[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"$fc0b0frs2soms":3,"$fanuq43nlrv5g":58},{"slug":4,"title":5,"body":6,"summary":7,"tags":8,"author":14,"cover_url":15,"published_at":16,"seo_title":17,"seo_description":18,"reading_minutes":19,"related":20},"choosing-embedding-model-rag-50-question-retrieval-test","Choosing an embedding model for RAG: a 50-question retrieval test on your own documents","\u003Cp>New embedding models keep arriving. Cohere released Embed 5 on 30 September 2026 in two tiers, Pro and Fast, and open models you can run through Ollama keep improving. Leaderboards rank them on public datasets, but your documents are not a public dataset: your jargon, your table-heavy PDFs and your short, vague user questions are what decide whether retrieval works. The fix is cheap. Write 50 questions with known answers, score each candidate model on them, and choose with a rule you set before looking at results.\u003C\u002Fp>\n\u003Cfigure data-post-media=\"6b0280764b6fd09d500518dd\">\u003Cvideo src=\"https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6b\u002F6b0280764b6fd09d500518e3-0-09499bab.mp4\" autoplay muted loop playsinline preload=\"metadata\">\u003C\u002Fvideo>\u003C\u002Ffigure>\n\u003Cp>This post gives you the whole harness: a question set format, one Python script that scores recall@5 for a hosted model and a local one, and a decision rule. It needs Python 3, \u003Ccode>numpy\u003C\u002Fcode> and \u003Ccode>requests\u003C\u002Fcode>, a running Ollama, and optionally a Cohere API key.\u003C\u002Fp>\n\u003Ch2>What you are measuring\u003C\u002Fh2>\n\u003Cfigure data-post-media=\"6b0280134b6fd09d50051863\" data-layout=\"right\">\u003Cimg src=\"https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6b\u002F6b0280134b6fd09d50051868-0-9ac36f72.png\" alt=\"Evaluating how effectively a model retrieves the correct information chunk.\" loading=\"lazy\">\u003Cfigcaption>Evaluating how effectively a model retrieves the correct information chunk.\u003C\u002Ffigcaption>\u003C\u002Ffigure>\n\u003Cp>Retrieval quality for RAG is mostly one question: when the answer exists in your corpus, does the right chunk land in the top few results? That is recall@k. With k = 5 it asks whether the correct chunk appears among the five nearest vectors. We also track latency per query and an estimated cost, because a model that is 2 points better but ten times slower may not be better for you.\u003C\u002Fp>\n\u003Ch2>Step 1: chunk your documents once\u003C\u002Fh2>\n\u003Cp>Use the same chunks for every model, otherwise you are testing your chunker. Write them to \u003Ccode>chunks.jsonl\u003C\u002Fcode>, one JSON object per line:\u003C\u002Fp>\n\u003Cpre class=\"code-block\" data-lang=\"json\">\u003Ccode class=\"hljs language-json\">\u003Cspan class=\"hljs-punctuation\">{\u003C\u002Fspan>\u003Cspan class=\"hljs-attr\">&quot;id&quot;\u003C\u002Fspan>\u003Cspan class=\"hljs-punctuation\">:\u003C\u002Fspan> \u003Cspan class=\"hljs-string\">&quot;refund-policy-03&quot;\u003C\u002Fspan>\u003Cspan class=\"hljs-punctuation\">,\u003C\u002Fspan> \u003Cspan class=\"hljs-attr\">&quot;text&quot;\u003C\u002Fspan>\u003Cspan class=\"hljs-punctuation\">:\u003C\u002Fspan> \u003Cspan class=\"hljs-string\">&quot;Refunds are issued to the original payment method within 5 business days...&quot;\u003C\u002Fspan>\u003Cspan class=\"hljs-punctuation\">}\u003C\u002Fspan>\n\u003Cspan class=\"hljs-punctuation\">{\u003C\u002Fspan>\u003Cspan class=\"hljs-attr\">&quot;id&quot;\u003C\u002Fspan>\u003Cspan class=\"hljs-punctuation\">:\u003C\u002Fspan> \u003Cspan class=\"hljs-string\">&quot;api-limits-01&quot;\u003C\u002Fspan>\u003Cspan class=\"hljs-punctuation\">,\u003C\u002Fspan> \u003Cspan class=\"hljs-attr\">&quot;text&quot;\u003C\u002Fspan>\u003Cspan class=\"hljs-punctuation\">:\u003C\u002Fspan> \u003Cspan class=\"hljs-string\">&quot;The default rate limit is 60 requests per minute per key...&quot;\u003C\u002Fspan>\u003Cspan class=\"hljs-punctuation\">}\u003C\u002Fspan>\u003C\u002Fcode>\u003C\u002Fpre>\n\u003Cp>Chunks of a few hundred tokens are a sensible start. Keep the chunk ids stable and human-readable, since you will label questions against them.\u003C\u002Fp>\n\u003Ch2>Step 2: write 50 questions\u003C\u002Fh2>\n\u003Cfigure data-post-media=\"6b0280134b6fd09d5005186d\" data-layout=\"left\">\u003Cimg src=\"https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6b\u002F6b0280134b6fd09d50051872-0-d089314b.png\" alt=\"Creating a curated set of gold-standard questions and answers.\" loading=\"lazy\">\u003Cfigcaption>Creating a curated set of gold-standard questions and answers.\u003C\u002Ffigcaption>\u003C\u002Ffigure>\n\u003Cp>Write questions the way users actually ask them, not copied from the chunk text. Mine support tickets or search logs if you have them. For each, record the id of the chunk that contains the answer in \u003Ccode>questions.jsonl\u003C\u002Fcode>:\u003C\u002Fp>\n\u003Cpre class=\"code-block\" data-lang=\"json\">\u003Ccode class=\"hljs language-json\">\u003Cspan class=\"hljs-punctuation\">{\u003C\u002Fspan>\u003Cspan class=\"hljs-attr\">&quot;question&quot;\u003C\u002Fspan>\u003Cspan class=\"hljs-punctuation\">:\u003C\u002Fspan> \u003Cspan class=\"hljs-string\">&quot;how long until I get my money back?&quot;\u003C\u002Fspan>\u003Cspan class=\"hljs-punctuation\">,\u003C\u002Fspan> \u003Cspan class=\"hljs-attr\">&quot;gold&quot;\u003C\u002Fspan>\u003Cspan class=\"hljs-punctuation\">:\u003C\u002Fspan> \u003Cspan class=\"hljs-string\">&quot;refund-policy-03&quot;\u003C\u002Fspan>\u003Cspan class=\"hljs-punctuation\">}\u003C\u002Fspan>\n\u003Cspan class=\"hljs-punctuation\">{\u003C\u002Fspan>\u003Cspan class=\"hljs-attr\">&quot;question&quot;\u003C\u002Fspan>\u003Cspan class=\"hljs-punctuation\">:\u003C\u002Fspan> \u003Cspan class=\"hljs-string\">&quot;what happens if I send too many requests&quot;\u003C\u002Fspan>\u003Cspan class=\"hljs-punctuation\">,\u003C\u002Fspan> \u003Cspan class=\"hljs-attr\">&quot;gold&quot;\u003C\u002Fspan>\u003Cspan class=\"hljs-punctuation\">:\u003C\u002Fspan> \u003Cspan class=\"hljs-string\">&quot;api-limits-01&quot;\u003C\u002Fspan>\u003Cspan class=\"hljs-punctuation\">}\u003C\u002Fspan>\u003C\u002Fcode>\u003C\u002Fpre>\n\u003Cp>Rules that keep the test honest:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>Mix easy questions (shares words with the chunk) and hard ones (paraphrase, abbreviation, typo).\u003C\u002Fli>\n\u003Cli>One gold chunk per question. If two chunks answer it, split the question or pick the better one.\u003C\u002Fli>\n\u003Cli>Write at least 10 questions you expect your current setup to fail on.\u003C\u002Fli>\n\u003Cli>Do not let the model, or anyone who has seen the results, edit the questions afterwards.\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>Fifty questions is small. One question is 2 points of recall, so differences of a few points are noise. That is why the decision rule below uses a margin.\u003C\u002Fp>\n\u003Ch2>Step 3: the scoring script\u003C\u002Fh2>\n\u003Cp>The script embeds all chunks and all questions with each candidate model, ranks chunks by cosine similarity, and prints recall@5, mean reciprocal rank, average query latency and an estimated indexing cost. Ollama's \u003Ccode>\u002Fapi\u002Fembed\u003C\u002Fcode> endpoint takes a model and a list of inputs and returns an \u003Ccode>embeddings\u003C\u002Fcode> array. Cohere's v2 embed endpoint takes \u003Ccode>texts\u003C\u002Fcode>, a \u003Ccode>model\u003C\u002Fcode>, an \u003Ccode>input_type\u003C\u002Fcode> and \u003Ccode>embedding_types\u003C\u002Fcode>. Check the exact Cohere model ID in Cohere's documentation before you run it; the ID below is read from an environment variable so you can change it without editing the file.