Self-taught SRE building small, sharp infrastructure tools for noisy systems: Kubernetes, MCP, local AI, and low-RAM Apple Silicon.
- KG-first SRE automation
- MCP supply-chain safety
- Local AI on Apple Silicon
- sre-ai-copilot — KG-first SRE intelligence layer: AlertManager → topology/deploy/pod-event enrichment → Discord incidents with blast radius, NATS impact, pod trail, owners, and daily digests.
- mcp-skills-vault — registry + supply-chain integrity gate for MCP servers. 112 pinned/vetted entries with sha512/sha256/Docker-digest pins, 4 advisory feeds, offline-first. Make MCP boring.
- mcp-trace — stdio flight recorder for MCP servers. Logs JSON-RPC metadata to local SQLite so you can see what a tool actually did.
- tweai — AI reply assistant for X (Twitter). Open-source Chrome extension that drafts replies in your voice — 8 personas, smart thread context, tweet translator. OpenAI, Grok, Gemini. BYOK, no subscription, nothing routed through a middleman.
Four repos built around one idea: a private local LLM daemon that knows what's on your screen.
| Repo | What it does |
|---|---|
| Froggy | Local LLM + screen OCR daemon for 8 GB Apple Silicon Macs. Memory management, sliding context window, private inference. Core of the ecosystem. |
| FroggyKit | Shared Swift package — FroggyClient IPC, used by froggy-mcp and froggy-sre. |
| froggy-mcp | MCP server — connects Claude Code to Froggy over a Unix socket. Cloud model, local eyes. |
| froggy-sre | SRE incident response agent — feed a Kubernetes alert, get back root cause + fix + risk in one MCP call. Routes LLM calls to Froggy first. |
Claude Code ←—MCP—→ froggy-mcp ←—socket—→ Froggy daemon (screen, OCR, local LLM, transcripts)
Claude Code ←—MCP—→ froggy-sre ←—socket (primary) / API (fallback)—→ incident analysis report
Archived router / IoT / malware-analysis repos from my network-security period. Kept for history; current work is SRE automation, MCP safety, and local AI.
In Frog We Trust 🐸



