The Git Times

Supabase, the flagship repository from supabase/supabase, has amassed 98,344 GitHub stars since its 2019 launch. Written primarily in TypeScript, it positions itself as the Postgres development platform, replicating Firebase's developer experience through enterprise-grade open source tools.

At its core, Supabase provides a dedicated Postgres database with extensions like PostGIS for geospatial data and pgvector for vector embeddings. Builders get auto-generated APIs via PostgREST for RESTful access and GraphQL support. Realtime subscriptions leverage an Elixir server polling Postgres replication over WebSockets, enabling live updates on inserts, updates, and deletes.

Authentication handles OAuth2 flows, while storage manages files. Edge Functions and database functions extend serverless capabilities. For AI workflows, it includes a vectors and embeddings toolkit. The platform supports both hosted deployment—sign up at supabase.com—and self-hosting for local development.

Supabase's architecture favors existing MIT/Apache-licensed tools, building new ones only when needed. Postgres anchors everything, drawing on 30+ years of reliability. Unlike Firebase's proprietary stack, this setup ensures data sovereignty and avoids vendor lock-in.

Recent activity underscores its momentum. The latest release, v1.26.02 (pushed March 2026), follows key updates:

• Supabase PrivateLink: Routes AWS traffic via VPC Lattice, eliminating public internet exposure for secure database connections. Details.
• Postgres Best Practices for AI Agents: 30 rules across eight categories to guide tools like Claude, Cursor, and GitHub Copilot in generating correct SQL.
• Ethereum Querying: SQL wrapper for real-time blockchain data via Infura FDW. Docs.
• Claude Connector: Official integration lets builders manage Supabase projects conversationally.

With topics spanning ai, realtime, vectors, and nextjs examples, Supabase targets full-stack apps. Community channels include a forum for best practices, GitHub issues for bugs, Discord for sharing, and email for infra issues. Documentation lives at supabase.com/docs.

This evolution matters for builders ditching Firebase: Supabase delivers comparable speed with Postgres's robustness, now enhanced for AI and enterprise security.

(Word count: 378)

Developers building spatial AI applications have a reliable workhorse in Kornia, a Python library layered on PyTorch. Launched in 2018, it now boasts 11,097 GitHub stars and continuous updates, with the latest push in March 2026 and release v0.8.2. Kornia delivers differentiable primitives for geometric computer vision, enabling seamless integration into deep learning pipelines via auto-differentiation, batch processing, and GPU acceleration.

At its core, Kornia offers differentiable image processing operators. These span filters like Gaussian, Sobel, and median; transformations including affine, homography, and perspective; enhancements such as histogram equalization and CLAHE; and edge detectors like Canny and Laplacian. Over 500 operations in total make it a one-stop shop for low-level vision tasks that backpropagate gradients directly.

Advanced augmentations stand out for training robust models. Builders can chain operations via AugmentationSequential, PatchSequential, or VideoSequential. Policies like AutoAugment, RandAugment, and TrivialAugment automate complexity, boosting data efficiency in computer vision workflows.

The library also bundles pre-trained AI models for production tasks:

• Face detection with YuNet
• Feature matching via LoFTR and LightGlue
• Descriptors including DISK, DeDoDe, and SOLD2
• Segmentation with SAM
• Classification using MobileViT and VisionTransformer

This matters because Kornia bridges classical vision algorithms with modern deep learning. Unlike rigid OpenCV bindings, its PyTorch-native design lets gradients flow through entire pipelines—crucial for tasks in robotics, image synthesis, and spatial AI.

Noteworthy now: Kornia is pivoting to end-to-end vision models, integrating Vision Language Models (VLMs) and Vision Language Agents (VLAs). Recent v0.8.2 release notes highlight maintenance rigor: dependency bumps (e.g., pytest to 8.3.5), documentation enhancements with SEO meta descriptions and docstrings, a Discord migration for community, and fixes like SOLD2Detector link corrections. Pre-commit CI ensures code quality across multiple PRs. These signal sustained momentum amid the shift, equipping builders for multimodal vision without library fragmentation.

For PyTorch users, Kornia reduces boilerplate, accelerates prototyping, and scales to enterprise-grade spatial reasoning.

(Word count: 362)

Developers and security builders now have a single, automated source for the latest CVEs paired with public proofs-of-concept (PoCs). The trickest/cve repository, launched in January 2022 and last updated in March 2026, has amassed 7,573 stars. It systematically collects CVE details from the official cvelist, then hunts for exploitable PoCs across the web.

The workflow splits CVEs by year into markdown files, such as those for 2022's high-profile bugs like CVE-2022-0001, CVE-2022-25297, CVE-2022-1388, and CVE-2022-25451. For each, it employs two PoC discovery methods:

• Reference scanning: Pulls CVE references and uses ffuf with a regex pattern (?i)[^a-z0-9]+(poc|proof of concept|proof[-_]of[-_]concept)[^a-z0-9]+ to flag likely PoC links. It also integrates HackerOne reports via AllVideoPocsFromHackerOne.
• GitHub search: Queries repositories mentioning the CVE ID using find-gh-poc.