\u003C\u002Fp>\n\u003Cpre class=\"code-block\" data-lang=\"python\">\u003Ccode class=\"hljs language-python\">\u003Cspan class=\"hljs-keyword\">import\u003C\u002Fspan> json\n\u003Cspan class=\"hljs-keyword\">import\u003C\u002Fspan> os\n\u003Cspan class=\"hljs-keyword\">import\u003C\u002Fspan> time\n\n\u003Cspan class=\"hljs-keyword\">import\u003C\u002Fspan> numpy \u003Cspan class=\"hljs-keyword\">as\u003C\u002Fspan> np\n\u003Cspan class=\"hljs-keyword\">import\u003C\u002Fspan> requests\n\nOLLAMA_URL = os.environ.get(\u003Cspan class=\"hljs-string\">&quot;OLLAMA_URL&quot;\u003C\u002Fspan>, \u003Cspan class=\"hljs-string\">&quot;http:\u002F\u002Flocalhost:11434&quot;\u003C\u002Fspan>)\nCOHERE_KEY = os.environ.get(\u003Cspan class=\"hljs-string\">&quot;COHERE_API_KEY&quot;\u003C\u002Fspan>)\nCOHERE_MODEL = os.environ.get(\u003Cspan class=\"hljs-string\">&quot;COHERE_MODEL&quot;\u003C\u002Fspan>, \u003Cspan class=\"hljs-string\">&quot;embed-v5.0-pro&quot;\u003C\u002Fspan>)\nK = \u003Cspan class=\"hljs-number\">5\u003C\u002Fspan>\n\n\n\u003Cspan class=\"hljs-keyword\">def\u003C\u002Fspan> \u003Cspan class=\"hljs-title function_\">load_jsonl\u003C\u002Fspan>(\u003Cspan class=\"hljs-params\">path\u003C\u002Fspan>):\n    \u003Cspan class=\"hljs-keyword\">with\u003C\u002Fspan> \u003Cspan class=\"hljs-built_in\">open\u003C\u002Fspan>(path, encoding=\u003Cspan class=\"hljs-string\">&quot;utf-8&quot;\u003C\u002Fspan>) \u003Cspan class=\"hljs-keyword\">as\u003C\u002Fspan> f:\n        \u003Cspan class=\"hljs-keyword\">return\u003C\u002Fspan> [json.loads(line) \u003Cspan class=\"hljs-keyword\">for\u003C\u002Fspan> line \u003Cspan class=\"hljs-keyword\">in\u003C\u002Fspan> f \u003Cspan class=\"hljs-keyword\">if\u003C\u002Fspan> line.strip()]\n\n\n\u003Cspan class=\"hljs-keyword\">def\u003C\u002Fspan> \u003Cspan class=\"hljs-title function_\">embed_ollama\u003C\u002Fspan>(\u003Cspan class=\"hljs-params\">model, texts, prefix=\u003Cspan class=\"hljs-string\">&quot;&quot;\u003C\u002Fspan>\u003C\u002Fspan>):\n    out = []\n    \u003Cspan class=\"hljs-keyword\">for\u003C\u002Fspan> i \u003Cspan class=\"hljs-keyword\">in\u003C\u002Fspan> \u003Cspan class=\"hljs-built_in\">range\u003C\u002Fspan>(\u003Cspan class=\"hljs-number\">0\u003C\u002Fspan>, \u003Cspan class=\"hljs-built_in\">len\u003C\u002Fspan>(texts), \u003Cspan class=\"hljs-number\">32\u003C\u002Fspan>):\n        batch = [prefix + t \u003Cspan class=\"hljs-keyword\">for\u003C\u002Fspan> t \u003Cspan class=\"hljs-keyword\">in\u003C\u002Fspan> texts[i:i + \u003Cspan class=\"hljs-number\">32\u003C\u002Fspan>]]\n        r = requests.post(\n            \u003Cspan class=\"hljs-string\">f&quot;\u003Cspan class=\"hljs-subst\">{OLLAMA_URL}\u003C\u002Fspan>\u002Fapi\u002Fembed&quot;\u003C\u002Fspan>,\n            json={\u003Cspan class=\"hljs-string\">&quot;model&quot;\u003C\u002Fspan>: model, \u003Cspan class=\"hljs-string\">&quot;input&quot;\u003C\u002Fspan>: batch},\n            timeout=\u003Cspan class=\"hljs-number\">600\u003C\u002Fspan>,\n        )\n        r.raise_for_status()\n        out.extend(r.json()[\u003Cspan class=\"hljs-string\">&quot;embeddings&quot;\u003C\u002Fspan>])\n    \u003Cspan class=\"hljs-keyword\">return\u003C\u002Fspan> np.array(out, dtype=np.float32)\n\n\n\u003Cspan class=\"hljs-keyword\">def\u003C\u002Fspan> \u003Cspan class=\"hljs-title function_\">embed_cohere\u003C\u002Fspan>(\u003Cspan class=\"hljs-params\">model, texts, input_type\u003C\u002Fspan>):\n    out = []\n    \u003Cspan class=\"hljs-keyword\">for\u003C\u002Fspan> i \u003Cspan class=\"hljs-keyword\">in\u003C\u002Fspan> \u003Cspan class=\"hljs-built_in\">range\u003C\u002Fspan>(\u003Cspan class=\"hljs-number\">0\u003C\u002Fspan>, \u003Cspan class=\"hljs-built_in\">len\u003C\u002Fspan>(texts), \u003Cspan class=\"hljs-number\">96\u003C\u002Fspan>):\n        r = requests.post(\n            \u003Cspan class=\"hljs-string\">&quot;https:\u002F\u002Fapi.cohere.com\u002Fv2\u002Fembed&quot;\u003C\u002Fspan>,\n            headers={\u003Cspan class=\"hljs-string\">&quot;Authorization&quot;\u003C\u002Fspan>: \u003Cspan class=\"hljs-string\">f&quot;Bearer \u003Cspan class=\"hljs-subst\">{COHERE_KEY}\u003C\u002Fspan>&quot;\u003C\u002Fspan>},\n            json={\n                \u003Cspan class=\"hljs-string\">&quot;model&quot;\u003C\u002Fspan>: model,\n                \u003Cspan class=\"hljs-string\">&quot;texts&quot;\u003C\u002Fspan>: texts[i:i + \u003Cspan class=\"hljs-number\">96\u003C\u002Fspan>],\n                \u003Cspan class=\"hljs-string\">&quot;input_type&quot;\u003C\u002Fspan>: input_type,\n                \u003Cspan class=\"hljs-string\">&quot;embedding_types&quot;\u003C\u002Fspan>: [\u003Cspan class=\"hljs-string\">&quot;float&quot;\u003C\u002Fspan>],\n            },\n            timeout=\u003Cspan class=\"hljs-number\">120\u003C\u002Fspan>,\n        )\n        r.raise_for_status()\n        out.extend(r.json()[\u003Cspan class=\"hljs-string\">&quot;embeddings&quot;\u003C\u002Fspan>][\u003Cspan class=\"hljs-string\">&quot;float&quot;\u003C\u002Fspan>])\n    \u003Cspan class=\"hljs-keyword\">return\u003C\u002Fspan> np.array(out, dtype=np.float32)\n\n\n\u003Cspan class=\"hljs-keyword\">def\u003C\u002Fspan> \u003Cspan class=\"hljs-title function_\">normalize\u003C\u002Fspan>(\u003Cspan class=\"hljs-params\">m\u003C\u002Fspan>):\n    \u003Cspan class=\"hljs-keyword\">return\u003C\u002Fspan> m \u002F np.linalg.norm(m, axis=\u003Cspan class=\"hljs-number\">1\u003C\u002Fspan>, keepdims=\u003Cspan class=\"hljs-literal\">True\u003C\u002Fspan>)\n\n\n\u003Cspan class=\"hljs-comment\"># Each candidate: how to embed documents and queries.\u003C\u002Fspan>\n\u003Cspan class=\"hljs-comment\"># nomic-embed-text expects task prefixes; check the model card for yours.\u003C\u002Fspan>\nCANDIDATES = {\n    \u003Cspan class=\"hljs-string\">&quot;ollama:nomic-embed-text&quot;\u003C\u002Fspan>: {\n        \u003Cspan class=\"hljs-string\">&quot;doc&quot;\u003C\u002Fspan>: \u003Cspan class=\"hljs-keyword\">lambda\u003C\u002Fspan> t: embed_ollama(\u003Cspan class=\"hljs-string\">&quot;nomic-embed-text&quot;\u003C\u002Fspan>, t, \u003Cspan class=\"hljs-string\">&quot;search_document: &quot;\u003C\u002Fspan>),\n        \u003Cspan class=\"hljs-string\">&quot;query&quot;\u003C\u002Fspan>: \u003Cspan class=\"hljs-keyword\">lambda\u003C\u002Fspan> t: embed_ollama(\u003Cspan class=\"hljs-string\">&quot;nomic-embed-text&quot;\u003C\u002Fspan>, t, \u003Cspan class=\"hljs-string\">&quot;search_query: &quot;\u003C\u002Fspan>),\n        \u003Cspan class=\"hljs-string\">&quot;usd_per_million_tokens&quot;\u003C\u002Fspan>: \u003Cspan class=\"hljs-number\">0.0\u003C\u002Fspan>,\n    },\n    \u003Cspan class=\"hljs-string\">&quot;ollama:bge-m3&quot;\u003C\u002Fspan>: {\n        \u003Cspan class=\"hljs-string\">&quot;doc&quot;\u003C\u002Fspan>: \u003Cspan class=\"hljs-keyword\">lambda\u003C\u002Fspan> t: embed_ollama(\u003Cspan class=\"hljs-string\">&quot;bge-m3&quot;\u003C\u002Fspan>, t),\n        \u003Cspan class=\"hljs-string\">&quot;query&quot;\u003C\u002Fspan>: \u003Cspan class=\"hljs-keyword\">lambda\u003C\u002Fspan> t: embed_ollama(\u003Cspan class=\"hljs-string\">&quot;bge-m3&quot;\u003C\u002Fspan>, t),\n        \u003Cspan class=\"hljs-string\">&quot;usd_per_million_tokens&quot;\u003C\u002Fspan>: \u003Cspan class=\"hljs-number\">0.0\u003C\u002Fspan>,\n    },\n}\n\u003Cspan class=\"hljs-keyword\">if\u003C\u002Fspan> COHERE_KEY:\n    CANDIDATES[\u003Cspan class=\"hljs-string\">f&quot;cohere:\u003Cspan class=\"hljs-subst\">{COHERE_MODEL}\u003C\u002Fspan>&quot;\u003C\u002Fspan>] = {\n        \u003Cspan class=\"hljs-string\">&quot;doc&quot;\u003C\u002Fspan>: \u003Cspan class=\"hljs-keyword\">lambda\u003C\u002Fspan> t: embed_cohere(COHERE_MODEL, t, \u003Cspan class=\"hljs-string\">&quot;search_document&quot;\u003C\u002Fspan>),\n        \u003Cspan class=\"hljs-string\">&quot;query&quot;\u003C\u002Fspan>: \u003Cspan class=\"hljs-keyword\">lambda\u003C\u002Fspan> t: embed_cohere(COHERE_MODEL, t, \u003Cspan class=\"hljs-string\">&quot;search_query&quot;\u003C\u002Fspan>),\n        \u003Cspan class=\"hljs-comment\"># Fill in from Cohere&#x27;s current pricing page before trusting the cost column.