Show HN: Linggen – A local-first memory layer for your AI (Cursor, Zed, Claude) https://ift.tt/uIb37Xq

Show HN: Linggen – A local-first memory layer for your AI (Cursor, Zed, Claude) Hi HN, Working with multiple projects, I got tired of re-explaining our complex multi-node system to LLMs. Documentation helped, but plain text is hard to search without indexing and doesn't work across projects. I built Linggen to solve this. My Workflow: I use the Linggen VS Code extension to "init my day." It calls the Linggen MCP to load memory instantly. Linggen indexes all my docs like it’s remembering them—it is awesome. One click loads the full architectural context, removing the "cold start" problem. The Tech: Local-First: Rust + LanceDB. Code and embeddings stay on your machine. No accounts required. Team Memory: Index knowledge so teammates' LLMs get context automatically. Visual Map: See file dependencies and refactor "blast radius." MCP-Native: Supports Cursor, Zed, and Claude Desktop. Linggen saves me hours. I’d love to hear how you manage complex system context! Repo: https://ift.tt/KOI3Fmp Website: https://linggen.dev https://ift.tt/KOI3Fmp December 19, 2025 at 11:24PM

Show HN: TinyPDF – 3KB PDF library (70x smaller than jsPDF) https://ift.tt/IE40lMO

Show HN: TinyPDF – 3KB PDF library (70x smaller than jsPDF) I needed to generate invoices in a Node.js app. jsPDF is 229KB. I only needed text, rectangles, lines, and JPEG images. So I wrote tinypdf: <400 lines of TypeScript, zero dependencies, 3.3KB minified+gzipped. What it does: - Text (Helvetica, colors, alignment) - Rectangles and lines - JPEG images - Multiple pages, custom sizes What it doesn't do: - Custom fonts, PNG/SVG, forms, encryption, HTML-to-PDF That's it. The 95% use case for invoices, receipts, reports, tickets, and labels. GitHub: https://github.com/Lulzx/tinypdf npm: npm install tinypdf https://github.com/Lulzx/tinypdf December 19, 2025 at 12:29AM

Show HN: Explore Prometheus /metrics endpoints from your terminal https://ift.tt/wBApo73

Show HN: Explore Prometheus /metrics endpoints from your terminal https://ift.tt/69tSwyG December 18, 2025 at 11:40PM

Show HN: The feature gap "Chat with PDF" tuts and a regulated enterprise system https://ift.tt/AzOTbi2

Show HN: The feature gap "Chat with PDF" tuts and a regulated enterprise system I've spent the last few months architecting a RAG system for a regulated environment. I am not a developer by trade, but I approached this with a strict "systems engineering" and audit mindset. While most tutorials stop at "LangChain + VectorDB", I found that making this legally defensible and operationally stable required about 40+ additional components. We moved from a simple ingestion script to a "Multi-Lane Consensus Engine" (inspired by Six Sigma) because standard OCR/extraction was too hallucination-prone for our use case. We had to build extensive auditing, RBAC down to the document level, and a hybrid Graph+Vector retrieval to get acceptable accuracy The current architecture includes: Ingestion: 4 parallel extraction lanes (Vision, Layout, Text, Legal) with a Consensus Engine ("Solomon") that only indexes data confirmed by multiple sources Retrieval: Hybrid Neo4j (Graph) + ChromaDB (Vector) with Reciprocal Rank Fusion Performance: Semantic Caching (Redis) specifically for similar-meaning queries (40x speedup) Security: Full RBAC, Audit Logging of every prompt/retrieval, and PII masking. I documented the complete feature list and gap analysis https://gist.github.com/2dogsandanerd/2a3d54085b2daaccbb1125... My question to the community: Looking at this list – where is the line between "robust production engineering" and "over-engineering"? For those working in Fintech/Medtech RAG: what critical failure modes am I still missing in this list? https://gist.github.com/2dogsandanerd/2a3d54085b2daaccbb1125601945ceeb December 17, 2025 at 11:20PM

Show HN: Solving the ~95% legislative coverage gap using LLM's https://ift.tt/LhpmTdB

