Curated Open-Source AI Repositories

“I prefer local-first desktops for client demos with patchy internet in Gujarat. Running the harness on-device keeps P95 stable when the line drops. I test it against my Rs 6K VPS setup for parity.”

“I use this as a starter shelf for new Claude Code seats, then trim hard. The Postgres and GitHub MCP entries saved me an hour per setup. My rule: read before install, keep the analytics dashboard on.”

“I query this radar from my lab before adding anything to this library. Velocity plus the MCP-skill taxonomy beats raw trending for agent work. I wired its MCP server into Cursor for morning scans.”

“I self-host research stacks so client queries never leak to third parties. The trending and compare tools feed my brief pipeline directly. I run it with Hishel caching to keep re-crawls cheap.”

“I keep agent memory in git in my Junagadh lab, so git log on memory beats any dashboard. Branching a personality to test a hunch feels natural. I run it in-process with file and Postgres MCP servers attached.”

“I keep agent memory in git in my Junagadh lab so git log beats any dashboard. Ouroboros carrying identity plus durable memory across restarts is the exact continuity pattern I prototype for Curro, and I run it lab-only with file plus Postgres MCP servers attached for evals.”

“I run RAG for Junagadh client docs where plain vector search misses multi-hop answers across invoices plus specs. M-flow scoring evidence paths instead of chunk similarity matches how I debug retrieval misses in my lab, and its MCP-server mode plugs straight into my Cursor morning-scan pipeline.”

“I maintain a 41-skill Junagadh catalog where discovery across 11 scattered repos wastes an hour per seat setup. Skills-mcp collapsing 9000 skills into one searchable MCP call is the index shape I want in front of my library pages, and I run it with tight install approvals to block broad-permission skills.”

“I watch recall degrade past 10K memories in long-running Junagadh agents where stale facts crowd out live decisions. YantrikDB adding temporal decay plus contradiction detection on top of HNSW is the maintenance loop my memory layer lacks, and its 87ms p50 skill recall fits my Rs 6K VPS latency budget.”

“I wire Codex seats for Gujarat SME pilots from Junagadh, and reference plugins beat hand-rolled prompts for repeatable deploys. I copied the Notion capture pattern for a Rajkot client site docs setup and had capture running the same evening on the Rs 6K VPS. I keep the Figma plugin disabled by default — design tokens without a reviewer in the loop produced three off-brand landing drafts before I gated it.”

“I render launch videos for Surat D2C drops where 100 creative variants beat one perfect cut. Deterministic re-renders saved a Diwali campaign when the client changed pricing at midnight — same HTML, new numbers, re-render, no editor awake. I run renders on the VPS with P95 under 4 minutes per 30-second clip.”

“I feed Gujarat SME quotations, invoices, and spec sheets into pgvector from Junagadh, and MarkItDown sits at the front of that pipe. A Rajkot RFQ inbox with mixed PDF and XLS attachments went from manual copy-paste to indexed with 42ms P95 HNSW search. I still hand-check scanned Gujarati PDFs — OCR misses cost me one wrong line item in April.”

“I track token burn per run in my P95 ledger on fixed-price builds, so a 20-50 percent read saving is margin. I trialed Vix plan mode on a Laravel RFQ module in Junagadh and the stem-agent cache reuse across explore plus plan phases cut one review loop from $3.65 to $1.86. The 3,000-line-file case in their own benchmark still lags — I keep big-file refactors on my standard harness.”

“I maintain a 41-skill Junagadh catalog, and the official spec is what I measure third-party skills against. The document-skills section standardized how my Curro pipeline handles DOCX briefs for two Surat clients. I install with tight approvals after auditing trending skills for broad permission requests — the same gate I run before any skill touches payment keys.”

“I point Junagadh interns here before they touch my client stack — rebuilding one single-file RAG starter removes magic faster than any lecture. Two interns shipped MCP tools within a month after working through the retrieval examples. For client work I vendor exactly one pattern per build and pin model versions.”

“I run a small team from Junagadh, so a staged sprint in slash commands fits how we already ship. The QA-lead command caught a broken UPI retry path before a Surat launch — P95 held 780ms because the checklist forced a load pass. I trimmed the CEO persona on day one; strategy prompts without profit-and-loss context generated noise for my fixed-price quotes.”