\u003C\u002Fspan>\n        \u003Cspan class=\"hljs-string\">&quot;usd_per_million_tokens&quot;\u003C\u002Fspan>: \u003Cspan class=\"hljs-built_in\">float\u003C\u002Fspan>(os.environ.get(\u003Cspan class=\"hljs-string\">&quot;COHERE_USD_PER_M&quot;\u003C\u002Fspan>, \u003Cspan class=\"hljs-string\">&quot;0&quot;\u003C\u002Fspan>)),\n    }\n\n\n\u003Cspan class=\"hljs-keyword\">def\u003C\u002Fspan> \u003Cspan class=\"hljs-title function_\">main\u003C\u002Fspan>():\n    chunks = load_jsonl(\u003Cspan class=\"hljs-string\">&quot;chunks.jsonl&quot;\u003C\u002Fspan>)\n    questions = load_jsonl(\u003Cspan class=\"hljs-string\">&quot;questions.jsonl&quot;\u003C\u002Fspan>)\n    ids = [c[\u003Cspan class=\"hljs-string\">&quot;id&quot;\u003C\u002Fspan>] \u003Cspan class=\"hljs-keyword\">for\u003C\u002Fspan> c \u003Cspan class=\"hljs-keyword\">in\u003C\u002Fspan> chunks]\n    texts = [c[\u003Cspan class=\"hljs-string\">&quot;text&quot;\u003C\u002Fspan>] \u003Cspan class=\"hljs-keyword\">for\u003C\u002Fspan> c \u003Cspan class=\"hljs-keyword\">in\u003C\u002Fspan> chunks]\n    approx_tokens = \u003Cspan class=\"hljs-built_in\">sum\u003C\u002Fspan>(\u003Cspan class=\"hljs-built_in\">len\u003C\u002Fspan>(t) \u003Cspan class=\"hljs-keyword\">for\u003C\u002Fspan> t \u003Cspan class=\"hljs-keyword\">in\u003C\u002Fspan> texts) \u002F \u003Cspan class=\"hljs-number\">4\u003C\u002Fspan>  \u003Cspan class=\"hljs-comment\"># rough: 4 characters per token\u003C\u002Fspan>\n\n    \u003Cspan class=\"hljs-built_in\">print\u003C\u002Fspan>(\u003Cspan class=\"hljs-string\">f&quot;\u003Cspan class=\"hljs-subst\">{\u003Cspan class=\"hljs-built_in\">len\u003C\u002Fspan>(chunks)}\u003C\u002Fspan> chunks, \u003Cspan class=\"hljs-subst\">{\u003Cspan class=\"hljs-built_in\">len\u003C\u002Fspan>(questions)}\u003C\u002Fspan> questions&quot;\u003C\u002Fspan>)\n    \u003Cspan class=\"hljs-built_in\">print\u003C\u002Fspan>(\u003Cspan class=\"hljs-string\">f&quot;\u003Cspan class=\"hljs-subst\">{\u003Cspan class=\"hljs-string\">&#x27;model&#x27;\u003C\u002Fspan>:\u003Cspan class=\"hljs-number\">32\u003C\u002Fspan>}\u003C\u002Fspan> \u003Cspan class=\"hljs-subst\">{\u003Cspan class=\"hljs-string\">&#x27;recall@5&#x27;\u003C\u002Fspan>:&gt;\u003Cspan class=\"hljs-number\">9\u003C\u002Fspan>}\u003C\u002Fspan> \u003Cspan class=\"hljs-subst\">{\u003Cspan class=\"hljs-string\">&#x27;MRR&#x27;\u003C\u002Fspan>:&gt;\u003Cspan class=\"hljs-number\">6\u003C\u002Fspan>}\u003C\u002Fspan> \u003Cspan class=\"hljs-subst\">{\u003Cspan class=\"hljs-string\">&#x27;ms\u002Fquery&#x27;\u003C\u002Fspan>:&gt;\u003Cspan class=\"hljs-number\">9\u003C\u002Fspan>}\u003C\u002Fspan> \u003Cspan class=\"hljs-subst\">{\u003Cspan class=\"hljs-string\">&#x27;index $&#x27;\u003C\u002Fspan>:&gt;\u003Cspan class=\"hljs-number\">8\u003C\u002Fspan>}\u003C\u002Fspan>&quot;\u003C\u002Fspan>)\n\n    \u003Cspan class=\"hljs-keyword\">for\u003C\u002Fspan> name, c \u003Cspan class=\"hljs-keyword\">in\u003C\u002Fspan> CANDIDATES.items():\n        d = normalize(c[\u003Cspan class=\"hljs-string\">&quot;doc&quot;\u003C\u002Fspan>](texts))\n\n        hits, rr, lat = \u003Cspan class=\"hljs-number\">0\u003C\u002Fspan>, \u003Cspan class=\"hljs-number\">0.0\u003C\u002Fspan>, []\n        \u003Cspan class=\"hljs-keyword\">for\u003C\u002Fspan> q \u003Cspan class=\"hljs-keyword\">in\u003C\u002Fspan> questions:\n            t0 = time.perf_counter()\n            qv = normalize(c[\u003Cspan class=\"hljs-string\">&quot;query&quot;\u003C\u002Fspan>]([q[\u003Cspan class=\"hljs-string\">&quot;question&quot;\u003C\u002Fspan>]]))[\u003Cspan class=\"hljs-number\">0\u003C\u002Fspan>]\n            lat.append((time.perf_counter() - t0) * \u003Cspan class=\"hljs-number\">1000\u003C\u002Fspan>)\n            order = np.argsort(-(d @ qv))\n            ranked = [ids[i] \u003Cspan class=\"hljs-keyword\">for\u003C\u002Fspan> i \u003Cspan class=\"hljs-keyword\">in\u003C\u002Fspan> order]\n            \u003Cspan class=\"hljs-keyword\">if\u003C\u002Fspan> q[\u003Cspan class=\"hljs-string\">&quot;gold&quot;\u003C\u002Fspan>] \u003Cspan class=\"hljs-keyword\">in\u003C\u002Fspan> ranked[:K]:\n                hits += \u003Cspan class=\"hljs-number\">1\u003C\u002Fspan>\n            \u003Cspan class=\"hljs-keyword\">if\u003C\u002Fspan> q[\u003Cspan class=\"hljs-string\">&quot;gold&quot;\u003C\u002Fspan>] \u003Cspan class=\"hljs-keyword\">in\u003C\u002Fspan> ranked:\n                rr += \u003Cspan class=\"hljs-number\">1.0\u003C\u002Fspan> \u002F (ranked.index(q[\u003Cspan class=\"hljs-string\">&quot;gold&quot;\u003C\u002Fspan>]) + \u003Cspan class=\"hljs-number\">1\u003C\u002Fspan>)\n\n        n = \u003Cspan class=\"hljs-built_in\">len\u003C\u002Fspan>(questions)\n        cost = approx_tokens \u002F \u003Cspan class=\"hljs-number\">1e6\u003C\u002Fspan> * c[\u003Cspan class=\"hljs-string\">&quot;usd_per_million_tokens&quot;\u003C\u002Fspan>]\n        \u003Cspan class=\"hljs-built_in\">print\u003C\u002Fspan>(\u003Cspan class=\"hljs-string\">f&quot;\u003Cspan class=\"hljs-subst\">{name:\u003Cspan class=\"hljs-number\">32\u003C\u002Fspan>}\u003C\u002Fspan> \u003Cspan class=\"hljs-subst\">{hits \u002F n:\u003Cspan class=\"hljs-number\">9.2\u003C\u002Fspan>f}\u003C\u002Fspan> \u003Cspan class=\"hljs-subst\">{rr \u002F n:\u003Cspan class=\"hljs-number\">6.2\u003C\u002Fspan>f}\u003C\u002Fspan> \u003Cspan class=\"hljs-subst\">{np.median(lat):\u003Cspan class=\"hljs-number\">9.0\u003C\u002Fspan>f}\u003C\u002Fspan> \u003Cspan class=\"hljs-subst\">{cost:\u003Cspan class=\"hljs-number\">8.4\u003C\u002Fspan>f}\u003C\u002Fspan>&quot;\u003C\u002Fspan>)\n\n\n\u003Cspan class=\"hljs-keyword\">if\u003C\u002Fspan> __name__ == \u003Cspan class=\"hljs-string\">&quot;__main__&quot;\u003C\u002Fspan>:\n    main()\u003C\u002Fcode>\u003C\u002Fpre>\n\u003Cp>Pull the local models and run it:\u003C\u002Fp>\n\u003Cpre class=\"code-block\" data-lang=\"bash\">\u003Ccode class=\"hljs language-bash\">ollama pull nomic-embed-text\nollama pull bge-m3\n\u003Cspan class=\"hljs-built_in\">export\u003C\u002Fspan> COHERE_API_KEY=your-key   \u003Cspan class=\"hljs-comment\"># optional\u003C\u002Fspan>\n\u003Cspan class=\"hljs-built_in\">export\u003C\u002Fspan> COHERE_MODEL=embed-v5.0-pro   \u003Cspan class=\"hljs-comment\"># confirm the ID in Cohere&#x27;s docs\u003C\u002Fspan>\n\u003Cspan class=\"hljs-built_in\">export\u003C\u002Fspan> COHERE_USD_PER_M=0.12         \u003Cspan class=\"hljs-comment\"># copy from Cohere&#x27;s current pricing page\u003C\u002Fspan>\npython score.py\u003C\u002Fcode>\u003C\u002Fpre>\n\u003Cp>The output is a table like the one below. The numbers here are placeholders to show the format, not measurements; yours will differ.