Curated Open-Source AI Repositories

“It is the first ASF-incubating agent workspace to nail local-first: append-only log of model messages/tool calls, one Runtime Host for Desktop/TUI/eval. +1,978 stars last week and +3,848 total proves teams want recoverable execution facts, not just chat. Rating: 4.7/5. When NOT to use: Skip if you need cloud multi-tenant now — Maka is macOS-first, Windows unsigned preview, no Linux yet, incubation WIP.”

“My Junagadh edge boxes run Pi5/4070 mixes — llmfit replaces guesswork with real fit + tok/s estimates across Q2_K→Q8_0. v1.1.12 now PRs measured tok/s back, which is why it held 34.5k stars and Repo Radar #2 Aug 28. Rating: 4.8/5. When NOT to use: Not for pure cloud API routing — llmfit shines for local/self-hosted sizing, not for OpenAI-only bill tracking.”

“Claude Code → Codex handoffs break for me; ai-memory fixes it with one binary, no vector DB, SQLite + Markdown versioned in git. v1.32.2 trending Aug 27 at 5,055 stars validates the file-not-vector approach for Gujarat SMEs on 4G. Rating: 4.6/5. When NOT to use: Avoid if you need semantic chunk retrieval at scale — this is file-level handoff memory, not pgvector GraphRAG.”

“I scan every MCP skill before prod — AI-Infra-Guard is the only guard that knows 146 AI components and 2K CVEs plus PAIR/GOAT jailbreaks. v4.6.0 Aug 26 is why it jumped to 6,032 stars on Repo Radar. Rating: 4.7/5. When NOT to use: Not for generic SAST — it is AI-infra specific (MCP, skills, agent supply chain), not your monolith lint.”

“My hybrid router keeps 80% on SLM Pi5; semantic-router is the production-grade version — neural domain classifiers, multi-backend. 5,376 stars Go, Apache-2.0, and the missing layer under Maka/llmfit is why it trended Aug 28. Rating: 4.5/5. When NOT to use: Skip if you route by static rules — gains appear when prompt intent actually needs classifier, not simple keyword.”

“After skills went viral (218k+ stars for skills repo), Anthropic shipped the official marketplace — immutable slugs, renames map, strict skill-bundle plugins. 35.3k stars + 3,438 commits is the canonical registry I wire in Curro. Rating: 4.9/5. When NOT to use: Not for non-Claude harnesses — it is Claude Code plugin directory; wire via MCP if you need cross-harness.”

“Most skill packs are coding-centric; this one targets science workflows — hypothesis, experiment, paper — which is where Gujarat research SMEs actually need agents. Trending #2 on GitHub today alongside archify proves niche skills are winning. Rating: 4.4/5. When NOT to use: Avoid if you only need web-dev personas — scientific skills add overhead without bench tasks.”

“I render blueprints for every library skill — archify gives me 38 diagram types as self-contained HTML/SVG that survive export, unlike Mermaid. Trending #1 today at 3,927 today stars over Ghidra tells you visual architecture is the bottleneck. Rating: 4.6/5. When NOT to use: Not for interactive dashboards — archify is static HTML/SVG for docs, not live Recharts.”

“The 92.5k-star fix for “every session starts from scratch” — claude-mem captures everything, compresses with AI and re-injects per agent. Works with Claude Code/OpenClaw/Codex/Gemini/Hermes/Copilot/OpenCode, which is why it leads AI memory trending at +218 today. Rating: 4.8/5. When NOT to use: Not if you already use file-based handoff (ai-memory) — claude-mem is session-capture heavy, heavier than git Markdown.”

“I run agent swarms for Gujarat SMEs — OpenMAIC showed me how to orchestrate 30+ classroom agents with shared memory and turn-taking. Its MIT classroom loop is the cleanest multi-agent handoff I have seen for education + meeting simulation.”

“I build offline-first agents for Junagadh where 4G drops — minimind lets me pre-train a 115M model on a ₹6K VPS and run 62 tok/s on Pi 5. Its from-scratch recipe demystifies tokenization and MoE better than any course.”

“I repurpose journal → video → podcast — notebooklm-py gives my n8n workflow direct NotebookLM source ingestion and audio overview generation without browser clicks. Its skill wrapper is how I auto-test MCP handshakes.”

“I run P95 42ms HNSW + 90-day OTel ledger from Junagadh where every token counts — context-mode cut my Claude Code context by 98% sandboxed and stopped re-reads after /compact. Its MCP hook routing across 17 platforms is the missing persistence layer I built manually for Gujarat SMEs on Pi 5 62 tok/s.”

“I ship Browser-Use automations for Surat RFQ inboxes — camofox gave me Cloudflare-bypass without rewriting Playwright scripts. Its stealth headless drop-in is how I keep vendor portal scrapes alive on 4G from Junagadh when sites block headless.”

“I run n8n + MCP workflows on ₹6K VPS — Lightpanda replaces Chromium bloat with Zig speed and tiny RAM, so my Junagadh ledger stays P95 780ms even with 500 browser calls. It is the browser I recommend when Playwright OOMs on Pi 5.”

“I run SEO & AEO Services where Google AI Overviews need cited answers — marketingskills gave my Claude Code an AEO checklist that maps to FAQPage + JSON-LD I ship for Gujarat SMEs. Its CRO/SEO skills turn my journal posts into 120% more clicks when cited.”