openclacky 1.1.3 → 1.1.4

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: ca451cdde2e1cc068bda60f6b56f3d534a0187a7fad0d0b42308ad391c028bdf
-  data.tar.gz: e828f029d229e20e4e41edbc6773c2f6e9d0bf66aae3ad5b088f563fa8bf4c51
+  metadata.gz: 4107ac9d894f9af9647f19152855f13854d4cfebd8013d2d22057ca98de87274
+  data.tar.gz: 3c86e0b865cccc27f8c785f39c49384744ad4e6cce6cdf257ca0485f1d62c260
 SHA512:
-  metadata.gz: 32ee351b327f14a64e66fa6d27926e99c148ee69965a41360528e8e68d69d5ea40e2377cb94f851d4f00bb1d445fed8e35474c35c620cdfe4cd2ae2a924af186
-  data.tar.gz: 8a08b30877664568e264f823fd328cab521dc7d3362c10dfb30b50aac76d951bb3c753df23f2d47f4ed631128f490810497c5166378d43c688bf814c81040867
+  metadata.gz: e6d66fd62f6434c05420f5f639a412fb953e49634c3f0740f155a2d20b70d91acb12a9844a61330f0761d88ff2c284a1d071f0590b3e53553bd11d57a035669f
+  data.tar.gz: c64f138d54a5397ba0a49b6878b79576f12b6fc2dec05f7e4b24b5eaf3490060743a0e3a20da17b6cf47476a85eaa093742b616b5be9ab65859db6e8db035404

data/CHANGELOG.md

@@ -5,6 +5,22 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [1.1.4] - 2026-05-22
+
+### Added
+- Thinking level control for AI models, configurable per provider
+- Free brand customization support
+- Syntax highlighting and per-block copy button for code blocks in Web UI (#152)
+- Font size setting (small/medium/large) with proportional UI scaling in Web UI (#147)
+- Chinese README documentation (#148)
+
+### Improved
+- Unify POST /api/file-action with download support for remote deployments (#153)
+- Hover interaction polish across Web UI
+
+### Fixed
+- Upgrade hot reload no longer leaves stale process (#143)
+
 ## [1.1.3] - 2026-05-20
 
 ### Added

data/README.md

@@ -6,6 +6,10 @@
 [![Downloads](https://img.shields.io/gem/dt/openclacky?label=downloads&style=flat-square&color=brightgreen)](https://rubygems.org/gems/openclacky)
 [![License](https://img.shields.io/badge/license-MIT-lightgrey?style=flat-square)](LICENSE.txt)
 
+<p align="center">
+  <a href="README.md">English</a> · <a href="README_CN.md">简体中文</a>
+</p>
+
 **The most Token-efficient open-source AI Agent.**
 
 OpenClacky matches Claude Code on capability at comparable cost, and saves significantly against other open-source agents (~50% vs OpenClaw, ~3× cheaper than Hermes). 100% open source (MIT), BYOK with any OpenAI-compatible model, built on two years of Agentic R&D and harness engineering.