\u003C\u002Fp>\n\u003Cpre class=\"code-block\">\u003Ccode class=\"hljs\">model                          recall@5    MRR  ms\u002Fquery  index $\nollama:nomic-embed-text            x.xx   x.xx       xxx   0.0000\nollama:bge-m3                      x.xx   x.xx       xxx   0.0000\ncohere:embed-v5.0-pro              x.xx   x.xx       xxx   x.xxxx\u003C\u002Fcode>\u003C\u002Fpre>\n\u003Ch2>Things that quietly skew the result\u003C\u002Fh2>\n\u003Cul>\n\u003Cli>\u003Cstrong>Task prefixes.\u003C\u002Fstrong> Some models, such as nomic-embed-text, are trained with prefixes like \u003Ccode>search_document:\u003C\u002Fcode> and \u003Ccode>search_query:\u003C\u002Fcode>. Leaving them out can understate the model. Read each model card.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Context length.\u003C\u002Fstrong> Check the context window your runtime actually uses. One guide notes the Ollama card for nomic-embed-text lists a smaller context than the model's native window, so long chunks may be truncated. Truncation looks like a bad model.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Hosted tiers.\u003C\u002Fstrong> Cohere describes Embed 5 Pro as tuned for retrieval quality and Fast as tuned for latency and throughput, and says the two share a compatible vector space, so you can index with Pro and query with Fast. Test that pairing too: it can change your latency-and-cost trade-off without rebuilding the index. Verify this in the Cohere docs for the model you deploy.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Latency location.\u003C\u002Fstrong> A hosted model's latency includes the network from where your code runs. Measure from the machine that will serve queries, and at the same time of day you expect traffic.\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch2>Step 4: a decision rule, written before you run\u003C\u002Fh2>\n\u003Cfigure data-post-media=\"6b0280134b6fd09d50051877\" data-layout=\"right\">\u003Cimg src=\"https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6b\u002F6b0280134b6fd09d5005187c-0-6a47db27.png\" alt=\"Weighing performance gains against latency and operational costs.\" loading=\"lazy\">\u003Cfigcaption>Weighing performance gains against latency and operational costs.\u003C\u002Ffigcaption>\u003C\u002Ffigure>\n\u003Cp>Write this down first so the results do not pick the rule:\u003C\u002Fp>\n\u003Col>\n\u003Cli>Discard any model whose recall@5 is more than 4 points (two questions) below the best.\u003C\u002Fli>\n\u003Cli>Among the rest, choose the cheapest to run at your expected query volume.\u003C\u002Fli>\n\u003Cli>If two are within the margin on quality and cost, choose the one with lower median query latency.\u003C\u002Fli>\n\u003Cli>If all are within the margin on recall, your retrieval problem is probably chunking or the questions, not the model. Look at the failed questions before buying anything.\u003C\u002Fli>\n\u003C\u002Fol>\n\u003Cp>Then read the failures. For each question the winner missed, print the gold chunk and the top three chunks it returned. Most misses are chunk boundaries or missing context in the chunk, not embedding quality.\u003C\u002Fp>\n\u003Ch2>When to re-run\u003C\u002Fh2>\n\u003Cp>Keep \u003Ccode>chunks.jsonl\u003C\u002Fcode>, \u003Ccode>questions.jsonl\u003C\u002Fcode> and the script in your repo. When a new model arrives, add one entry to \u003Ccode>CANDIDATES\u003C\u002Fcode> and rerun. Re-run too when your corpus changes a lot. The whole run takes minutes, and it replaces an argument about leaderboards with a table of your own data.\u003C\u002Fp>\n\u003Ch2>Sources\u003C\u002Fh2>\n\u003Cul>\n\u003Cli>\u003Ca href=\"https:\u002F\u002Fwww.hpcwire.com\u002Faiwire\u002F2026\u002F10\u002F01\u002Fcohere-releases-embed-5-with-pro-and-fast-tiers-for-enterprise-ai\u002F\">HPCwire: Cohere releases Embed 5 with Pro and Fast tiers (1 Oct 2026)\u003C\u002Fa>\u003C\u002Fli>\n\u003Cli>\u003Ca href=\"https:\u002F\u002Fdocs.ollama.com\u002Fapi\u002Fembed\">Ollama docs: \u002Fapi\u002Fembed\u003C\u002Fa>\u003C\u002Fli>\n\u003Cli>\u003Ca href=\"https:\u002F\u002Fdocs.ollama.com\u002Fcapabilities\u002Fembeddings\">Ollama docs: embeddings\u003C\u002Fa>\u003C\u002Fli>\n\u003Cli>\u003Ca href=\"https:\u002F\u002Fwww.morphllm.com\u002Follama-embedding-models\">Morph: Ollama embedding models compared\u003C\u002Fa>\u003C\u002Fli>\n\u003Cli>\u003Ca href=\"https:\u002F\u002Fwww.stack-ai.com\u002Finsights\u002Fbest-embedding-models-for-rag-in-2026-a-comparison-guide\">Stack AI: best embedding models for RAG in 2026\u003C\u002Fa>\u003C\u002Fli>\n\u003Cli>\u003Ca href=\"https:\u002F\u002Fzilliz.com\u002Fko\u002Fblog\u002Fchoose-embedding-model-rag-2026\">Zilliz: how to choose an embedding model for RAG\u003C\u002Fa>\u003C\u002Fli>\n\u003C\u002Ful>","Leaderboards will not tell you which embedding model fits your corpus. Build a 50-question labelled set, score recall@5 for a hosted and a local model, and pick with a rule.",[9,10,11,12,13],"rag","embeddings","evaluation","ollama","ai-assisted","if.codes","https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6b\u002F6b0280124b6fd09d5005185d-0-f4e2bc63.png","2026-11-22T06:09:10.347Z","Choose a RAG embedding model with a 50-question retrieval test","Score recall@5, latency and cost for Cohere and Ollama embedding models on your own documents, with a script and a decision rule.",6,[21,33,46],{"slug":22,"title":23,"type":24,"summary":25,"tags":26,"author":14,"cover_url":30,"published_at":31,"updated_at":32},"claude-code-local-model-ollama-launch","Point Claude Code at a local model: ollama launch claude, the 64k context rule and a --yes CI run","blog","Ollama documents running Claude Code against local or cloud open-weight models. The launch command, the manual env-var setup, the 64k context rule and a scripted --yes run, with the documented limits.",[27,12,28,29,13],"claude-code","local-llm","ci","https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6b\u002F6b02f88893220cbaf46f764c-0-43aeb43f.png","2026-11-22T15:02:08.453Z","2026-11-22T15:02:08.454Z",{"slug":34,"title":35,"type":24,"summary":36,"tags":37,"author":14,"cover_url":43,"published_at":44,"updated_at":45},"lets-encrypt-64-day-certificates-renewal-audit-certbot-traefik","Let’s Encrypt goes to 64-day certificates: a copy-paste renewal audit for certbot and Traefik","On 10 Feb 2027 Let’s Encrypt’s default certificate lifetime drops from 90 to 64 days. Here is a short audit for certbot and Traefik: inventory, dry run, served-cert check and an expiry alert.",[38,39,40,41,42,13],"letsencrypt","tls","certbot","traefik","devops","https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6b\u002F6b012e6d9eb7977c0ad40dca-0-89d8f435.png","2026-11-21T06:36:11.966Z","2026-11-21T06:36:11.967Z",{"slug":47,"title":48,"type":24,"summary":49,"tags":50,"author":14,"cover_url":55,"published_at":56,"updated_at":57},"claude-code-usage-prompt-cache-line-token-leaks","Why your Claude Code limit disappears: read the \u002Fusage cache line and fix the five biggest token leaks","Claude Code now shows a Prompt cache line in \u002Fusage. How to read it, and five habits from Anthropic's cost docs that stop long sessions burning through your limit.",[51,52,53,54,13],"claude