pygpt-net 2.6.59__py3-none-any.whl → 2.6.61__py3-none-any.whl

{pygpt_net-2.6.59.dist-info → pygpt_net-2.6.61.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: pygpt-net
-Version: 2.6.59
+Version: 2.6.61
 Summary: Desktop AI Assistant powered by: OpenAI GPT-5, GPT-4, o1, o3, Gemini, Claude, Grok, DeepSeek, and other models supported by Llama Index, and Ollama. Chatbot, agents, completion, image generation, vision analysis, speech-to-text, plugins, MCP, internet access, file handling, command execution and more.
 License: MIT
 Keywords: ai,api,api key,app,assistant,bielik,chat,chatbot,chatgpt,claude,dall-e,deepseek,desktop,gemini,gpt,gpt-3.5,gpt-4,gpt-4-vision,gpt-4o,gpt-5,gpt-oss,gpt3.5,gpt4,grok,langchain,llama-index,llama3,mistral,o1,o3,ollama,openai,presets,py-gpt,py_gpt,pygpt,pyside,qt,text completion,tts,ui,vision,whisper
@@ -117,7 +117,7 @@ Description-Content-Type: text/markdown
 
 [![pygpt](https://snapcraft.io/pygpt/badge.svg)](https://snapcraft.io/pygpt)
 
-Release: **2.6.59** | build: **2025-09-23** | Python: **>=3.10, <3.14**
+Release: **2.6.61** | build: **2025-09-26** | Python: **>=3.10, <3.14**
 
 > Official website: https://pygpt.net | Documentation: https://pygpt.readthedocs.io
 > 
@@ -180,6 +180,7 @@ You can download compiled 64-bit versions for Windows and Linux here: https://py
 - Provides an intuitive operation and interface.
 - Includes a notepad.
 - Includes simple painter / drawing tool.
+- Includes an node-based Agents Builder.
 - Supports multiple languages.
 - Requires no previous knowledge of using AI models.
 - Simplifies image generation using image models like `DALL-E` and `Imagen`.
@@ -2088,6 +2089,7 @@ PyGPT features several useful tools, including:
 - HTML/JS Canvas (built-in HTML renderer)
 - Translator
 - Web Browser (Chromium)
+- Agents Builder (beta)
 
 ![v2_tool_menu](https://github.com/szczyglis-dev/py-gpt/raw/master/docs/source/images/v2_tool_menu.png)
 
@@ -2165,6 +2167,71 @@ Enables translation between multiple languages using an AI model.
 
 A built-in web browser based on Chromium, allowing you to open webpages directly within the app. **SECURITY NOTICE:** For your protection, avoid using the built-in browser for sensitive or critical tasks. It is intended for basic use only.
 
+# Agents Builder (beta)
+
+To launch the Agent Editor, navigate to:
+
+`Tools -> Agents Builder`
+
+![nodes](https://github.com/szczyglis-dev/py-gpt/raw/master/docs/source/images/nodes.png)
+
+This tool allows you to create workflows for agents using a node editor, without writing any code. You can add a new agent type, and it will appear in the list of presets.
+
+To add a new element, right-click on the editor grid and select `Add` to insert a new node.
+
+**Types of Nodes:**
+
+- **Flow/Start**: The starting point for agents (user input).
+- **Flow/Agent**: A single agent with customizable default parameters, such as system instructions and tool usage. These settings can be overridden in the preset.
+- **Flow/Memory**: Shared memory between agents (shared Context).
+- **Flow/End**: The endpoint, returning control to the user.
+
+Agents with connected shared memory share it among themselves. Agents without shared memory only receive the latest output from the previous agent.
+
+The first agent in the sequence always receives the full context passed by the user.
+
+Connecting agents and memory is done using node connections via slots. To connect slots, simply drag from the input port to the output port (Ctrl + mouse button removes a connection).
+
+**Node Editor Navigation:**
+
+- **Right-click**: Add node, undo, redo, clear
+- **Middle-click + drag**: Pan view
+- **Ctrl + Mouse wheel**: Zoom
+- **Left-click a port**: Create connection
+- **Ctrl + Left-click a port**: Rewire or detach connection
+- **Right-click or DELETE a node/connection**: Remove node/connection
+
+**Tip:** Enable agent debugging in `Settings -> Debug -> Log Agents usage to console` to log the full workflow to the console.
+
+Agents built using this tool are compatible with both OpenAI Agents and LlamaIndex.
+
+**Notes:**
+
+Routing and system instruction: for every agent that has more than one connection leading to the next agent, a routing instruction is automatically injected just before your system prompt:
+
+```
+You are a routing-capable agent in a multi-agent flow.
+Your id is: <current_id>, name: <agent_name>.
+You MUST respond ONLY with a single JSON object and nothing else.
+Schema:
+{
+  "route": "<ID of the next agent from allowed_routes OR the string 'end'>",
+  "content": "<final response text for the user (or tool result)>"
+}
+Rules:
+- allowed_routes: [<allowed>]
+- If you want to finish the flow, set route to "end".
+- content must contain the user-facing answer (you may include structured data as JSON or Markdown inside content).
+- Do NOT add any commentary outside of the JSON. No leading or trailing text.
+- If using tools, still return the final JSON with tool results summarized in content.
+- Human-friendly route names: <names>
+- Human-friendly route roles (optional): <roles>
+
+<here begins your system instruction>
+```
+
+**INFO:** Agents Builder is in beta.
+
 
 # Token usage calculation
 
@@ -2389,8 +2456,6 @@ Config -> Settings...
 