data/README_CN.md

@@ -0,0 +1,198 @@
+# OpenClacky
+
+[![Build](https://img.shields.io/github/actions/workflow/status/clacky-ai/openclacky/main.yml?label=build&style=flat-square)](https://github.com/clacky-ai/openclacky/actions)
+[![Release](https://img.shields.io/gem/v/openclacky?label=release&style=flat-square&color=blue)](https://rubygems.org/gems/openclacky)
+[![Ruby](https://img.shields.io/badge/ruby-%3E%3D%203.1.0-red?style=flat-square)](https://www.ruby-lang.org)
+[![Downloads](https://img.shields.io/gem/dt/openclacky?label=downloads&style=flat-square&color=brightgreen)](https://rubygems.org/gems/openclacky)
+[![License](https://img.shields.io/badge/license-MIT-lightgrey?style=flat-square)](LICENSE.txt)
+
+**最省 Token 的开源 AI Agent。**
+
+OpenClacky 在任务能力上对齐 Claude Code，成本相当，同时相比其他开源 Agent 有显著优势（约节省 50% vs OpenClaw，约便宜 3× vs Hermes）。100% 开源（MIT），支持 BYOK 接入任意 OpenAI 兼容模型，背后是两年 Agentic 研发与 Harness 工程积累。
+
+> 官网：https://www.openclacky.com/ · 投资方：**奇绩创坛 · 真格基金 · 红杉中国 · 高瓴资本**
+
+## 为什么选 OpenClacky？
+
+同一个任务，你要花多少钱？在可比的 Agent 工作负载下，OpenClacky 相比主流方案节省了大量 Token 费用。
+
+| Agent | 相对成本 | 备注 |
+|---|---|---|
+| **OpenClacky** | **~0.8×** | 16 个工具 · 近 100% 缓存命中 · 子 Agent 路由 |
+| Claude Code | 1.0×（基准） | 世界级 Harness，闭源订阅制 |
+| OpenClaw | ~1.5× | 能力对标的 Harness Agent |
+| Hermes | ~3× | 52 个内置工具，Schema 体积膨胀 ~3–4× |
+
+*数据为内部常见 Agent 任务均值，以 Claude Code 为基准。完整基准测试报告将在 GitHub 发布。*
+
+## 功能对比
+
+核心 Agent 能力各家大致对齐，真正的差异在于**成本、开放性、Skill 进化能力和集成支持**。
+
+| 功能 | Claude Code | OpenClaw | Hermes | **OpenClacky** |
+|---|:---:|:---:|:---:|:---:|
+| Token 成本 | 1.0× | ~1.5× | ~3× | **~0.8×** |
+| 开源 | ❌ 闭源 | ✅ 开源 | ✅ 开源 | ✅ MIT |
+| BYOK / 自由选模型 | ❌ 仅限 Anthropic | ✅ | ✅ | ✅ |
+| Skill 自我进化 | ❌ | ❌ | ✅ | ✅ |
+| IM 集成（飞书/企微/微信/Discord/Telegram） | ❌ | ✅ | ✅ | ✅ |
+
+## 成本是怎么降下来的
+
+不是靠裁剪功能——而是在每一层都做了正确的取舍，效果叠加。
+
+### 1. 极高缓存命中率
+Session 不重启、双缓存标记、**先插入再压缩**——System Prompt 从不被修改，压缩后仍能复用缓存。**实测缓存命中率：接近 100%。**
+
+### 2. 极简工具集
+仅 **16 个核心工具**。扩展能力通过一个 `invoke_skill` 元工具交给 Skill 生态承载。工具数量不是指标——任务完成率才是。
+
+| OpenClacky | Claude Code | OpenClaw | Hermes |
+|:--:|:--:|:--:|:--:|
+| **16** | 40+ | 23 | 52 |
+
+### 3. 空闲时自动压缩
+去开个会、倒杯咖啡——Agent 在后台压缩长上下文并预热缓存。你回来发第一条消息就能直接命中缓存。**冷启动首 Token 成本降低 50%+。**
+
+### 4. BYOK——你选模型，你定成本
+任意 OpenAI 兼容 API，即插即用。官方直连、聚合路由、兼容中转——100% 由你决定。代码用 Claude，子任务自动路由到 DeepSeek，再省一截。
+
+背后是 **2 年 · 3 代 Agentic 架构 · 6 个核心 Harness 工程决策**的积累。
+
+## Skills——Agent 的灵魂
+
+- **`/` 唤起** — 即时浏览、模糊搜索、直接调用。数百个 Skill 触手可及。
+- **用自然语言创建 Skill** — 描述你想要的，Agent 自动起草 `SKILL.md`、拆解步骤、跑验证。无需写代码。
+- **自我进化** — 每次运行后，Agent 根据执行上下文和结果更新 Skill。下次调用更稳定、更精准。
+- **开放兼容** — 支持 Claude Skills / Markdown Pack / 自定义格式。
+- **可变现** — 打磨好的 Skill 可打包出售，支持加密分发、License 管理、创作者自定价。
+
+## 安装
+
+### 桌面安装器（推荐）
+
+双击安装，环境、依赖、Skill 全部自动配置好。
+
+- **macOS** — [下载 `.dmg`](https://oss.1024code.com/openclacky-installer/official/openclacky-installer.dmg)（Apple Silicon / Intel）
+- **Windows** — [下载 `.exe`](https://oss.1024code.com/openclacky-installer/official/openclacky-installer.exe)（Windows 10 2004+ / Windows 11）
