+The installer creates:
+
+- `config/initializers/turnkit.rb`
+- `app/models/turnkit/conversation.rb`
+- `app/models/turnkit/turn.rb`
+- `app/models/turnkit/message.rb`
+- `app/models/turnkit/tool_execution.rb`
+- a migration for TurnKit persistence
+
+The generated migration currently uses `ActiveRecord::Migration[7.1]`. In a newer Rails app, update that version if your app requires it, for example `ActiveRecord::Migration[8.1]`.
+
 Run migrations:
 
 ```sh
@@ -307,12 +443,88 @@ Configure Rails:
 TurnKit.store = TurnKit::ActiveRecordStore.new
 ```
 
+Suggested Rails file layout for your application AI code:
+
+```text
+app/models/assistant/
+  tools/
+    web_search.rb
+    read_web_page.rb
+  skills/
+  prompts/
+```
+
+If you prefer to keep AI infrastructure out of `app/models`, add an autoloaded directory such as:
+
+```text
+app/ai/
+  tools/
+  skills/
+  prompts/
+```
+
 Reconcile stale turns:
 
 ```ruby
 TurnKit.reconcile_stale!
 ```
 
+#### Debugging Rails persistence
+
+Inspect the latest persisted turn in a Rails console:
+
+```ruby
+turn = Turnkit::Turn.order(created_at: :desc).first
+turn.status
+turn.error
+turn.output_text
+```
+
+Check whether the model actually called tools:
+
+```ruby
+Turnkit::ToolExecution
+  .where(turn_uid: turn.uid)
+  .order(:created_at)
+  .map { |execution|
+    {
+      name: execution.tool_name,
+      status: execution.status,
+      arguments: execution.arguments,
+      result_keys: execution.result&.keys,
+      error: execution.error
+    }
+  }
+```
+
+#### Live smoke test
+
+Use a model whose provider key is configured, then run a real tool-using turn:
+
+```ruby
+TurnKit.default_model = "gpt-4.1-mini"
+
+agent = TurnKit::Agent.new(
+  name: "researcher",
+  instructions: "Use web_search, then read_web_page, before answering.",
+  tools: [
+    Assistant::Tools::WebSearch,
+    Assistant::Tools::ReadWebPage
+  ]
+)
+
+turn = agent.conversation.ask(
+  "Search for the TurnKit Ruby gem, read the first useful result, then summarize it."
+)
+
+puts turn.output_text
+
+pp Turnkit::ToolExecution
+  .where(turn_uid: turn.id)
+  .order(:created_at)
+  .pluck(:tool_name, :status, :error)
+```
+
 ## Options
 
 Configure defaults:
@@ -337,7 +549,8 @@ agent = TurnKit::Agent.new(
   model: "gpt-4.1-mini",
   max_iterations: 10,
   timeout: 60,
-  cost_limit: 0.25
+  cost_limit: 0.25,
+  thinking: { effort: :low }
 )
 ```
 
@@ -350,6 +563,7 @@ agent = TurnKit::Agent.new(
 | `timeout` | Limit seconds per root turn. |
 | `max_tool_executions` | Limit tool calls per root turn. |
 | `cost_limit` | Limit cost per root turn. |
+| `thinking` | Configure provider reasoning or extended thinking per agent. |
 | `cost_rates` | Override prices by model. |
 | `cost_calculator` | Override cost calculation. |
 | `prompt_cache` | Use provider prompt caching. |

data/lib/turnkit/adapters/ruby_llm.rb

@@ -3,7 +3,7 @@
 module TurnKit
   module Adapters
     class RubyLLM < Client
-      def chat(model:, messages:, tools:, instructions:, temperature: nil, metadata: nil)
+      def chat(model:, messages:, tools:, instructions:, temperature: nil, thinking: nil, metadata: nil)
         require "ruby_llm"
 
         configure_from_environment
@@ -11,6 +11,7 @@ module TurnKit
         chat = ::RubyLLM.chat(model: model)
         add_instructions(chat, instructions, model: model)
         chat.with_temperature(temperature) if temperature
+        apply_thinking(chat, thinking)
         Array(tools).each { |tool| chat.with_tool(ruby_llm_tool(tool)) }
         Array(messages).each { |message| add_message(chat, message) }
 
@@ -27,6 +28,11 @@ module TurnKit
           config.openrouter_api_key ||= ENV["OPENROUTER_API_KEY"]
         end
 
+        def apply_thinking(chat, thinking)
+          thinking = Agent.normalize_thinking(thinking)
+          chat.with_thinking(**thinking) if thinking
+        end
+
         def complete_without_tool_execution(chat)
           provider = chat.instance_variable_get(:@provider)
           provider.complete(
@@ -123,6 +129,7 @@ module TurnKit
