DeepSeek has introduced two new tools, a Search API and a Fetch API, giving developers programmatic access to live web content for AI agents and retrieval-augmented generation applications. The Search API returns structured results from DeepSeek's own search engine, while the Fetch API pulls a single webpage's content and converts it to Markdown; both use the same bearer-token authentication as DeepSeek's existing API and are priced at $0.75 per 1,000 successful requests.
api-docs.deepseek.com
· 2026-10-18
Google launched EmbeddingGemma 2, a follow-up to its original EmbeddingGemma model that reportedly surpassed 20 million downloads. The new 740-million-parameter model is built on the Gemma 4 architecture, released under an Apache 2.0 license, and extends beyond text to embed code, images, video, and audio in a single shared space for on-device inference.
blog.google
· 2026-10-06
A technical essay critiques the common architecture behind AI agent 'memory' plugins, describing them as systems that chop conversation transcripts into snippets, store them in vector databases, and retrieve the most similar ones per prompt. The author argues this retrieval-by-similarity approach fails to give agents real understanding of a project, and proposes that agents instead need structured documentation rather than memory snippets.
liao.gg
· 2026-10-03
Kapa, a platform that indexes company knowledge for AI agents, built an internal benchmark of 1,000 annotated eval cases from real production data to compare retrieval methods. The company tested seven retrievers—including traditional hybrid search, agentic grep-based search, and its own Kapa systems—measuring both accuracy and query time. Results showed Kapa's 'Deep' retriever scoring 0.65 in about five seconds, while a frontier model using only grep matched a tuned pipeline's 0.61 score but took roughly five times longer.
kapa.ai
· 2026-10-02
A new PDF parsing tool called papero converts PDFs into Markdown, JSON, Word or Excel while preserving reading order, tables, formulas, figures and the bounding-box position of every block. It runs entirely on CPU using geometric analysis rather than machine learning models, and is available as a browser app, a Python package (pip install papero-extract), or an API. The browser version processes files locally so the PDF does not leave the user's machine.
github.com
· 2026-10-01
A new textbook titled Foundation Model Engineering has been released, aimed at AI engineers and research-minded readers who want to understand foundation models beyond basic API use. It ties together topics like attention, mixture-of-experts, RLHF, multimodality, long-context inference, retrieval-augmented generation, and agents into a single engineering narrative, using PyTorch examples, quizzes, and interactive visualizers.
sungeuns.github.io
· 2026-09-14
Frigade, a YC W23 startup, has released an Assist API that gives in-app AI agents a single tool call to generate on-screen visual guides instead of relying on outdated help-center text or web searches. The tool uses a browser-based agent trained on a test account to understand a product's UI, then either produces a step-by-step visual walkthrough, returns a conceptual explanation, or declines if it can't help. A demo shows integration with the Vercel AI SDK.
news.ycombinator.com
· 2026-09-09
Egiziago Cioffi, CEO of Microsoft partner SynSphere Italia, discovered that an Azure OpenAI email assistant he built was returning SharePoint content to a low-privilege test account that the account could not access directly in SharePoint. The assistant had passed all evaluation scores and unit tests, but comparing outputs from a high-privilege and a low-privilege account against identical queries exposed the mismatch, revealing that retrieval was happening under the indexer's permissions rather than the requester's.
venturebeat.com
· 2026-09-01
A free set of runnable Colab notebooks called AI Engineer Notebooks has been released to teach applied LLM engineering skills like RAG, agents, evals, fine-tuning, and serving. The curriculum deliberately avoids frameworks like LangChain or LlamaIndex, instead having learners build agent loops and retrieval pipelines from raw API calls using the free Groq API. Two GPU-heavy topics, LoRA fine-tuning and self-hosted serving, include optional appendices tested on Colab's free T4 GPU.
github.com
· 2026-08-27
A proposed technique lets websites detect AI agents through HTTP Accept headers and respond with stripped-down Markdown instead of full HTML pages. This version omits navigation bars, stylesheets, scripts, and layout elements, leaving only the core textual content for automated systems to process.
acceptmarkdown.com
· 2026-08-26
A technical guide argues that most retrieval-augmented generation (RAG) implementations are over-engineered, with teams reaching for embeddings, vector databases and reranking before checking if simpler tools suffice. It lays out decision factors—data freshness, corpus size and churn, query patterns, scale and team ML skill—to determine which retrieval approach fits, starting with plain full-text search like BM25, Elasticsearch or Postgres search before adding complexity.
lighthousenewsletter.com
· 2026-08-26
GoKawiil