+
+更多选项：https://www.openclacky.com/
+
+### 命令行安装
+
+一键安装（Mac/Ubuntu）：
+
+```bash
+/bin/bash -c "$(curl -sSL https://raw.githubusercontent.com/clacky-ai/openclacky/main/scripts/install.sh)"
+```
+
+Windows：
+
+```bash
+powershell -c "& ([scriptblock]::Create((irm 'https://raw.githubusercontent.com/clacky-ai/openclacky/main/scripts/install.ps1')))"
+```
+
+或使用 Ruby（3.x/4.x）：
+
+**环境要求：** Ruby >= 3.1.0
+
+```bash
+gem install openclacky
+```
+
+详见：https://www.openclacky.com/docs/installation
+
+## 快速开始
+
+### 终端（CLI）
+
+```bash
+openclacky            # 在当前目录启动交互式 Agent
+```
+
+### Web UI
+
+```bash
+openclacky server     # 默认地址：http://localhost:7070
+```
+
+打开 **http://localhost:7070**，享受完整的聊天界面，支持多 Session 并行——同时跑编码、文案、研究等多个任务。
+
+选项：
+
+```bash
+openclacky server --port 8080        # 自定义端口
+openclacky server --host 0.0.0.0     # 监听所有接口（支持远程访问）
+```
+
+## 配置
+
+```bash
+$ openclacky
+> /config
+```
+
+设置你的 **API Key**、**模型**和 **Base URL**（任意 OpenAI 兼容提供商）。
+
+开箱即支持：**Claude (Anthropic) · GPT (OpenAI) · DeepSeek · Kimi (Moonshot) · MiniMax · OpenRouter**，或任意自定义端点。
+
+## 代码开发场景
+
+OpenClacky 是一款通用 AI 编程助手——搭建全栈应用脚手架、添加功能，或快速探索陌生代码库：
+
+```bash
+$ openclacky
+> /new my-app        # 创建新项目脚手架
+> 添加邮箱密码登录功能
+> 支付模块是怎么实现的？
+```
+
+## Star 历史
+
+<a href="https://www.star-history.com/?repos=clacky-ai%2Fopenclacky&type=date&legend=top-left">
+ <picture>
+   <source media="(prefers-color-scheme: dark)" srcset="https://api.star-history.com/chart?repos=clacky-ai/openclacky&type=date&theme=dark&legend=top-left" />
+   <source media="(prefers-color-scheme: light)" srcset="https://api.star-history.com/chart?repos=clacky-ai/openclacky&type=date&legend=top-left" />
+   <img alt="Star History Chart" src="https://api.star-history.com/chart?repos=clacky-ai/openclacky&type=date&legend=top-left" />
+ </picture>
+</a>
+
+## 进阶——创作者计划
+
+已有深度用户将自己的工作流打磨成垂直 AI 专家在 OpenClacky 上发布——支持加密分发、License 管理、自定义定价。法律、医疗、财务规划等领域均有落地。
+
+了解更多：https://www.openclacky.com/ → Creators
+
+## 从源码安装
+
+```bash
+git clone https://github.com/clacky-ai/openclacky.git
+cd openclacky
+bundle install
+bin/clacky
+```
+
+## 信任与背书
+
+- **100% 开源** — MIT 协议，所有代码公开，所有决策可溯源
+- **2 年 Agentic 研发** — 经历 3 代架构演进
+- **16 个核心工具** — 极简设计
+- **投资方** — 奇绩创坛 · 真格基金 · 红杉中国 · 高瓴资本
+
+## 关注作者公众号
+
+本项目由 **李亚飞** 创立并主导开发。如果你对 AI Agent 工程、Harness 设计、创业经历感兴趣，欢迎关注微信公众号： **技术达人李亚飞**
+
+近期文章：
+
+- [从 ShowMeBug 到 OpenClacky：我对 AI 时代的 4 次下注](https://mp.weixin.qq.com/s/wTW-IU5Czu-OpJTFh_mwgA)
+- [我把 AI 账单从 30 美金打到 5 美金](https://mp.weixin.qq.com/s/BDhE0y8xbX0ea3vLlV37Ig)
+- [100% Cache 命中的 Harness 怎么设计：一个开源 AI Agent 的 7 个工程决策](https://mp.weixin.qq.com/s/Rc1xk0Qw168D4Y07kkBiGQ)
+
+## 参与贡献
+
+欢迎在 GitHub 提交 Bug 报告和 Pull Request：https://github.com/clacky-ai/openclacky 。参与贡献者须遵守[行为准则](https://github.com/clacky-ai/openclacky/blob/main/CODE_OF_CONDUCT.md)。
+
+## 许可证
+
+基于 [MIT 协议](https://opensource.org/licenses/MIT) 开源发布。