             output_tokens: token_value(response, :output_tokens),
             cached_tokens: token_value(response, :cached_tokens),
             cache_write_tokens: token_value(response, :cache_creation_tokens),
+            thinking_tokens: thinking_token_value(response),
             cost: response_cost(response)
           )
           Result.new(
@@ -137,6 +144,10 @@ module TurnKit
           response.respond_to?(method) ? response.public_send(method).to_i : 0
         end
 
+        def thinking_token_value(response)
+          token_value(response, :thinking_tokens).nonzero? || token_value(response, :reasoning_tokens)
+        end
+
         def response_cost(response)
           return unless response.respond_to?(:cost)
 

data/lib/turnkit/agent.rb

@@ -4,11 +4,11 @@ module TurnKit
   class Agent
     attr_reader :name, :description, :model, :instructions, :tools, :skills, :available_skills, :sub_agents
     attr_reader :client, :store, :max_iterations, :timeout, :cost_limit, :max_depth, :max_tool_executions
-    attr_reader :prompt_sections, :system_prompt, :prompt_mode
+    attr_reader :prompt_sections, :system_prompt, :prompt_mode, :thinking
 
     def initialize(name:, description: "", model: nil, instructions: "", tools: [], skills: [], available_skills: [], sub_agents: [],
       system_prompt: nil, prompt_sections: nil, prompt_mode: nil, client: nil, store: nil,
-      max_iterations: nil, timeout: nil, cost_limit: nil, max_depth: nil, max_tool_executions: nil)
+      max_iterations: nil, timeout: nil, cost_limit: nil, max_depth: nil, max_tool_executions: nil, thinking: nil)
       @name = name.to_s
       @description = description.to_s
       @model = model
@@ -27,9 +27,25 @@ module TurnKit
       @cost_limit = cost_limit
       @max_depth = max_depth
       @max_tool_executions = max_tool_executions
+      @thinking = self.class.normalize_thinking(thinking)
       raise ArgumentError, "name is required" if @name.empty?
     end
 
+    def self.normalize_thinking(value)
+      return nil if value.nil?
+
+      attrs = value.respond_to?(:to_h) ? value.to_h : value
+      raise ArgumentError, "thinking must be a hash" unless attrs.is_a?(Hash)
+
+      attrs = attrs.transform_keys(&:to_sym)
+      unknown = attrs.keys - %i[effort budget]
+      raise ArgumentError, "unknown thinking attributes: #{unknown.join(", ")}" if unknown.any?
+      raise ArgumentError, "thinking requires :effort or :budget" if attrs[:effort].nil? && attrs[:budget].nil?
+      raise ArgumentError, "thinking budget must be an Integer" if attrs[:budget] && !attrs[:budget].is_a?(Integer)
+
+      attrs.slice(:effort, :budget).compact
+    end
+
     def conversation(model: nil, subject: nil, metadata: {})
       store = effective_store
       record = store.create_conversation(
@@ -53,6 +69,10 @@ module TurnKit
       model || TurnKit.default_model
     end
 
+    def effective_thinking
+      thinking
+    end
+
     def effective_client
       client || TurnKit.client
     end

data/lib/turnkit/client.rb

@@ -2,7 +2,7 @@
 
 module TurnKit
   class Client
-    def chat(model:, messages:, tools:, instructions:, temperature: nil, metadata: nil)
+    def chat(model:, messages:, tools:, instructions:, temperature: nil, thinking: nil, metadata: nil)
       raise NotImplementedError
     end
   end

data/lib/turnkit/conversation.rb

@@ -2,6 +2,8 @@
 
 module TurnKit
   class Conversation
+    THINKING_UNSET = Object.new.freeze
+
     attr_reader :agent, :id, :store, :model, :subject, :metadata
 
     def initialize(agent:, record:, store:, model:, subject: nil, metadata: {})
@@ -24,12 +26,15 @@ module TurnKit
       async ? turn : turn.run!
     end
 
-    def run!(trigger_message_id: nil, model: nil, budget: nil, parent_turn: nil, parent_tool_execution: nil, depth: 0, agent: self.agent)
-      build_turn(trigger_message_id: trigger_message_id, model: model, budget: budget, parent_turn: parent_turn, parent_tool_execution: parent_tool_execution, depth: depth, agent: agent).run!