Show HN: Solving the ~95% legislative coverage gap using LLM's Hi HN, I'm Jacek, the solo founder behind this project (Lustra). The Problem: 95% of legislation goes unnoticed because raw legal texts are unreadable. Media coverage is optimized for outrage, not insight. The Solution. I built a digital public infrastructure that: 1. Ingests & Sterilizes: Parses raw bills (PDF/XML) from US & PL APIs. Uses LLMs (Vertex AI, temp=0, strict JSON) to strip political spin. 2. Civic Algorithm: The main feed isn't sorted by an editorial board. It's sorted by user votes ("Shadow Parliament"). What the community cares about rises to the top. 3. Civic Projects: An incubator for citizen legislation. Users submit drafts (like our Human Preservation Act ), which are vetted by AI scoring and displayed with visual parity alongside government bills. Tech Stack: Frontend: Flutter (Web & Mobile Monorepo), Backend: Firebase + Google Cloud Run, AI: Vertex AI (Gemini 2.5 Flash), License: PolyForm Noncommercial (Source Available). I am looking for contributors. I have the US and Poland live. EU, UK, FR, DE in pipeline, partially available. I need help building Data Adapters for other parliaments (the core logic is country-agnostic). If you want to help audit the code or add a country, check the repo. The goal is to complete the database as much as possible with current funding. Live App: https://lustra.news Repo: https://ift.tt/B6CuXES Dev Log: https://ift.tt/sGwciy6 https://lustra.news/ December 16, 2025 at 08:09PM

Show HN: Zenflow – orchestrate coding agents without "you're right" loops https://ift.tt/BrcHsRF

Show HN: Zenflow – orchestrate coding agents without "you're right" loops Hi HN, I’m Andrew, Founder of Zencoder. While building our IDE extensions and cloud agents, we ran into the same issue many of you likely face when using coding agents in complex repos: agents getting stuck in loops, apologizing, and wasting time. We tried to manage this with scripts, but juggling terminal windows and copy-paste prompting was painful. So we built Zenflow, a free desktop tool to orchestrate AI coding workflows. It handles the things we were missing in standard chat interfaces: Cross-Model Verification: You can have Codex review Claude’s code, or run them in parallel to see which model handles the specific context better. Parallel Execution: Run five different approaches on a backlog item simultaneously—mix "Human-in-the-Loop" for hard problems with "YOLO" runs for simple tasks. Dynamic Workflows: Configured via simple .md files. Agents can actually "rewire" the next steps of the workflow dynamically based on the problem at hand. Project list/kanban views across all workload What we learned building this To tune Zenflow, we ran 100+ experiments across public benchmarks (SWE-Bench-*, T-Bench) and private datasets. Two major takeaways that might interest this community: Benchmark Saturation: Models are becoming progressively overtrained on all versions of SWE-Bench (even Pro). We found public results are diverging significantly from performance on private datasets. If you are building workflows, you can't rely on public benches. The "Goldilocks" Workflow: In autonomous mode, heavy multi-step processes often multiply errors rather than fix them. Massive, complex prompt templates look good on paper but fail in practice. The most reliable setups landed in a narrow “Goldilocks” zone of just enough structure without over-orchestration. The app is free to use and supports Claude Code, Codex, Gemini, and Zencoder. We’ve been dogfooding this heavily, but I'd love to hear your thoughts on the default workflows and if they fit your mental model for agentic coding. Download: https://ift.tt/e2o0ucq YT flyby: https://www.youtube.com/watch?v=67Ai-klT-B8 https://ift.tt/e2o0ucq December 16, 2025 at 10:02PM

Show HN: Cordon – Reduce large log files to anomalous sections https://ift.tt/PkxpMJ9

Show HN: Cordon – Reduce large log files to anomalous sections Cordon uses transformer embeddings and density scoring to identify what's semantically unique in log files, filtering out repetitive noise. The core insight: a critical error repeated 1000x is "normal" (semantically dense). A strange one-off event is anomalous (semantically isolated). Outputs XML-tagged blocks with anomaly scores. Designed to reduce large logs as a form of pre-processing for LLM analysis. Architecture: https://ift.tt/gv8IX3w... Benchmark: https://ift.tt/cMh6GH7... Trade-offs: intentionally ignores repetitive patterns, uses percentile-based thresholds (relative, not absolute). https://ift.tt/LD4YcQz December 16, 2025 at 02:06AM