data/docs/engineering-article.md

@@ -4,7 +4,7 @@
 
 ---
 
-I'm Yafei Lee, founder of [OpenClacky](https://github.com/clacky-ai/openclacky), an open-source AI agent written in Ruby. We wanted an agent with skills, memory, sub-agents, browser automation, dynamic model switching, and long-running sessions. Each of those features made prompt caching worse in a different way.
+I'm Yafei Lee, founder of [OpenClacky](https://github.com/clacky-ai/openclacky), an open-source AI Agent written in Ruby. We wanted an agent with skills, memory, sub-agents, browser automation, dynamic model switching, and long-running sessions. Each of those features made prompt caching worse in a different way.
 
 That was the real architecture problem. Not how to call an LLM, not how to add another tool, not how to orchestrate more agents — how to keep the cache prefix stable while the product keeps changing.
 

data/lib/clacky/agent/llm_caller.rb

@@ -104,6 +104,7 @@ module Clacky
             tools: tools_to_send,
             max_tokens: @config.max_tokens,
             enable_caching: @config.enable_prompt_caching,
+            reasoning_effort: @reasoning_effort,
             on_chunk: build_progress_on_chunk
           )
 

data/lib/clacky/agent/session_serializer.rb

@@ -63,6 +63,9 @@ module Clacky
           @pending_error_rollback = true
         end
 
+        saved_reasoning = session_data.dig(:config, :reasoning_effort)
+        self.reasoning_effort = saved_reasoning if saved_reasoning
+
         # Restore the session's original model if it still exists in the current
         # config. This prevents all sessions from silently switching to the new
         # default model when the user changes it and restarts. Falls back to the
@@ -128,6 +131,7 @@ module Clacky
             enable_prompt_caching: @config.enable_prompt_caching,
             max_tokens: @config.max_tokens,
             verbose: @config.verbose,
+            reasoning_effort: @reasoning_effort,
             # Persist the current model identity so the session can restore its
             # original model on restart. model_name + model_base_url form a
             # composite key to avoid matching a different provider's model of

data/lib/clacky/agent.rb

@@ -43,13 +43,30 @@ module Clacky
     attr_reader :session_id, :name, :history, :iterations, :total_cost, :working_dir, :created_at, :total_tasks, :todos,
                 :cache_stats, :cost_source, :ui, :skill_loader, :agent_profile,
                 :status, :error, :updated_at, :source,
-                :latest_latency  # Hash of latency metrics from the most recent LLM call (see Client#send_messages_with_tools)
+                :latest_latency,  # Hash of latency metrics from the most recent LLM call (see Client#send_messages_with_tools)
+                :reasoning_effort
     attr_accessor :pinned
 