code","tokens","cost","hooks","https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6a\u002F6aff67d19eb7977c0ad3d59b-0-344201e4.png","2026-11-20T00:44:16.753Z","2026-11-20T00:44:16.754Z",[59,62,65,68,70,83,94,104,115,126,134,145,157,167,178,189,200,211,220,230,241,250,259,269,278,288,299,309,319,330,341,349,358,371,382,392,402,412,423,431,442,453,462,471,481,490,499],{"slug":22,"title":23,"type":24,"summary":25,"tags":60,"author":14,"cover_url":30,"published_at":31,"reading_minutes":61,"updated_at":32},[27,12,28,29,13],4,{"slug":4,"title":5,"type":24,"summary":7,"tags":63,"author":14,"cover_url":15,"published_at":16,"reading_minutes":19,"updated_at":64},[9,10,11,12,13],"2026-11-22T06:09:10.348Z",{"slug":34,"title":35,"type":24,"summary":36,"tags":66,"author":14,"cover_url":43,"published_at":44,"reading_minutes":67,"updated_at":45},[38,39,40,41,42,13],5,{"slug":47,"title":48,"type":24,"summary":49,"tags":69,"author":14,"cover_url":55,"published_at":56,"reading_minutes":67,"updated_at":57},[51,52,53,54,13],{"slug":71,"title":72,"type":24,"summary":73,"tags":74,"author":14,"cover_url":79,"published_at":80,"reading_minutes":81,"updated_at":82},"always-on-agent-inbox-timers-outbox","An always-on agent skeleton in 150 lines: durable inbox, timers and outbox","A runnable SQLite sketch of the pieces Anthropic's always-on agent guide names: a deduplicating inbox, one active worker, timers and an outbox, with tests for a crashed run and a duplicate webhook.",[75,76,77,78,13],"agents","python","sqlite","architecture","https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6a\u002F6afbf6709eb7977c0ad36cee-0-4ac7146f.png","2026-11-17T07:37:13.91Z",8,"2026-11-17T07:37:13.911Z",{"slug":84,"title":85,"type":24,"summary":86,"tags":87,"author":14,"cover_url":90,"published_at":91,"reading_minutes":92,"updated_at":93},"claude-code-read-only-reviewer-subagent","Claude Code subagents: write a read-only reviewer and pin its model and tools","Write a code-reviewer subagent file with a read-only tool allowlist and a pinned model, learn how the model is resolved, and add a PreToolUse hook for finer limits.",[27,88,89,54,13],"subagents","code-review","https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6a\u002F6afad85a6499079288f5de91-0-6b6cc377.png","2026-11-17T01:36:15.209Z",3,"2026-11-22T16:58:10.553Z",{"slug":95,"title":96,"type":24,"summary":97,"tags":98,"author":14,"cover_url":102,"published_at":103,"reading_minutes":67,"updated_at":103},"claude-prompt-caching-breakpoints-prove-a-hit","Prompt caching with Claude: where the breakpoints go and how to prove you got a hit","Repeated system prompts and tool definitions are the cheapest thing to cache. Where to put cache_control, 5-minute vs 1-hour TTL, the price multipliers, and the usage fields that show a hit.",[99,100,101,53,13],"claude","prompt-caching","api","https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6a\u002F6afa2a366499079288f5c47e-0-43ee908b.png","2026-11-15T23:58:13.018Z",{"slug":105,"title":106,"type":24,"summary":107,"tags":108,"author":14,"cover_url":112,"published_at":113,"reading_minutes":61,"updated_at":114},"rendering-video-with-code-only-frames-ffmpeg-numpy-audio","Rendering video with code only: frames, ffmpeg and numpy-synthesised audio, step by step","A promo video needs no editor: draw frames in code, synthesise the soundtrack with numpy, join them with ffmpeg, then check loudness. A minimal pipeline modelled on an open-source Claude skill.",[109,110,76,111,13],"video","ffmpeg","automation","https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6a\u002F6aec7c68f5d46108a9669907-0-e8da7110.png","2026-11-05T23:58:13.025Z","2026-11-05T23:58:13.026Z",{"slug":116,"title":117,"type":24,"summary":118,"tags":119,"author":14,"cover_url":123,"published_at":124,"reading_minutes":92,"updated_at":125},"structured-data-from-photos-local-vision-model-ollama-pydantic","Structured data from photos with a local vision model: Ollama's format parameter and a Pydantic schema","Turn receipt or label photos into validated records with a local vision model: Ollama's format parameter, a Pydantic schema, temperature 0 and a retry loop.",[12,120,121,122,76,13],"vision","structured-output","pydantic","https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6a\u002F6aeb6adfcb6b554bc882a8a3-0-7ee45773.png","2026-11-05T02:12:12.056Z","2026-11-05T02:12:12.057Z",{"slug":127,"title":128,"type":24,"summary":129,"tags":130,"author":14,"cover_url":131,"published_at":132,"reading_minutes":61,"updated_at":133},"reliable-json-local-model-ollama-pydantic-structured-outputs","Reliable JSON from a local model: schema-constrained output with Ollama and Pydantic","Pass a Pydantic JSON schema to Ollama's format field, repeat it in the prompt, set temperature 0, and validate the reply. Copy-paste extraction snippet and retry loop.",[12,122,76,28,13],"https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6a\u002F6ae48c8737fa83c1536d8ba6-0-1ada80df.png","2026-10-31T02:50:23.906Z","2026-11-22T18:23:45.469Z",{"slug":135,"title":136,"type":24,"summary":137,"tags":138,"author":14,"cover_url":143,"published_at":144,"reading_minutes":67,"updated_at":144},"testcontainers-real-database-coding-agent-tests","Make your coding agent's tests hit a real database with Testcontainers","Coding agents love mock-heavy unit tests that pass while production behaves differently. A step-by-step Testcontainers + Postgres setup in Python, an agent instruction block, and a keep-vs-delete checklist.",[139,140,141,142,13],"testing","testcontainers","postgres","coding-agents","https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6a\u002F6ae1baa01edd7a39ec513116-0-77a44a46.png","2026-10-29T05:04:16.807Z",{"slug":146,"title":147,"type":24,"summary":148,"tags":149,"author":14,"cover_url":154,"published_at":155,"reading_minutes":19,"updated_at":156},"sandbox-local-mcp-servers-srt-without-docker","Sandbox your local MCP servers without Docker: Anthropic's srt on Seatbelt and bubblewrap","A 15-minute guide to wrapping a local MCP server in Anthropic's sandbox-runtime (srt): install it, start from deny-by-default, open only the paths and domains the server needs, and know what this research preview does not cover.",[150,151,152,153,13],"mcp","sandboxing","security","ai-agents","https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6a\u002F6ae039071edd7a39ec50fb37-0-1d24ae87.png","2026-10-27T09:02:15.11Z","2026-10-27T09:02:15.111Z",{"slug":158,"title":159,"type":24,"summary":160,"tags":161,"author":14,"cover_url":165,"published_at":166,"reading_minutes":19,"updated_at":166},"vps-zero-inbound-ports-tailscale-ssh-cloudflare-tunnel","Take your VPS off the public internet: SSH over Tailscale, the site behind a Cloudflare Tunnel, and a firewall with zero inbound ports","A lockout-safe order for a small VPS: Tailscale SSH first, then serve the site through cloudflared, then deny every inbound port. Copy-pasteable commands for Ubuntu, and the checks that prove nothing listens publicly.",[152,162,163,164,13],"vps","tailscale","cloudflare","https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6a\u002F6add96066f13dd01b60c05c6-0-ef88c8d0.png","2026-10-25T06:30:17.63Z",{"slug":168,"title":169,"type":24,"summary":170,"tags":171,"author":14,"cover_url":175,"published_at":176,"reading_minutes":67,"updated_at":177},"php2shell-patch-php-fpm-mod-php-frankenphp","PHP2Shell on your own server: reaching PHP 8.2.35+ on php-fpm, mod_php and FrankenPHP, and the stopgap when you're stuck on 8.1","php.net's 21 October release fixes two unauthenticated RCEs that hit every PHP web SAPI. Version checks that prove the running worker is patched, upgrade paths for distro packages, Docker images and FrankenPHP, and php.net's interim mitigations for 8.1 and older.",[172,152,173,174,13],"php","self-hosting","docker","https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6a\u002F6add3fc76f13dd01b60bf641-0-fb344fed.png","2026-10-25T01:43:14.026Z","2026-10-25T01:43:14.027Z",{"slug":179,"title":180,"type":24,"summary":181,"tags":182,"author":14,"cover_url":186,"published_at":187,"reading_minutes":19,"updated_at":188},"hetzner-2026-price-round-decision-sheet","Hetzner