simonw pushed to simonw/llm: code update
code update
Chronological feed of everything captured from Simon Willison.
code update
code update
f.k.a. Awesome ChatGPT Prompts. Share, discover, and collect prompts from the community. Free and open source — self-host for your organization with complete privacy.. Stars: 159499
Data scraped from LG&E/KU's outage map. Stars: 6
An up-to-date archive of the data from https://petition.parliament.uk & http://petitions.number10.gov.uk. Stars: 16
The highest-scoring AI memory system ever benchmarked. And it's free.. Stars: 43697
MapLibre GL JS - Interactive vector tile maps in the browser. Stars: 10364
Open Lakehouse Format for Multimodal AI. Convert from Parquet in 2 lines of code for 100x faster random access, vector index, and data versioning. Compatible with Pandas, DuckDB, Polars, Pyarrow, and PyTorch with more integrations coming... Stars: 6314
AutoGPT is the vision of accessible AI for everyone, to use and to build on. Our mission is to provide the tools, so that you can focus on what matters.. Stars: 183356
Mobile and Web client for Codex and Claude Code, with realtime voice, encryption and fully featured. Stars: 17812
The easy-to-use open source Business Intelligence and Embedded Analytics tool that lets everyone work with data :bar_chart:. Stars: 46829
code update
code update
Infisical is the open-source platform for secrets, certificates, and privileged access management.. Stars: 25841
A course of learning LLM inference serving on Apple Silicon for systems engineers: build a tiny vLLM + Qwen.. Stars: 4081
The live data layer for apps and AI agents. Create up-to-the-second views into your business, just using SQL. Stars: 6269
Test your prompts, agents, and RAGs. Red teaming/pentesting/vulnerability scanning for AI. Compare performance of GPT, Claude, Gemini, Llama, and more. Simple declarative configs with command line and CI/CD integration. Used by OpenAI and Anthropic.. Stars: 19992
Asynchronous HTTP client/server framework for asyncio and Python. Stars: 16398
Blazing-fast Data-Wrangling toolkit. Stars: 3594
Fyn is a fork of uv for fast Python package management, dependency resolution, virtual environments, and pyproject.toml workflows.. Stars: 301
code update
code update
code update
code update
code update
🚀 The fast, Pythonic way to build MCP servers and clients.. Stars: 24485
claude-code-action. Stars: 7020
Playwright is a framework for Web Testing and Automation. It allows testing Chromium, Firefox and WebKit with a single API. . Stars: 86218
A modern replacement for Redis and Memcached. Stars: 30336
Open source, privacy-first web analytics. Lightweight, cookie-free Google Analytics alternative. Self-hosted or cloud.. Stars: 24570
LLM inference in C/C++. Stars: 103273
An extremely fast Python linter and code formatter, written in Rust.. Stars: 47008
Go ahead and axolotl questions. Stars: 11655
Questionably Accurate Course Scheduler for RPI. Stars: 44
A query and indexing engine for Redis, providing secondary indexing, full-text search, vector similarity search and aggregations.. Stars: 6118
A Cross-Site Scripting (XSS) vulnerability was identified in the Claude Code Timeline Viewer's markdown rendering function. The original implementation failed to properly sanitize user-supplied URLs and attribute values within markdown links, allowing for JavaScript injection. The vulnerability was mitigated by integrating the DOMPurify library, which now sanitizes the HTML output of the markdown renderer, effectively preventing malicious script execution and attribute manipulation while preserving intended markdown functionality, including newly added table support.
Simon Willison iteratively enhanced claude-code-timeline.html, a client-side viewer for Claude Code .jsonl session logs, by adding a "Copy prompts" button above the timeline/detail panels. The button filters events by type='user', role='user', and content-type='text', extracts text content from matching messages, concatenates with double newlines, and copies to clipboard using existing copyText utility. A 2-second "Copied!" state feedback was added via setTimeout for improved UX. Implementation used precise Edit tool calls on HTML structure, els object, and script logic without disrupting core timeline/filtering functionality.