+    REASONING_EFFORTS = %w[low medium high].freeze
+
     def permission_mode
       @config&.permission_mode&.to_s || ""
     end
 
+    def reasoning_effort=(value)
+      @reasoning_effort = normalize_reasoning_effort(value)
+    end
+
+    private def normalize_reasoning_effort(value)
+      return nil if value.nil?
+      str = value.to_s.strip.downcase
+      return nil if str.empty? || str == "off" || str == "none"
+      return str if REASONING_EFFORTS.include?(str)
+      nil
+    end
+
+    public
+
     def initialize(client, config, working_dir:, ui:, profile:, session_id:, source:)
       @client = client  # Client for current model
       @config = config.is_a?(AgentConfig) ? config : AgentConfig.new(config)
@@ -79,6 +96,7 @@ module Clacky
       @task_cost_source = :estimated  # Track cost source for current task
       @previous_total_tokens = 0  # Track tokens from previous iteration for delta calculation
       @latest_latency = nil  # Most recent LLM call's latency metrics (see Client#send_messages_with_tools)
+      @reasoning_effort = nil  # Per-session reasoning effort override; nil = provider default
       @ui = ui  # UIController for direct UI interaction
       @debug_logs = []  # Debug logs for troubleshooting
       @pending_injections = []     # Pending inline skill injections to flush after observe()
@@ -102,6 +120,9 @@ module Clacky
       # Background sync: compare remote skill versions and download updates quietly.
       # Runs in a daemon thread so Agent startup is never blocked.
       @brand_config.sync_brand_skills_async!
+      # Free-mode counterpart: branded but not activated → fetch unencrypted skills
+      # via the public endpoint so users get a working install with no serial number.
+      @brand_config.sync_free_skills_async!
 