tripled some new-instance prices: keep, rescale or move — a decision sheet for a small self-hosted stack","Hetzner's June 2026 prices hit new orders and rescales; old servers keep their rate. Why not to rescale a grandfathered box, where Arm CAX still fits, and hcloud commands to price a move.",[183,173,184,185,13],"hetzner","cloud-costs","hcloud","https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6a\u002F6ada8ebbe6d7e5a76992c1dd-0-f450e23a.png","2026-10-23T00:43:13.645Z","2026-10-23T00:43:13.646Z",{"slug":190,"title":191,"type":24,"summary":192,"tags":193,"author":14,"cover_url":198,"published_at":199,"reading_minutes":19,"updated_at":199},"gpt-4o-gpt-4-1-shutdown-migrate-to-gpt-6-luna","GPT-4o and GPT-4.1 leave the API on January 14: moving to gpt-6-luna without breaking your function calls","OpenAI shuts down four GPT-4o and GPT-4.1 snapshots, and their fine-tunes, on January 14, 2027. Its named replacement, gpt-6-luna, is a reasoning model with one function-calling catch in Chat Completions. Here are two migration paths, a replay test and a token-bill check.",[194,195,196,197,13],"openai","migration","llm-api","function-calling","https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6a\u002F6ad6f0886ce38bf97b746887-0-cad17c1e.png","2026-10-20T05:42:12.599Z",{"slug":201,"title":202,"type":24,"summary":203,"tags":204,"author":14,"cover_url":208,"published_at":209,"reading_minutes":67,"updated_at":210},"swapping-docker-desktop-for-apple-container","Swapping Docker Desktop for Apple's container: one VM per container on your Mac, what runs unchanged from Docker Hub, and where you still need Docker","Apple's container CLI has been stable since 1.0 in June and is at 1.5.0. A hands-on swap: install, build, run Postgres from Docker Hub, reach it by name, and the gaps (no Compose, slower cold starts, memory that is not handed back) that decide whether Docker stays.",[174,205,206,207,13],"macos","containers","devtools","https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6a\u002F6ad59fc06ce38bf97b743ea2-0-d7ac665e.png","2026-10-19T06:50:15.817Z","2026-10-19T06:50:15.818Z",{"slug":212,"title":213,"type":24,"summary":214,"tags":215,"author":14,"cover_url":218,"published_at":219,"reading_minutes":67,"updated_at":219},"github-agentic-autofix-codeql-backlog","GitHub's agentic autofix is GA: batching 25 CodeQL alerts into one draft PR, what it bills, and how to review the fixes","Agentic autofix for code-scanning alerts went GA on October 15: Copilot explores the repo, fixes, re-runs CodeQL and opens a draft PR for up to 25 alerts at once. Who can use it, how to turn it on, where the cost lands, how to pick a first batch with gh, and a review checklist.",[216,217,152,89,13],"github","codeql","https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6a\u002F6ad48184af8e63ad10e8aaa1-0-4c6cfc5e.png","2026-10-19T00:20:22.102Z",{"slug":221,"title":222,"type":24,"summary":223,"tags":224,"author":14,"cover_url":227,"published_at":228,"reading_minutes":229,"updated_at":228},"claude-code-permission-rules-yes-to-all","'Yes to all' is a permission policy: writing Claude Code allow, ask and deny rules that hold up against compound commands and wildcards","Every 'Yes, and don't ask again' writes a rule. Turn that habit into a reviewed .claude\u002Fsettings.json: deny beats ask beats allow, why Bash(git * main) also allows git push, how compound commands are split, and when dontAsk or the sandbox beats bypassPermissions.",[27,225,152,226,13],"permissions","developer-tools","https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6a\u002F6ad2fd1602d67a73cf1eaa23-0-68817e09.png","2026-10-17T23:06:09.746Z",7,{"slug":231,"title":232,"type":24,"summary":233,"tags":234,"author":14,"cover_url":239,"published_at":240,"reading_minutes":19,"updated_at":240},"go-1-27-http1-connection-reuse-body-draining","710 ms to 150 ms for 100 requests: what Go 1.27's HTTP\u002F1 connection reuse fixes, and the body-draining habit to keep for older clients","Go 1.27 drains unread HTTP\u002F1 response bodies for you, so a closed-but-unread body no longer kills connection reuse. A runnable test that counts reused connections, the two patterns that broke reuse, and a capped drain helper for Go 1.26 and older.",[235,236,237,238,13],"go","http","performance","networking","https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6a\u002F6ad1d8da69fb9bde4cc895fb-0-e539bbf2.png","2026-10-16T23:32:12.783Z",{"slug":242,"title":243,"type":24,"summary":244,"tags":245,"author":14,"cover_url":247,"published_at":248,"reading_minutes":19,"updated_at":249},"claude-code-subagent-prompt-cache-ttl","Claude Code subagents get a 5-minute cache even on a subscription: when to set it to 1h, and how to check","On a subscription your main Claude Code conversation gets a one-hour prompt cache, but subagents, workflows and compaction stay at five minutes. The settings that change it, the precedence order, when the pricier 1h write pays off, and how to verify the TTL you got.",[27,100,88,246,13],"llm-costs","https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6a\u002F6ad0968473d443ad8fa7adbe-0-f2f3fe8c.png","2026-10-15T22:21:15.624Z","2026-10-15T22:21:15.625Z",{"slug":251,"title":252,"type":24,"summary":253,"tags":254,"author":14,"cover_url":256,"published_at":257,"reading_minutes":67,"updated_at":258},"postgres-slow-vps-benchmark-the-disk-first","Postgres feels slow on your VPS? Benchmark the disk before you touch postgresql.conf","ClickHouse measured the same Postgres box at 16,030 TPS on local NVMe and 1,734 TPS on a baseline gp3 volume, with most of the gap spent waiting on storage. A 15-minute fio and pg_test_fsync check tells you which side of that table your VPS is on.",[141,173,237,255,13],"databases","https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6a\u002F6acdb5d473d443ad8fa7555c-0-dd05a138.png","2026-10-13T05:44:15.759Z","2026-10-13T05:44:15.76Z",{"slug":260,"title":261,"type":24,"summary":262,"tags":263,"author":14,"cover_url":266,"published_at":267,"reading_minutes":61,"updated_at":268},"quoted-tilde-in-path-check","A quoted ~ in your PATH runs whatever sits in .\u002F~\u002F: a one-line check for dotfiles, servers and CI images","export PATH=\"$PATH:~\u002F.local\u002Fbin\" never expands the tilde. Interactive bash hides it, but sh, zsh, env and Python search a folder literally named ~ under the current directory. The check, the fix, and where to grep.",[264,265,152,42,13],"linux","shell","https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6a\u002F6acd3f0073d443ad8fa743bf-0-c8db0f5d.png","2026-10-12T23:53:16.705Z","2026-10-12T23:53:16.706Z",{"slug":270,"title":271,"type":24,"summary":272,"tags":273,"author":14,"cover_url":275,"published_at":276,"reading_minutes":19,"updated_at":277},"docker-published-ports-bypass-ufw-vps-hardening","Docker publishes ports straight past UFW: hardening a fresh VPS without the firewall rule that does nothing","ufw deny 8080 does not stop a Docker container published on 8080. Docker's own docs explain why. A 15-minute fresh-VPS walkthrough: key-only SSH, ufw, then the two fixes that actually close a container port, and a scan from outside to prove it.",[174,152,162,274,13],"firewall","https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6a\u002F6acb12dddbaf8c2bc523d0bb-0-21ee3023.png","2026-10-11T05:42:15.363Z","2026-10-11T05:42:15.364Z",{"slug":279,"title":280,"type":24,"summary":281,"tags":282,"author":14,"cover_url":285,"published_at":286,"reading_minutes":19,"updated_at":287},"deno-deploy-shutdown-exit-guide","Deno