Results merge automatically without overwriting manual commits, filtered by a blacklist.txt to cut false positives. Shields.io generates version-specific badges for affected software. An atom feed enables product-specific monitoring, and a template script builds searchable HTML tables.

This matters for builders because it centralizes scattered PoCs, saving hours of manual GitHub trawling or NVD browsing. Red teams and pentesters can "watch" the repo for real-time notifications on new exploits. Product owners scan for vulnerabilities in their stack—search by software name or version to reveal all public PoCs.

What sets it apart now? Amid rising supply-chain attacks, its automation ensures freshness: daily pulls from cvelist and GitHub keep it ahead of fragmented CVE trackers. With topics spanning cve-poc, exploit, pentesting, and red-team, it's tailored for infosec workflows. Nearly all content generates via Trickest's pipeline, refined for accuracy, though users should verify PoCs independently.

The full history lives in year-sorted directories; hottest CVEs highlight front-page risks. For teams building secure software, this repo signals where attackers strike first.

(Word count: 348)

Developers seeking control over their notes have a robust option in memos, the open-source, self-hosted service from usememos/memos. Launched in December 2021, it has amassed 57,374 GitHub stars and remains actively maintained, with the latest push on March 1, 2026, and release v0.26.2 addressing key stability issues.

At its core, memos functions as a privacy-first knowledge base for personal notes, team wikis, and memos-style microblogging. Data stores as plain text Markdown, ensuring portability without vendor lock-in. Built as a single Go binary paired with a React frontend, it boasts a low memory footprint and starts in seconds—ideal for resource-constrained environments.

Key technical strengths include:

• Deployment simplicity: A one-line Docker command (docker run -d --name memos -p 5230:5230 -v ~/.memos:/var/opt/memos neosmemo/memos:stable) gets it running on http://localhost:5230. Supports SQLite (default), MySQL, and PostgreSQL; Docker Compose, Kubernetes Helm charts, and pre-built binaries for Linux/macOS/Windows are available.
• APIs for integration: Full REST and gRPC endpoints enable embedding into workflows.
• Minimalist UI: Dark mode, mobile-responsive, with Markdown editing and task lists.

The v0.26.2 release underscores its production readiness. Fixes tackle authentication persistence across tabs via localStorage, spurious logouts on token expiry, and SSRF vulnerabilities in webhook dispatchers. UI refinements include calendar navigation improvements, task list scoping, attachment handling, and compound tag support with ampersands. Backend tweaks ensure delete operations only trigger post-actions on success, while eliminating redundant fetches in editors.

This maturation matters now as builders push back against proprietary tools like Notion or Evernote, which impose tracking, ads, or subscriptions. Memos delivers zero telemetry, self-hosting on your infrastructure, and FOSS extensibility—tagged for Docker, Go, Markdown, self-hosted setups, and SQLite. With topics spanning note-taking to social-network primitives, it suits solo devs to teams building custom knowledge systems.

Contributions remain open: bug reports, PRs, and docs updates are encouraged, reflecting a healthy community.

(Word count: 348)

OpenC3 COSMOS is a Ruby framework for commanding and monitoring embedded systems, dubbed "targets." It handles communication over TCP/IP, UDP, serial, and other protocols, providing a full suite of tools for telemetry display, graphing, operational scripting, command sending, logging, and log playback. Developers configure interfaces to their hardware, then access a ready-to-use dashboard for system status, raw packet inspection, and connection management.

At its core, the Command and Telemetry Server oversees target connections. The Limits Monitor tracks telemetry points exceeding thresholds, logging violations even after recovery. Command Sender offers dropdown-driven interfaces for one-off packets, complete with parameter descriptions. Script Runner executes Ruby test procedures in a graphical stepper, with pause/stop controls and telemetry checks.

The project shines in automated testing of embedded devices, from spacecraft components to IoT gadgets and cell phones. With 206 GitHub stars and topics spanning automation, hardware integration, and scripting, it targets builders needing robust control loops without custom infrastructure.

Version 7.0.0-rc2, released recently, marks a pivotal upgrade. The headline addition is a QuestDB-powered time series database (TSDB), slashing data retrieval times by an order of magnitude. This eliminates reducer microservices, unlocks SQL querying via QuestDB console, and paves the way for new integrations. RC2 refines this with SCOPE-prefixed TSDB table names (requiring openc3.sh cleanup from RC1, which drops logged data), a TEMPLATE_BASE64 keyword, admin themes, and fixes for TSDB, STRUCTURE keyword, Python microservices, and XTCE support.

COSMOS also shifts from AGPL to the COSMOS Builder's License, easing commercial adaptations while preserving open contributions. Last pushed March 1, 2026, it remains actively maintained since its 2022 launch.

For developers, COSMOS matters now as v7 stabilizes high-performance telemetry at RC stage. QuestDB's efficiency addresses scaling pain points in hardware testbeds, where historical data queries often bottleneck workflows. Production users should await final release, but testers gain immediate value in prototyping command-control pipelines.

(Word count: 378)