       # Initialize Time Machine
       init_time_machine

data/lib/clacky/brand_config.rb

@@ -410,6 +410,86 @@ module Clacky
       end
     end
 
+    # Fetch the list of free (unencrypted, published) skills available for the
+    # configured package_name. Anonymous endpoint — no license key required.
+    # This is what powers the "no serial number" free mode: a branded install
+    # that is not activated still gets the creator's free skills automatically.
+    #
+    # Returns { success: bool, skills: [], error: }. Each skill in the returned
+    # array carries the same shape as fetch_brand_skills! (name, latest_version,
+    # description, etc.) so install_brand_skill! can consume it directly.
+    def fetch_free_skills!
+      return { success: false, error: "Not branded", skills: [] } unless branded?
+      if @package_name.nil? || @package_name.strip.empty?
+        return { success: false, error: "package_name not configured", skills: [] }
+      end
+
+      encoded_pkg = URI.encode_www_form_component(@package_name.strip)
+      response    = platform_client.get("/api/v1/distributions/free_skills?package_name=#{encoded_pkg}")
+
+      if response[:success] && response[:data].is_a?(Hash)
+        installed = installed_brand_skills
+        skills    = (response[:data]["skills"] || []).map do |skill|
+          normalized   = skill["name"].to_s.downcase.gsub(/[\s_]+/, "-").gsub(/[^a-z0-9-]/, "").gsub(/-+/, "-")
+          name         = installed.keys.find { |k| k == normalized } || normalized
+          local        = installed[name]
+          latest_ver   = (skill["latest_version"] || {})["version"] || skill["version"]
+          needs_update = local ? version_older?(local["version"], latest_ver) : false
+          skill.merge(
+            "name"              => name,
+            "installed_version" => local ? local["version"] : nil,
+            "needs_update"      => needs_update
+          )
+        end
+        { success: true, skills: skills, paid_skills_count: response[:data]["paid_skills_count"].to_i }
+      else
+        { success: false, error: response[:error] || "Failed to fetch free skills", skills: [], paid_skills_count: 0 }
+      end
+    end
+
+    # Install a single free (unencrypted) skill. Thin wrapper around
+    # install_brand_skill! that records the skill as encrypted: false so the
+    # loader reads SKILL.md directly without attempting decryption.
+    def install_free_skill!(skill_info)
+      install_brand_skill!(skill_info, encrypted: false)
+    end
+
+    # Synchronise free skills in the background for unactivated branded installs.
+    #
+    # Mirrors sync_brand_skills_async! but uses the public free_skills endpoint
+    # so no license is required. Only runs when the install is branded and NOT
+    # activated — once a license is activated the regular brand-skill sync
+    # takes over (and may include additional encrypted skills).
+    #
+    # @return [Thread, nil]
+    def sync_free_skills_async!(on_complete: nil)
+      return nil unless branded?
+      return nil if activated?
+      return nil if ENV["CLACKY_TEST"] == "1"
+
+      Thread.new do
+        Thread.current.abort_on_exception = false
+
+        begin
+          result = fetch_free_skills!
+          next unless result[:success]
+
+          remote_skill_names = result[:skills].map { |s| s["name"] }
+          installed_brand_skills.each_key do |local_name|
+            send(:delete_brand_skill!, local_name) unless remote_skill_names.include?(local_name)
+          end
+
+          installed = installed_brand_skills
+          to_install = result[:skills].select { |s| installed[s["name"]].nil? || s["needs_update"] }
+          results    = to_install.map { |skill_info| install_free_skill!(skill_info) }
+
+          on_complete&.call(results)
+        rescue StandardError
+          # Background sync failures are intentionally swallowed.
+        end
+      end
+    end
+
     # Upload (publish) a custom skill ZIP to the OpenClacky Cloud API.
     # Calls POST /api/v1/client/skills (system-license endpoint).
     # zip_data is the raw binary content of the ZIP file.
@@ -629,7 +709,9 @@ module Clacky
 
     # Install (or update) a single brand skill by downloading and extracting its zip.
     # skill_info: a hash from fetch_brand_skills! with at least name + latest_version.download_url + version
-    def install_brand_skill!(skill_info)
+    # encrypted: whether the ZIP contains AES-encrypted .enc files + MANIFEST.enc.json (true)
+    #            or plaintext SKILL.md and supporting files (false, used by free-mode).
+    def install_brand_skill!(skill_info, encrypted: true)
       require "net/http"
       require "uri"
 
@@ -698,10 +780,10 @@ module Clacky
 
       FileUtils.rm_f(tmp_zip)
 
-      # Record installed version in brand_skills.json (including description for
-      # offline display when the remote API is unreachable).
-      # encrypted: true because the ZIP contains MANIFEST.enc.json + AES-256-GCM encrypted files.
-      record_installed_skill(slug, version, skill_info["description"], encrypted: true, description_zh: skill_info["description_zh"], name_zh: skill_info["name_zh"])
+      record_installed_skill(slug, version, skill_info["description"],
+                             encrypted: encrypted,
+                             description_zh: skill_info["description_zh"],
+                             name_zh: skill_info["name_zh"])
 
       { success: true, name: slug, version: version }
     rescue StandardError, ScriptError => e

data/lib/clacky/client.rb

@@ -125,7 +125,7 @@ module Clacky
     #   path. When given but streaming is not yet wired for the active provider,
     #   a single synthetic invocation is fired after the response is received,
     #   so UI plumbing can be exercised end-to-end without the proxy work.
-    def send_messages_with_tools(messages, model:, tools:, max_tokens:, enable_caching: false, on_chunk: nil)
+    def send_messages_with_tools(messages, model:, tools:, max_tokens:, enable_caching: false, reasoning_effort: nil, on_chunk: nil)
       caching_enabled = enable_caching && supports_prompt_caching?(model)
       cloned = deep_clone(messages)
 