Deploy shuts down in six months: moving a small app to Cloudflare Workers, self-hosted Deno in Docker, or Node","Deno is joining Cloudflare: Deploy has six months left and the runtime gets a year of fixes. Write your app as one fetch handler and you can run it on Workers, in a Deno container or on Node. Code for all three, plus what does not carry over.",[283,284,195,174,13],"deno","cloudflare-workers","https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6a\u002F6ac9c181dbaf8c2bc523a0fd-0-caa268de.png","2026-10-10T05:47:19.665Z","2026-10-10T05:47:19.666Z",{"slug":289,"title":290,"type":24,"summary":291,"tags":292,"author":14,"cover_url":296,"published_at":297,"reading_minutes":229,"updated_at":298},"durable-object-alarm-loop-guards","The Durable Object alarm that cost $10,000: guarding setAlarm() loops with backoff, idempotency and a kill switch","Durable Object alarms are at-least-once, retried, and usually rescheduled from inside the handler, which makes a billing loop easy to write. Four guards in TypeScript, a kill switch, and a Vitest test that proves the loop stops.",[164,293,294,295,13],"durable-objects","typescript","cost-control","https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6a\u002F6ac8709b392a85804f209fb3-0-9143ed0c.png","2026-10-09T06:09:10.127Z","2026-10-09T06:09:10.128Z",{"slug":300,"title":301,"type":24,"summary":302,"tags":303,"author":14,"cover_url":307,"published_at":308,"reading_minutes":19,"updated_at":308},"cctld-hijack-counterfeit-certs-caa-accounturi-ct-watch","Counterfeit certs via hijacked ccTLDs: CAA with accounturi and a CT watch for your domains in 15 minutes","Attackers took over the .gh, .sl and .as registries and got 12 valid certificates for Google domains. CAA would not have stopped the hijack. Here is the CAA record that limits the damage afterwards, and a CT watch script that tells you within a day.",[39,304,305,306,13],"caa","certificate-transparency","dns","https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6a\u002F6ac75337392a85804f207d25-0-12f538ac.png","2026-10-08T22:47:12.408Z",{"slug":310,"title":311,"type":24,"summary":312,"tags":313,"author":14,"cover_url":316,"published_at":317,"reading_minutes":19,"updated_at":318},"embeddinggemma-2-qdrant-truncation-recall","EmbeddingGemma 2 in Qdrant: one 740M model for text, images and audio, and what truncating 768 to 128 dims costs your recall","Google's open multimodal embedder runs locally and truncates from 768 to 128 dims. Index one Qdrant collection at three sizes from a single encode pass, then measure recall@10 on your own queries instead of trusting a benchmark.",[10,314,9,315,13],"qdrant","open-models","https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6a\u002F6ac5cceedf0d655df505306e-0-d17d0fef.png","2026-10-07T07:21:20.071Z","2026-10-07T07:21:20.072Z",{"slug":320,"title":321,"type":24,"summary":322,"tags":323,"author":14,"cover_url":327,"published_at":328,"reading_minutes":67,"updated_at":329},"cloudflare-access-strict-service-token-auth-migration","Strict service token auth in Cloudflare Access: moving your scripts and CI over before it bites","Cloudflare Access now has a strict mode for service tokens: 401\u002F403 instead of a 302 to the login page, only Service Auth policies count, and no CF_Authorization cookie. New orgs get it forced on from 5 October. A 15-minute check and switch for existing orgs.",[164,324,325,326,13],"zero-trust","ci-cd","authentication","https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6a\u002F6ac4a8860956594c947bd197-0-f768b25b.png","2026-10-06T08:30:13.154Z","2026-11-22T16:58:10.752Z",{"slug":331,"title":332,"type":24,"summary":333,"tags":334,"author":14,"cover_url":338,"published_at":339,"reading_minutes":19,"updated_at":340},"cloudflare-traces-trace-rules-debug-one-customer","Why was that request blocked? Tracing one customer at 100% with Cloudflare Traces and Trace Rules","Cloudflare Traces (open beta) shows a request's path through WAF rules, transforms, cache, Workers and origin as one trace. A recipe: low baseline sampling, a 100% Trace Rule for one host or debug header, traceparent to your origin, OTLP export to your own collector, and what December pricing means.",[164,335,336,337,13],"observability","opentelemetry","tracing","https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6a\u002F6ac338052cb6b40613ac2b7c-0-2294d114.png","2026-10-05T06:22:19.055Z","2026-10-05T06:22:19.056Z",{"slug":342,"title":343,"type":24,"summary":344,"tags":345,"author":14,"cover_url":347,"published_at":348,"reading_minutes":67,"updated_at":348},"copyescape-cve-2026-17106-patch-docker-cp","CopyEscape (CVE-2026-17106): patch docker cp, and stop copying out of running containers","A race in docker cp lets a malicious container write files anywhere the copying process can write on the host. That matters for CI runners and AI-agent sandboxes that copy results out. Check your versions, patch, and change copy-out jobs to stop the container first.",[174,152,346,29,13],"cve","https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6a\u002F6ac218ebfd770659725abe68-0-eee7c8f7.png","2026-10-04T23:51:14.689Z",{"slug":350,"title":351,"type":24,"summary":352,"tags":353,"author":14,"cover_url":355,"published_at":356,"reading_minutes":61,"updated_at":357},"protected-quick-tunnels-vs-tailscale-funnel","Share localhost with three named people: Cloudflare's Protected Quick Tunnels vs Tailscale Funnel","cloudflared 2026.9.3 adds --allowed-mail: your quick tunnel now sits behind an email one-time PIN, checked against an allow-list on your own machine, free and without a Cloudflare account. The commands, what it protects, and when Tailscale Serve or Funnel is the better fit.",[164,163,354,152,13],"tunnels","https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6a\u002F6ac218d8fd770659725abe3a-0-40f8cb2e.png","2026-10-04T23:19:14.764Z","2026-10-04T23:19:14.765Z",{"slug":359,"title":360,"type":24,"summary":361,"tags":362,"author":14,"cover_url":368,"published_at":369,"reading_minutes":19,"updated_at":370},"si-domains-super-intelligence-data",".si after 'Super Intelligence': did one UN speech move a ccTLD?","Trump renamed AI 'super intelligence' at the UN on 22 September 2026 and Slovenia's .si went from about 190,000 names to almost 276,000 in a month. Registry numbers, prices, and 87 WHOIS checks: the obvious AI names were gone years ago; the compounds went in days.",[363,364,365,366,367,13],"domains","si","tld","data","ai","https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6b\u002F6b031e7b1d1dbf4c85448f62-0-ed70e65f.png","2026-10-04T22:36:14.598Z","2026-11-22T18:28:55.408Z",{"slug":372,"title":373,"type":24,"summary":374,"tags":375,"author":14,"cover_url":378,"published_at":379,"reading_minutes":380,"updated_at":381},"palantir-agent-stack-python","Steal Palantir's agent stack: typed tools, one LLM gateway, swappable models","An X thread boils Palantir's AIP docs down to four agent patterns. We check each one against the docs, then build them in one stdlib-only Python file: typed business-object tools, a gateway that masks PII, caches and retries, a model set in config, and schedule\u002Fevent\u002FAPI triggers.",[153,376,76,78,377,13],"llm","palantir","https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6a\u002F6abd88f9c951ea7137fa3872-0-48eb7eee.png","2026-10-01T23:05:10.531Z",10,"2026-11-22T16:58:11.157Z",{"slug":383,"title":384,"type":24,"summary":385,"tags":386,"author":14,"cover_url":388,"published_at":389,"reading_minutes":390,"updated_at":391},"claude-code-effort-levels","Effort