{"id":51045170,"url":"https://github.com/chipi/agentic-ai-homelab","last_synced_at":"2026-06-22T13:01:09.972Z","repository":{"id":364326304,"uuid":"1267273198","full_name":"chipi/agentic-ai-homelab","owner":"chipi","description":"How I work with agentic AI: personal patterns, homelab infra (vLLM, Ollama, Grafana Alloy, LibreChat), project scaffolding. Not best practices - what I run.","archived":false,"fork":false,"pushed_at":"2026-06-20T20:08:07.000Z","size":399,"stargazers_count":0,"open_issues_count":0,"forks_count":0,"subscribers_count":0,"default_branch":"main","last_synced_at":"2026-06-20T20:16:52.684Z","etag":null,"topics":["agentic-ai","claude-code","grafana-alloy","homelab","librechat","mcp","ollama","opencode","self-hosted","vllm"],"latest_commit_sha":null,"homepage":null,"language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"mit","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/chipi.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":null,"maintainers":null,"copyright":null,"agents":"AGENTS.md","dco":null,"cla":null}},"created_at":"2026-06-12T11:40:19.000Z","updated_at":"2026-06-20T20:08:11.000Z","dependencies_parsed_at":null,"dependency_job_id":null,"html_url":"https://github.com/chipi/agentic-ai-homelab","commit_stats":null,"previous_names":["chipi/agentic-ai-homelab"],"tags_count":0,"template":false,"template_full_name":null,"purl":"pkg:github/chipi/agentic-ai-homelab","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/chipi%2Fagentic-ai-homelab","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/chipi%2Fagentic-ai-homelab/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/chipi%2Fagentic-ai-homelab/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/chipi%2Fagentic-ai-homelab/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/chipi","download_url":"https://codeload.github.com/chipi/agentic-ai-homelab/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/chipi%2Fagentic-ai-homelab/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":34649822,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-05-26T15:22:16.424Z","status":"online","status_checked_at":"2026-06-22T02:00:06.391Z","response_time":106,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["agentic-ai","claude-code","grafana-alloy","homelab","librechat","mcp","ollama","opencode","self-hosted","vllm"],"created_at":"2026-06-22T13:01:08.822Z","updated_at":"2026-06-22T13:01:09.967Z","avatar_url":"https://github.com/chipi.png","language":"Python","funding_links":[],"categories":[],"sub_categories":[],"readme":"# agentic-ai-homelab\n\nHow I work with agentic AI, end-to-end: cloud LLM APIs, local self-hosted\nmodels, project scaffolding, and the connective tissue between them.\n\nNot a tutorial. Not best practices. **This is what I run.** Patterns here\nhave been used in real projects; configs here are the actual templates I\ndeploy (with secrets stripped). Take what's useful, leave the rest.\n\n## Status\n\n**v0.2 — all four pillars are real.** Templates bootstrap; infra\ncomposes deploy; cloud workflow examples run; agent harness configs\ndrop in. Plus a top-level `provider-bakeoff/` that compares 10 LLM\nproviders across 4 countries on the same tasks with real cost numbers.\n\nWhat's still open: ops items that need live access (Grafana Cloud\ncreds, image-tag pinning post-boot), Claude Code `settings.json`\nextraction, and existing project AGENTS.md dedup. See\n[`docs/wip/NEXT_STEPS.md`](docs/wip/NEXT_STEPS.md).\n\nContinuity log: [`docs/history/`](docs/history/) — `0001-genesis.md`\nis the founding session, `0003-v0.2-arc.md` is the arc that brought\nthe four pillars to real.\n\n## What's here\n\n```\n.\n├── AGENTS.md                 — repo-scoped agent rules (harness-agnostic)\n├── CLAUDE.md                 — Claude Code entry; imports AGENTS.md so every harness picks up the same rules\n├── LICENSE                   — MIT\n├── docs/                     — narrative + recipes + history (MkDocs site)\n│   ├── philosophy.md         — the underlying \"how I work\" 1-pager\n│   ├── project-setup.md      — pillar 1: scaffolding for new projects\n│   ├── local-ai-infra.md     — pillar 2: self-hosted LLM stack (vLLM + Ollama + observability)\n│   ├── cloud-ai-workflow.md  — pillar 3: working with Claude / OpenAI / Gemini APIs\n│   ├── agent-harnesses.md    — pillar 4: opencode / Claude Code / MCP registry\n│   ├── recipes/              — operator-facing recipes (DGX dashboard, GPU mode-swap, ...)