@@ -140,13 +140,13 @@ module Clacky
       response =
         if bedrock?
           streaming_used = !on_chunk.nil?
-          send_bedrock_request(cloned, model, tools, max_tokens, caching_enabled, on_chunk: wrapped_on_chunk)
+          send_bedrock_request(cloned, model, tools, max_tokens, caching_enabled, reasoning_effort: reasoning_effort, on_chunk: wrapped_on_chunk)
         elsif anthropic_format?
           streaming_used = !on_chunk.nil?
-          send_anthropic_request(cloned, model, tools, max_tokens, caching_enabled, on_chunk: wrapped_on_chunk)
+          send_anthropic_request(cloned, model, tools, max_tokens, caching_enabled, reasoning_effort: reasoning_effort, on_chunk: wrapped_on_chunk)
         else
           streaming_used = !on_chunk.nil?
-          send_openai_request(cloned, model, tools, max_tokens, caching_enabled, on_chunk: wrapped_on_chunk)
+          send_openai_request(cloned, model, tools, max_tokens, caching_enabled, reasoning_effort: reasoning_effort, on_chunk: wrapped_on_chunk)
         end
       t1 = Process.clock_gettime(Process::CLOCK_MONOTONIC)
 
@@ -217,8 +217,8 @@ module Clacky
 
     # ── Bedrock Converse request / response ───────────────────────────────────
 
-    def send_bedrock_request(messages, model, tools, max_tokens, caching_enabled, on_chunk: nil)
-      body = MessageFormat::Bedrock.build_request_body(messages, model, tools, max_tokens, caching_enabled)
+    def send_bedrock_request(messages, model, tools, max_tokens, caching_enabled, reasoning_effort: nil, on_chunk: nil)
+      body = MessageFormat::Bedrock.build_request_body(messages, model, tools, max_tokens, caching_enabled, reasoning_effort: reasoning_effort)
       return send_bedrock_stream_request(body, model, on_chunk) if on_chunk
 
       response = bedrock_connection.post(bedrock_endpoint(model)) { |r| r.body = body.to_json }
@@ -248,7 +248,10 @@ module Clacky
         end
       end
 
-      raise_error(response) unless response.status == 200
+      unless response.status == 200
+        response.env.body = sse_buf if response.body.to_s.empty?
+        raise_error(response)
+      end
       MessageFormat::Bedrock.parse_response(aggregator.to_h)
     end
 
@@ -263,11 +266,11 @@ module Clacky
 
     # ── Anthropic request / response ──────────────────────────────────────────
 
-    def send_anthropic_request(messages, model, tools, max_tokens, caching_enabled, on_chunk: nil)
+    def send_anthropic_request(messages, model, tools, max_tokens, caching_enabled, reasoning_effort: nil, on_chunk: nil)
       # Apply cache_control to the message that marks the cache breakpoint
       messages = apply_message_caching(messages) if caching_enabled
 
-      body = MessageFormat::Anthropic.build_request_body(messages, model, tools, max_tokens, caching_enabled)
+      body = MessageFormat::Anthropic.build_request_body(messages, model, tools, max_tokens, caching_enabled, reasoning_effort: reasoning_effort)
       return send_anthropic_stream_request(body, on_chunk) if on_chunk
 
       response = anthropic_connection.post(anthropic_messages_path) { |r| r.body = body.to_json }
@@ -304,14 +307,15 @@ module Clacky
 