A Python script, `claude_code_to_gist.py`, was modified to ensure consistent project folder referencing. Previously, it used a relative path, but now it explicitly points to `~/.claude/projects`, enhancing predictability and portability for locating `.jsonl` session files for Gist publication.
The Claude Code Timeline Viewer is being upgraded to include a 'Show prompts' button that reveals a modal dialog containing user prompts. This enhancement, inspired by the Codex timeline functionality, improves prompt visibility and user experience by decoupling prompt viewing from immediate clipboard operations. The update involves adding new HTML elements, CSS styling, and JavaScript functions to manage the modal's state and interactions, ensuring a more intuitive way to access and copy user inputs.
The `llm logs` command-line interface has been enhanced to include a dedicated "## Options" section in its Markdown output for individual log entries. This new section, formatted as a YAML-like list, displays model-specific options when they are present in the `options_json` field of the log record. This improvement provides clearer visibility into the parameters used for each model interaction directly within the human-readable log view, streamlining analysis for technical users.
The conversation focuses on optimizing how the `llm-gemini` plugin handles media resolution for Gemini models, particularly to accurately reflect user intent in logging and efficiently interact with the API. The key change involves modifying the default `media_resolution` option to `None` and refining the request-building logic to only explicitly set a resolution when specified by the user or when a YouTube URL necessitates a low-resolution default for processing.
A minimal Uvicorn-based "Hello, World!" ASGI application was developed and thoroughly tested using a Test-Driven Development (TDD) approach. The project adheres to a strict dependency constraint, only utilizing Uvicorn for application runtime, with pytest and pytest-asyncio serving as development dependencies for robust testing. Comprehensive project documentation including a specification, README, and pyproject.toml was generated.
A FastAPI and Uvicorn-based HTTP proxy server has been developed and tested. This proxy transparently forwards all HTTP methods and preserves path, query strings, and request bodies to `simonwillison.net`. It enhances responses with `x-proxied-by: uvicorn` and `x-response-time-microseconds` headers, effectively demonstrating how to inject custom metadata and performance metrics into proxied traffic.
The Claude Code Timeline Viewer, a web-based tool for visualizing Claude Code sessions, has been updated. The `extractUserPrompts` function in the `claude-code-timeline.html` file was modified to filter out user prompts marked as `isCompactSummary: true` or `isMeta: true`. This enhancement prevents auto-generated context summaries and meta-messages from appearing in the user prompt timeline, improving the clarity of user interaction analysis.
The provided content explores a polyfill for the experimental CSS Grid Lanes feature, aimed at enabling masonry-style layouts across browsers. The core challenge lies in accurately parsing and applying CSS properties like `display: grid-lanes`, `grid-template-columns`, `gap`, and `item-tolerance` in environments lacking native support. The polyfill must overcome browser-specific interpretations of invalid CSS to correctly emulate the layout algorithm, which places items in the shortest available "lane" to optimize visual flow.
Simon Willison has not been granted access to the X platform, contrary to potential speculation. This insight clarifies the current status of his platform access.
The potential for "streaming experts" within a Mixture-of-Experts (MoE) model suggests a capability to dynamically allocate computational resources. This approach could enable more efficient processing by engaging specialized expert models only when relevant to the input stream. It implies an architectural evolution towards adaptive and on-demand expert utilization in large language models.
Running large language models (LLMs) for inference, especially those with high parameter counts, typically necessitates significant GPU memory. While some quantized models can operate on consumer-grade hardware like a 256GB or 512GB Mac Studio, larger, unquantized models predominantly require high-end NVIDIA GPU servers to ensure sufficient memory and computational throughput.
Image generation models are incorporating social media comments and descriptive text as training data, as evidenced by precise textual details appearing in generated images that mirror comments from posts. This suggests that the models are not only processing visual information but also integrating contextual textual descriptions from online interactions, even for unusual prompts. This expansive approach to data ingestion implies a broader definition of
Anthropic has made its advanced Opus-beating model exclusively available to partnered security research organizations under "Project Glasswing." This selective distribution strategy is likely a response to recent concerns from credible security experts, aiming to control access to powerful AI models for responsible research and development.