Curing AI

Quit Claude Code mid-task, open Codex in the same folder, and it picks up where you left off: what failed, what's still open. ai-memory gives 20+ coding agents one shared memory, stored as plain markdown on a server you run. github.com/akitaonrails/ai-m…

Assign an issue to Claude Code, Codex or Cursor the way you'd assign a teammate. Multica has them pick it up, post progress, flag blockers and hand back a PR for review. Self-hostable, and the agents run on your own machine. github.com/multica-ai/multic…

Keep your budget on your own server. ezBookkeeping runs on a Raspberry Pi with one Docker command and imports CSV, OFX, QIF, GnuCash, Firefly III and Beancount. It even reads receipt photos, and has MCP so your AI can query your spending. github.com/mayswind/ezbookke…

Drop in your PDFs, slides, spreadsheets and scanned docs, and RAGFlow turns them into answers with traceable citations back to the source chunk. Open source, self-hostable with Docker, and you can see and fix how every doc got chunked. github.com/infiniflow/ragflo…

Connect your laptop, home server and cloud VMs over WireGuard without opening a single port. NetBird is a Tailscale-style mesh you can self-host on a 1 CPU, 2 GB VM. Access policies, SSO and private DNS all live in one dashboard. github.com/netbirdio/netbird

Make videos the way you build web apps: React components, a timeline in code, rendered to MP4. Remotion has agent skills, so your coding agent can draft the video and you fine-tune it in the Studio. github.com/remotion-dev/remo…

Run your own email server without gluing Postfix, Dovecot and a spam filter together. Stalwart is one Rust server for mail, calendars, contacts and files. Spam filtering, DKIM/DMARC and a web admin are built in. github.com/stalwartlabs/stal…

TeamViewer, but open source, and you can run the relay server yourself. RustDesk works out of the box with zero config, and if you don't want your sessions going through anyone else's servers, you point it at your own. github.com/rustdesk/rustdesk

Schedule posts to X, LinkedIn, Instagram, TikTok, Bluesky, Reddit and more from one calendar, on your own server. Postiz is free to self-host, has an API for n8n/Zapier, and AI post generation if you bring your own key. github.com/gitroomhq/postiz-…

Perplexity, but self-hosted. One docker run gets you an AI answer engine with cited sources, private SearxNG search, and Ollama or cloud models. github.com/ItzCrazyKns/Vane

Run Claude Code, Codex and Pi from one app, then pick up the same session from your phone or another computer. The client is open source and uses the subscriptions or API keys you already pay for, or a local model. github.com/makecindy/cindy

Stack the free tiers of 34 LLM providers (Groq, Cerebras, Mistral, OpenRouter...) behind one OpenAI-compatible /v1 endpoint. Roughly 7.4B tokens a month. When one rate-limits you it fails over, so Claude Code or Aider keep going. github.com/tashfeenahmed/fre…

Write CUDA kernels in plain Rust, straight from NVIDIA. No DSL, no C++ bindings, and Rust's ownership rules follow your buffers onto the GPU. Pick thread-level SIMT or NumPy-style tiles. Still an early research project, so APIs will break. github.com/NVIDIA/cuda-rust

tmux, but built for coding agents. Every pane shows whether Claude Code, Codex or Cursor is working, blocked or idle, so you stop hunting for the stuck one. Drop SSH or close the laptop and they keep running. One Rust binary. github.com/herdrdev/herdr

Microsoft open-sourced an app that watches you do a task once, then writes it up as a SKILL.md your agent can repeat. Record your screen, Copilot turns it into steps, and it prefers real tools like gh over replaying your clicks. github.com/microsoft/skill-r…

Give Claude Code or Codex a /humanizer command that strips the tells of AI writing, built on Wikipedia's own "Signs of AI writing" guide. It catches "It's not just X, it's Y" and "Let that sink in", then rewrites without touching your facts. github.com/blader/humanizer

Paying for Loom just to send a 2 minute bug repro? Cap does the same thing and it's open source. - Instant Mode uploads while you record, so the share link is ready the second you stop - Studio Mode records locally and gives you an editor with zooms, backgrounds, trimming and captions - Bring your own S3 bucket (R2, B2, MinIO...) or self-host the whole thing with Docker Compose - AI titles, summaries, chapters and transcripts - Can import your old Loom videos Native app for Mac, Windows and Linux. Rust + TypeScript, AGPLv3, ~24k stars. github.com/CapSoftware/Cap

Got a gaming PC, an old laptop and a Mac mini sitting around? Mesh LLM turns them into one LLM server. It pools GPUs and memory across machines and gives you a single OpenAI-compatible API on localhost. - Model fits on one box? It runs there. Too big? It gets split into layer stages across peers - Requests route by model name to whichever node can serve it - Traffic between nodes is end-to-end encrypted over QUIC - Private mesh with invite tokens, or join the public one with one command - Works with Goose, OpenCode, Claude Code and Pi Rust, Apache-2.0, ~3.8k stars and climbing. github.com/Mesh-LLM/mesh-llm

NotebookLM, but you host it and pick the model. Open Notebook is a self-hosted research workspace: - throw in PDFs, web pages, audio and video, then chat with all of it with citations - 20+ providers (OpenAI, Anthropic, Gemini, Mistral) or fully local with Ollama - podcast generation with 1-4 speakers instead of the fixed 2 - full-text + vector search and a REST API - one docker compose up and it's running 41k stars, MIT. github.com/lfnovo/open-noteb…

Voice typing that never leaves your machine. Hex from anomalyco is a tiny local dictation app: - hold a shortcut, talk, release, text gets pasted wherever your cursor is - transcription runs on-device, nothing sent anywhere after the model download - Mac, Windows and Linux (brew, AUR, deb/rpm, Nix) - written in Rust, MIT licensed ~3.2k stars and climbing on GitHub Trending today. github.com/anomalyco/hex