     # ── OpenAI request / response ─────────────────────────────────────────────
 
-    def send_openai_request(messages, model, tools, max_tokens, caching_enabled, on_chunk: nil)
+    def send_openai_request(messages, model, tools, max_tokens, caching_enabled, reasoning_effort: nil, on_chunk: nil)
       # Apply cache_control markers to messages when caching is enabled.
       # OpenRouter proxies Claude with the same cache_control field convention as Anthropic direct.
       messages = apply_message_caching(messages) if caching_enabled
 
       body = MessageFormat::OpenAI.build_request_body(
         messages, model, tools, max_tokens, caching_enabled,
-        vision_supported: @vision_supported
+        vision_supported: @vision_supported,
+        reasoning_effort: reasoning_effort
       )
       return send_openai_stream_request(body, on_chunk) if on_chunk
 

data/lib/clacky/message_format/anthropic.rb

@@ -48,7 +48,7 @@ module Clacky
       # @param max_tokens [Integer]
       # @param caching_enabled [Boolean]
       # @return [Hash] ready to serialize as JSON body
-      def build_request_body(messages, model, tools, max_tokens, caching_enabled)
+      def build_request_body(messages, model, tools, max_tokens, caching_enabled, reasoning_effort: nil)
         system_messages = messages.select { |m| m[:role] == "system" }
         regular_messages = messages.reject { |m| m[:role] == "system" }
 
@@ -64,9 +64,21 @@ module Clacky
         body = { model: model, max_tokens: max_tokens, messages: api_messages }
         body[:system] = system_text unless system_text.empty?
         body[:tools]  = api_tools   if api_tools&.any?
+
+        if (effort = normalized_effort(reasoning_effort))
+          body[:thinking] = { type: "adaptive" }
+          body[:output_config] = { effort: effort }
+        end
+
         body
       end
 
+      private_class_method def self.normalized_effort(effort)
+        return nil if effort.nil? || effort.to_s.empty?
+        s = effort.to_s
+        %w[low medium high].include?(s) ? s : nil
+      end
+
       # ── Response parsing ──────────────────────────────────────────────────────
 
       # Parse Anthropic API response into canonical internal format.

data/lib/clacky/message_format/bedrock.rb

@@ -52,7 +52,7 @@ module Clacky
       # @param max_tokens [Integer]
       # @param caching_enabled [Boolean] (currently unused for Bedrock)
       # @return [Hash] ready to serialize as JSON body
-      def build_request_body(messages, model, tools, max_tokens, caching_enabled = false)
+      def build_request_body(messages, model, tools, max_tokens, caching_enabled = false, reasoning_effort: nil)
         system_messages = messages.select { |m| m[:role] == "system" }
         regular_messages = messages.reject { |m| m[:role] == "system" }
 
@@ -83,9 +83,21 @@ module Clacky
           body[:toolConfig] = { tools: tools.map { |t| to_api_tool(t) } }
         end
 
+        extra = additional_fields_for_effort(reasoning_effort)
+        body[:additionalModelRequestFields] = extra if extra
+
         body
       end
 
+      private_class_method def self.additional_fields_for_effort(effort)
+        return nil if effort.nil? || effort.to_s.empty?
+        return nil unless %w[low medium high].include?(effort.to_s)
+        {
+          thinking: { type: "adaptive" },
+          output_config: { effort: effort.to_s }
+        }
+      end
+
       # ── Response parsing ──────────────────────────────────────────────────────
 
       # Parse Bedrock Converse API response into canonical internal format.

data/lib/clacky/message_format/open_ai.rb

@@ -44,7 +44,7 @@ module Clacky
       # @param vision_supported [Boolean] whether the target model accepts
       #   image_url content blocks (default true, conservative)
       # @return [Hash]
-      def build_request_body(messages, model, tools, max_tokens, caching_enabled, vision_supported: true)
+      def build_request_body(messages, model, tools, max_tokens, caching_enabled, vision_supported: true, reasoning_effort: nil)
         api_messages = messages.map { |msg| normalize_message_content(msg, vision_supported: vision_supported) }
 
         body = { model: model, max_tokens: max_tokens, messages: api_messages }
@@ -59,6 +59,10 @@ module Clacky
           end
         end
 
+        if reasoning_effort && !reasoning_effort.to_s.empty?
+          body[:reasoning_effort] = reasoning_effort.to_s
+        end
+
         body
       end