\n│   ├── adr/                  — architecture decision records for this repo\n│   ├── rfc/                  — open proposals not yet decided\n│   ├── wip/                  — work-in-progress notes, next-steps plans\n│   └── history/              — session-by-session continuity log\n├── templates/\n│   ├── new-project/          — copy this dir to bootstrap a new repo\n│   └── opencode/             — drop into ~/.config/opencode/\n├── infra/\n│   ├── vllm/                 — hardened vLLM compose template (Qwen3-Coder-Next-FP8)\n│   ├── observability/        — Grafana Alloy + DCGM + cAdvisor + Ollama metrics\n│   └── dgx/bin/              — DGX-host operator scripts (gpu-mode-swap.sh)\n├── examples/                 — small concrete code samples (cloud + local)\n│   ├── claude-api-with-caching/  — prompt-cache discipline demo\n│   ├── mcp-tool-template/        — FastMCP skeleton with 3 tool shapes\n│   └── eval-harness/             — provider-agnostic 4-op harness\n└── provider-bakeoff/         — top-level lift-able: 10 providers × 2 tasks\n```\n\n## Four pillars\n\n1. **Project setup** — how I scaffold a new repo. AGENTS.md template, ADR\n   convention, docs/wip/ pattern, layered Makefile gates (`docs-*` /\n   `lint` / `test-*` / `ci-fast` / `ci`), PR template, GH Pages docs site\n   workflow, pre-commit baseline.\n   → [Pillar 1](docs/project-setup.md) · `templates/new-project/`\n2. **Local AI infra** — DGX homelab stack: hardened vLLM compose, Ollama\n   for catalog + smaller models, observability layer (Alloy → Grafana\n   Cloud). Three recipes (terminal dashboard, GPU mode-swap,\n   observability boot) cover daily operation.\n   → [Pillar 2](docs/local-ai-infra.md) · `infra/vllm/{template,coder-next,openwebui}/` + `infra/observability/`\n3. **Cloud AI workflow** — patterns for Claude API, OpenAI, Gemini:\n   prompt caching as discipline, batch API, multi-provider routing, eval\n   harnesses, cost gates.\n   → [Pillar 3](docs/cloud-ai-workflow.md) · `examples/claude-api-with-caching/`,\n   `examples/mcp-tool-template/`, `examples/eval-harness/`\n4. **Agent harnesses** — the connective tissue. Global AGENTS.md,\n   opencode provider config, MCP server registry pattern. Two recipes\n   (lean-ctx + RTK token management, Chrome DevTools MCP agent loop)\n   cover the daily-driver tooling.\n   → [Pillar 4](docs/agent-harnesses.md) · `templates/opencode/`\n\n## Plus: `provider-bakeoff/`\n\nA self-contained mini-project (designed to lift into its own GH repo)\nthat compares **10 LLM providers across 4 countries** — Claude, GPT,\nGemini, Grok (🇺🇸) · Mistral, Hugging Face (🇫🇷) · DeepSeek, Qwen,\nKimi (🇨🇳) · local vLLM (🏠) — on the same tasks, with real cost\nnumbers, in one command.\n\n```bash\ncd provider-bakeoff\ncp .env.example .env       # fill in whichever keys you have\nmake bakeoff               # ~$2-5 for a full sweep\n```\n\n## Who this is for\n\nMe, primarily. Future-me when I forget why I made a decision. Other\nengineers running similar setups who want a working reference instead of a\nblog post.\n\n## How to use\n\n- **Reading**: start with [`docs/philosophy.md`](docs/philosophy.md), then the pillar that maps to\n  your need.\n- **Bootstrapping a new project**: `cp -r templates/new-project/ ~/Projects/\u003cname\u003e/`\n  then sed-substitute the placeholders (`\u003cproject-name\u003e`, `\u003cowner\u003e`,\n  `\u003cproject-description\u003e`).\n- **Setting up the homelab stack**: start with [`docs/recipes/observability-boot.md`](docs/recipes/observability-boot.md)\n  (lightest deploy, zero ACL change) → then [`infra/vllm/`](infra/vllm/README.md) (template + the operator's two deploys)\n  → then the operator tools per [`docs/recipes/dgx-terminal-dashboard.md`](docs/recipes/dgx-terminal-dashboard.md).\n- **Comparing LLM providers**: `provider-bakeoff/` (see above).\n- **Resuming after a break**: [`docs/history/`](docs/history/) is the\n  continuity log. Latest entry has what's current.\n\n## What this is not\n\n- Not a curated \"best practices\" guide. Patterns rot. What's here works for me\n  right now; if it stops working, I update it or delete it.\n- Not a generic infrastructure-as-code repo. It assumes a single-operator\n  homelab with one DGX-class box and one tailnet.\n- Not a learning resource. If you don't already know what vLLM, MCP, or\n  Tailscale are, this won't teach you.\n\n## License\n\nMIT — see [LICENSE](LICENSE). Same license I use across my projects.\n\n## Repo conventions\n\nSee [`AGENTS.md`](AGENTS.md) for the repo-scoped rules that govern work\nin this repo (harness-agnostic; Claude Code picks them up via\n[`CLAUDE.md`](CLAUDE.md)). For decisions that shape the repo itself,\nsee [`docs/adr/`](docs/adr/).\n\nSite: \u003chttps://chipi.github.io/agentic-ai-homelab/\u003e.\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fchipi%2Fagentic-ai-homelab","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fchipi%2Fagentic-ai-homelab","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fchipi%2Fagentic-ai-homelab/lists"}