+    def run!(trigger_message_id: nil, model: nil, budget: nil, parent_turn: nil, parent_tool_execution: nil, depth: 0, agent: self.agent, thinking: THINKING_UNSET)
+      build_turn(trigger_message_id: trigger_message_id, model: model, budget: budget, parent_turn: parent_turn, parent_tool_execution: parent_tool_execution, depth: depth, agent: agent, thinking: thinking).run!
     end
 
-    def build_turn(trigger_message_id: nil, model: nil, budget: nil, parent_turn: nil, parent_tool_execution: nil, depth: 0, agent: self.agent)
+    def build_turn(trigger_message_id: nil, model: nil, budget: nil, parent_turn: nil, parent_tool_execution: nil, depth: 0, agent: self.agent, thinking: THINKING_UNSET)
       snapshot = latest_message_sequence
+      effective_thinking = thinking.equal?(THINKING_UNSET) ? agent.effective_thinking : Agent.normalize_thinking(thinking)
+      options = { "trigger_message_id" => trigger_message_id }.compact
+      options["thinking"] = effective_thinking
       record = store.create_turn(
         "conversation_id" => id,
         "agent_name" => agent.name,
@@ -39,7 +44,7 @@ module TurnKit
         "context_message_sequence" => snapshot,
         "status" => "pending",
         "model" => model || self.model || agent.effective_model,
-        "options" => { "trigger_message_id" => trigger_message_id }.compact
+        "options" => options
       )
       Turn.new(agent: agent, conversation: self, record: record, store: store, budget: budget, depth: depth)
     end

data/lib/turnkit/cost.rb

@@ -2,10 +2,10 @@
 
 module TurnKit
   class Cost
-    COMPONENTS = %i[input output cache_read cache_write].freeze
+    COMPONENTS = %i[input output cache_read cache_write thinking].freeze
     PER_MILLION = 1_000_000.0
 
-    attr_reader :input, :output, :cache_read, :cache_write
+    attr_reader :input, :output, :cache_read, :cache_write, :thinking
 
     def self.aggregate(costs)
       costs = costs.compact
@@ -55,6 +55,7 @@ module TurnKit
         output: amount(usage.output_tokens, rates[:output] || rates[:output_per_million]),
         cache_read: amount(usage.cached_tokens, rates[:cache_read] || rates[:cached_input] || rates[:cache_read_input_per_million] || rates[:cached_input_per_million]),
         cache_write: amount(usage.cache_write_tokens, rates[:cache_write] || rates[:cache_creation] || rates[:cache_write_input_per_million] || rates[:cache_creation_input_per_million]),
+        thinking: amount(usage.thinking_tokens, rates[:thinking] || rates[:reasoning] || rates[:thinking_output] || rates[:reasoning_output] || rates[:thinking_output_per_million] || rates[:reasoning_output_per_million]),
         strict: true
       )
     end
@@ -70,7 +71,8 @@ module TurnKit
           input: usage.input_tokens,
           output: usage.output_tokens,
           cached: usage.cached_tokens,
-          cache_creation: usage.cache_write_tokens
+          cache_creation: usage.cache_write_tokens,
+          thinking: usage.thinking_tokens
         )
         from_hash(::RubyLLM::Cost.new(tokens: tokens, model: model_info).to_h)
       else
@@ -92,6 +94,7 @@ module TurnKit
         output: hash[:output],
         cache_read: hash[:cache_read] || hash[:cached_input],
         cache_write: hash[:cache_write] || hash[:cache_creation],
+        thinking: hash[:thinking] || hash[:reasoning] || hash[:thinking_output] || hash[:reasoning_output],
         total: hash[:total]
       )
     end
@@ -119,11 +122,12 @@ module TurnKit
       tokens.to_i * price.to_f / PER_MILLION
     end
 
-    def initialize(input: nil, output: nil, cache_read: nil, cache_write: nil, total: nil, strict: false)
+    def initialize(input: nil, output: nil, cache_read: nil, cache_write: nil, thinking: nil, total: nil, strict: false)
       @input = number(input)
       @output = number(output)
       @cache_read = number(cache_read)
       @cache_write = number(cache_write)
+      @thinking = number(thinking)
       @total = number(total)
       @strict = strict
     end
@@ -142,6 +146,7 @@ module TurnKit
         "output" => output,
         "cache_read" => cache_read,
         "cache_write" => cache_write,
+        "thinking" => thinking,
         "total" => total
       }.compact
     end

data/lib/turnkit/turn.rb

@@ -6,7 +6,7 @@ module TurnKit
 
     attr_reader :agent, :conversation, :store, :budget, :depth
     attr_reader :id, :conversation_id, :agent_name, :parent_turn_id, :parent_tool_execution_id