 - `Store dialog window positions`: Enable or disable dialogs positions store/restore, Default: True.
 
-- `Use theme colors in chat window`: Use color theme in chat window, Default: True.
-
 **Code syntax**
 
 - `Code syntax highlight`: Syntax highlight theme in code blocks. `WebEngine / Chromium` render mode only.
@@ -2457,8 +2522,6 @@ Config -> Settings...
 
 - `Open URLs in built-in browser`: Enable this option to open all URLs in the built-in browser (Chromium) instead of an external browser. Default: False.
 
-- `Convert lists to paragraphs`: If enabled, lists (ul, ol) will be converted to paragraphs (p), Default: True.
-
 - `Model used for auto-summary`: Model used for context auto-summary (generating titles in context list) (default: *gpt-4o-mini*). **Tip:** If you prefer to use local models, you should change the model here as well
 
 **Remote tools**
@@ -3658,6 +3721,17 @@ may consume additional tokens that are not displayed in the main window.
 
 ## Recent changes:
 
+**2.6.61 (2025-09-26)**
+
+- Enhanced the agents node editor, custom agent flow, and instruction following.
+- Added drag-and-drop and reordering functionality to the presets list.
+- Added statistics for response tokens, including time elapsed and tokens per second.
+- Improved UI/UX.
+
+**2.6.60 (2025-09-25)**
+
+- Added a new tool: Agents Builder - allowing visual design of agent workflows using nodes - available in Tools -> Agents Builder (beta).
+
 **2.6.59 (2025-09-23)**
 
 - LlamaIndex has been upgraded to v0.13.6.
@@ -3674,93 +3748,6 @@ may consume additional tokens that are not displayed in the main window.
 - Improved: The local web search plugin has been enhanced to retrieve multiple URLs at once.
 - Added: Use proxy switch in Settings.
 
-**2.6.56 (2025-09-22)**
-
-- Optimized: Memory usage and performance in streaming and rendering large contexts.
-- Added: Copy to clipboard functionality in user messages.
-- Added: Manual scroll in plain-text mode.
-
-**2.6.55 (2025-09-18)**
-
-- Fixed: Unnecessary context loading from the database.
-- Optimized: Token count in the input field.
-
-**2.6.54 (2025-09-18)**
-
-- Added: Remote tools (like web search) are now also available in the Chat with Files and Agents (LlamaIndex) modes.
-- Added: Two new plugins: Wolfram Alpha and OpenStreetMap.
-- Fixed: Enabled local file-like schemes in links/images in the markdown-it parser.
-
-**2.6.53 (2025-09-17)**
-
-- Added: An icon to enable/disable the web search remote tool in the icon bar, along with remote web search functionality in OpenRouter (#135).
-- Added: The ability to mute audio in real-time mode via the audio icon.
-
-**2.6.52 (2025-09-17)**
-
-- Added MCP plugin: Provides access to remote tools via the Model Context Protocol (MCP), including stdio, SSE, and Streamable HTTP transports, with per-server allow/deny filtering, Authorization header support, and a tools cache.
-- Fixed: tab tooltips reload on profile switch.
-
-**2.6.51 (2025-09-16)**
-
-- Fix: Automatically reloading calendar notes.
-- Fix: Context menu CSS background color for calendar items.
-
-**2.6.50 (2025-09-16)**
-
-- Optimized: Improved memory cleanup when switching profiles and unloading tabs.
-- Fix: Resolved missing PID data in text output.
-- Fix: Enhanced real-time parsing of execute tags.
-
-**2.6.49 (2025-09-16)**
-
-- Fixed: Occasional crashes when focusing on an output container unloaded from memory in the second column.
-
-**2.6.48 (2025-09-15)**
-
-- Added: auto-loading of next items to the list of contexts when scrolling to the end of the list.
-
-**2.6.47 (2025-09-15)**
-
-- Improved: Parsing of custom markup tags.
-- Optimized: Switching profiles.
-
-**2.6.46 (2025-09-15)**
-
-- Added: Global proxy settings for all API SDKs.
-- Fixed: xAI client configuration in Chat with Files.
-- Fixed: Top margin in streaming container.
-- Refactored: Debug workers.
-
-**2.6.45 (2025-09-13)**
-
-- Improved: Parsing of custom markup in the stream.
-- Improved: Message block parsing moved to JavaScript.
-
-**2.6.44 (2025-09-12)**
-
-- Added: Auto-collapse for large user input blocks.
-- Added: Configuration for syntax highlighting intervals.
-- Improved: Visibility of label icons.
-- Improved: Scrolling of code blocks.
-- Fixed: Parsing of quotes in custom markdown blocks.
-
-**2.6.43 (2025-09-12)**
-
-- Fixed: preset restoration when switching profiles.
-- Improved: faster application launch and exit.
-
-**2.6.42 (2025-09-11)**
-
-- Fixed: Save/load zoom factor in the chat window when switched via Ctrl + mouse wheel.
-- Fixed: Break the first line in code blocks.
-- Added: Configuration options for syntax highlight limits.
-- Updated: SVG icons.
-
-**2.6.41 (2025-09-11)**
-
-- Rendering engine optimizations: markdown parsing moved to JS, reduced CPU usage, added auto-memory clearing, and more.
-
 # Credits and links
 
 **Official website:** <https://pygpt.net>