Codeberg Proposes ToS Update to Ban LLM Training Data Extraction

·1 views

Codeberg, an open-source code hosting platform, is considering an extension to its Terms of Use that would explicitly prohibit the extraction of data for training large language models. The proposed change is outlined in a pull request on Codeberg's own repository, inviting community review and discussion. The move reflects growing concern among open-source communities about AI companies scraping hosted code without contributor consent. If adopted, the updated terms would add a new layer of protection for developers who host their projects on the platform.

Read the full story at Hacker News

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Developer launches AI platform to zero signups, discovers 5 critical hidden bugs

A developer launched an AI agent platform on Product Hunt and received zero signups, prompting a deep dive into the product after user feedback highlighted a lack of cost transparency. Investigation revealed a bug where integer truncation zeroed out nearly all cost data, since most LLM calls cost less than one cent per execution. Further digging uncovered that workflow-level cost and token fields had never been written to the database, a model was being billed at roughly 30 times its actual rate, and cost data returned by the API was never displayed in the frontend. Most critically, the platform's four analytics pages were entirely unreachable — no navigation links pointed to them anywhere in the app, making them effectively invisible to all users.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer builds .NET library to fill Mistral Agents API gap for enterprise workflows

A developer has released Mistral.Agents.Net, an open-source .NET library designed to bridge the missing agentic layer in existing community SDKs for Mistral's platform. Unlike stateless chat completions, Mistral's Agents API supports persistent agents, stateful conversation threads, and built-in tools such as web search and a code interpreter. The library enables .NET developers to create multi-agent workflows with handoff support, allowing one agent to delegate tasks to another without manual routing logic. It also exposes function tool support, letting developers connect agents to private internal data sources via C# methods. The project follows a typed-first, raw-fallback design philosophy, offering structured outputs for common cases and a JSON escape hatch for edge cases.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

whoimports tool scans Python codebases to trace all imports of a module

A new open-source command-line tool called whoimports helps Python developers identify every file that imports a given module before renaming, moving, or deleting it. Unlike basic text searches, it uses Python's Abstract Syntax Tree (AST) to accurately detect import statements without triggering false positives from strings or comments. The tool supports both file paths and dotted module names, and can output results in plain text, Markdown, or JSON formats. It requires no external dependencies and works with Python 3.10 and above. Available on GitHub under the MIT license, it is designed to complement other refactoring tools such as gitchurn and redactx.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Deep Agents Uses Middleware Stack to Handle Real-World AI Agent Complexity

Building functional AI agents requires far more than a basic model-tool loop, and the Deep Agents framework addresses this through a built-in middleware system. Middleware consists of independent code layers that wrap the core agent loop, each handling a specific concern such as context management, file access, or task tracking. When developers create an agent using Deep Agents, they automatically receive a default middleware stack that includes tools for todo-list management, file system access, sub-agent coordination, conversation summarization, and broken tool-call recovery. The stack runs in a defined sequence, with each layer processing requests before they reach the model and responses on the way back out. Developers can also insert custom middleware into the pipeline without modifying the agent's core logic, enabling tailored behaviors like human-approval gates or custom summarization strategies.

0 comments Read more at DEV Community