-    attr_reader :root_turn_id, :context_message_sequence, :model
+    attr_reader :root_turn_id, :context_message_sequence, :model, :thinking
     attr_reader :started_at
 
     def initialize(agent:, conversation:, record:, store:, budget: nil, depth: 0)
@@ -22,6 +22,7 @@ module TurnKit
       @root_turn_id = @record["root_turn_id"] || id
       @context_message_sequence = @record["context_message_sequence"].to_i
       @model = @record["model"] || agent.effective_model
+      @thinking = thinking_from_options
       @started_at = @record["started_at"]
       @budget = budget || agent.build_budget
       @depth = depth
@@ -40,6 +41,7 @@ module TurnKit
           messages: llm_messages,
           tools: agent.effective_tools,
           instructions: agent.system_prompt_for(turn: self, conversation: conversation),
+          thinking: thinking,
           metadata: { turn_id: id, conversation_id: conversation.id }
         )
         result_cost = Cost.from_usage(result.usage, model: result.model || model)
@@ -94,6 +96,7 @@ module TurnKit
 
     def reload
       @record = store.load_turn(id)
+      @thinking = thinking_from_options
       self
     end
 
@@ -106,6 +109,13 @@ module TurnKit
         MessageProjection.for(conversation.messages_for_turn(self))
       end
 
+      def thinking_from_options
+        options = (@record["options"] || {}).transform_keys(&:to_s)
+        return Agent.normalize_thinking(options["thinking"]) if options.key?("thinking")
+
+        agent.effective_thinking
+      end
+
       def persist_assistant_message(result)
         if result.tool_calls?
           conversation.append_message(
@@ -133,6 +143,7 @@ module TurnKit
           "output_tokens" => current["output_tokens"].to_i + usage.output_tokens,
           "cached_tokens" => current["cached_tokens"].to_i + usage.cached_tokens,
           "cache_write_tokens" => current["cache_write_tokens"].to_i + usage.cache_write_tokens,
+          "thinking_tokens" => current["thinking_tokens"].to_i + usage.thinking_tokens,
           "total_tokens" => current["total_tokens"].to_i + usage.total_tokens
         }
         totals["cost_details"] = aggregate_cost(current["cost_details"], cost).to_h if cost&.total

data/lib/turnkit/usage.rb

@@ -2,7 +2,7 @@
 
 module TurnKit
   class Usage
-    attr_reader :input_tokens, :output_tokens, :cached_tokens, :cache_write_tokens, :cost
+    attr_reader :input_tokens, :output_tokens, :cached_tokens, :cache_write_tokens, :thinking_tokens, :cost
 
     def self.aggregate(usages)
       usages = usages.compact
@@ -13,6 +13,7 @@ module TurnKit
         output_tokens: usages.sum(&:output_tokens),
         cached_tokens: usages.sum(&:cached_tokens),
         cache_write_tokens: usages.sum(&:cache_write_tokens),
+        thinking_tokens: usages.sum(&:thinking_tokens),
         cost: cost
       )
     end
@@ -29,20 +30,22 @@ module TurnKit
         output_tokens: attrs["output_tokens"],
         cached_tokens: attrs["cached_tokens"],
         cache_write_tokens: attrs["cache_write_tokens"],
+        thinking_tokens: attrs["thinking_tokens"] || attrs["reasoning_tokens"],
         cost: cost
       )
     end
 
-    def initialize(input_tokens: 0, output_tokens: 0, cached_tokens: 0, cache_write_tokens: 0, cost: nil)
+    def initialize(input_tokens: 0, output_tokens: 0, cached_tokens: 0, cache_write_tokens: 0, thinking_tokens: 0, cost: nil)
       @input_tokens = input_tokens.to_i
       @output_tokens = output_tokens.to_i
       @cached_tokens = cached_tokens.to_i
       @cache_write_tokens = cache_write_tokens.to_i
+      @thinking_tokens = thinking_tokens.to_i
       @cost = cost
     end
 
     def total_tokens
-      input_tokens + output_tokens + cached_tokens + cache_write_tokens
+      input_tokens + output_tokens + cached_tokens + cache_write_tokens + thinking_tokens
     end
 
     def to_h
@@ -51,6 +54,7 @@ module TurnKit
         "output_tokens" => output_tokens,
         "cached_tokens" => cached_tokens,
         "cache_write_tokens" => cache_write_tokens,
+        "thinking_tokens" => thinking_tokens,
         "total_tokens" => total_tokens,
         "cost" => cost
       }.compact

data/lib/turnkit/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module TurnKit
-  VERSION = "0.2.4"
+  VERSION = "0.2.5"
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: turnkit
 version: !ruby/object:Gem::Version
-  version: 0.2.4
+  version: 0.2.5
 platform: ruby
 authors:
 - Sam Couch