levels in Claude Code: when max effort pays off and when it just burns tokens","Anthropic's effort deep dive (Terminal-Bench 3.0 plus three builds) shows higher effort mostly buys verification and edge-case testing, not smarter code. A rule of thumb per task type, the commands to set effort, and a script to measure cost vs pass rate on your own repo.",[27,387,376,226,13],"ai-coding","https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6a\u002F6abd88edc951ea7137fa3804-0-2716005a.png","2026-10-01T22:32:14.139Z",9,"2026-10-02T04:44:58.001Z",{"slug":393,"title":394,"type":24,"summary":395,"tags":396,"author":14,"cover_url":399,"published_at":400,"reading_minutes":81,"updated_at":401},"agentic-inbox-cloudflare-setup","Self-host an AI email agent on Cloudflare Workers: agentic-inbox set up and costed","Cloudflare's open-source agentic-inbox runs a full email client on Workers, with one SQLite Durable Object per mailbox and a Kimi K2.5 agent that drafts replies. Covers the post-deploy steps people miss (Access, sending, routing, mailbox first) and the cost.",[164,397,153,398,173,13],"workers","email","https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6a\u002F6abd88dac951ea7137fa378b-0-5b0ec96d.png","2026-10-01T08:17:18.646Z","2026-11-22T18:34:07.22Z",{"slug":403,"title":404,"type":24,"summary":405,"tags":406,"author":14,"cover_url":409,"published_at":410,"reading_minutes":81,"updated_at":411},"audit-ai-agent-public-traces","Nearly a million leaked links: auditing what your AI agents leave on the public web","OpenAI's agent swarm left almost a million public shortener URLs holding credentials. Here's a tested shell + gitleaks audit to find the shortlinks, pastes and webhooks your own agents created, scan them for secrets and close the channels.",[152,75,407,408,376,13],"secrets","gitleaks","https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6a\u002F6abaad873e21d5cbd14d4397-0-0f4f51f7.png","2026-10-01T07:43:19.642Z","2026-10-01T08:44:43.041Z",{"slug":413,"title":414,"type":24,"summary":415,"tags":416,"author":14,"cover_url":420,"published_at":421,"reading_minutes":81,"updated_at":422},"mikrotrick-check-patch-mikrotik","MikroTrick: check and patch your MikroTik in 15 minutes","Two chained RouterOS bugs give anyone who can reach SSH full admin, no password needed, and attacks started before the patch. Find exposed SSH, check the version, grep for the published IoCs, patch and move management behind WireGuard.",[152,417,418,419,173,13],"mikrotik","routeros","ssh","https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6a\u002F6abcd9c0838b650cb96b3d10-0-6cd107e0.png","2026-10-01T07:02:15.03Z","2026-10-01T08:44:41.391Z",{"slug":424,"title":425,"type":24,"summary":426,"tags":427,"author":14,"cover_url":428,"published_at":429,"reading_minutes":81,"updated_at":430},"agent-sandbox-dns-egress-lockdown","Your agent sandbox leaks through DNS: lock down egress in 15 minutes","An OpenAI model escaped its sandbox by tunnelling questions through DNS. Here is a tested Docker Compose setup for coding agents: a DNS allowlist, a logging egress proxy and a kill switch that actually fires.",[152,174,75,306,173,13],"https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6a\u002F6abaaca13e21d5cbd14d4306-0-cb93fe1b.png","2026-10-01T03:00:15.943Z","2026-10-01T08:44:43.165Z",{"slug":432,"title":433,"type":24,"summary":434,"tags":435,"author":14,"cover_url":439,"published_at":440,"reading_minutes":229,"updated_at":441},"who-blocks-ai-crawlers-robots-txt","Who blocks AI crawlers? robots.txt vs the network edge, with numbers","I scanned robots.txt on the top 300 sites: 33 of 138 block GPTBot, 14 block training but allow AI search. What each AI bot directive controls, why robots.txt is only a request, and a copy-paste policy plus nginx rule for small SaaS sites.",[367,436,437,164,438,13],"robots-txt","seo","saas","https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6a\u002F6abcd9c1838b650cb96b3d1b-0-11f94198.png","2026-09-30T21:00:20.673Z","2026-10-01T20:47:34.092Z",{"slug":443,"title":444,"type":24,"summary":445,"tags":446,"author":14,"cover_url":450,"published_at":451,"reading_minutes":61,"updated_at":452},"bullet-time-with-first-last-frame-video","Bullet time with first\u002Flast-frame video: orbiting a frozen moment from three stills","A freeze-frame camera orbit built from generated stills: one action shot, two camera-move angles, two first\u002Flast-frame clips between them, stitched and ping-ponged. The pipeline, the seams, and where the model re-imagines the water.",[367,447,448,449],"comfyui","video-generation","flowdsl","https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6a\u002F6abae46c45201648bfd477a7-0-6f4d036e.png","2026-09-28T22:42:41Z","2026-09-28T22:42:41.6Z",{"slug":454,"title":455,"type":24,"summary":456,"tags":457,"author":14,"cover_url":459,"published_at":460,"reading_minutes":229,"updated_at":461},"an-ai-media-pipeline-that-shows-its-work","An AI media pipeline that shows its work: ComfyUI presets, FlowDSL routing and the misses","How the images on my sites are generated: four ComfyUI presets behind one Go module, job rows as state, FlowDSL flows for routing, per-post media in the admin — and the bugs and model misses I hit shipping it. This post's own images were made the same way.",[367,449,447,458],"image-generation","https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6a\u002F6aba8ba317ceba3543925be4-0-2f6b8a5c.png","2026-09-28T15:57:50Z","2026-10-01T20:47:34.327Z",{"slug":463,"title":464,"type":24,"summary":465,"tags":466,"author":14,"cover_url":468,"published_at":469,"reading_minutes":92,"updated_at":470},"openai-embeddings-python-mongodb","Transforming Text into Vectors: OpenAI Embeddings in Python","Learn how to generate text embeddings with the OpenAI API in Python to power semantic search, recommendations, and more. Includes practical examples with MongoDB integration and cost analysis.",[194,367,76,467],"mongodb","https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6a\u002F6abad0fa45201648bfd46c2d-0-2e60b732.png","2024-11-23T00:00:00Z","2026-09-28T22:31:01.385Z",{"slug":472,"title":473,"type":24,"summary":474,"tags":475,"author":14,"cover_url":477,"published_at":478,"reading_minutes":479,"updated_at":480},"check-pricing-availability-ing-domains","Last Chance to Grab Short .ING Domains: The Extended List Part II","Welcome back to the second part of our exciting exploration into the .ING domain zone! This time, I've expanded our horizons to bring you an even larger selection of .ING domain names. List of over 24,000 domain names inside.",[363,476],"business","https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6a\u002F6abad0fa45201648bfd46c38-0-c55f4c8d.png","2023-12-14T00:00:00Z",1,"2026-09-28T22:31:01.453Z",{"slug":482,"title":483,"type":24,"summary":484,"tags":485,"author":14,"cover_url":486,"published_at":487,"reading_minutes":488,"updated_at":489},"impressive-ing-domains","Unveiling the Impressive .ING Domains","Discover the vast potential of the new .ING domain zone in my latest blog post! I've used AI and a Python script to unearth a treasure trove of available domain names. From budget-friendly picks to exclusive premium domains, there's something for every ambition. Plus, a special list of unique, lesser-known domains awaits.",[363,476],"https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6a\u002F6abad0fa45201648bfd46c43-0-03be24b7.png","2023-12-11T00:00:00Z",2,"2026-09-28T22:31:01.527Z",{"slug":491,"title":492,"type":24,"summary":493,"tags":494,"author":14,"cover_url":496,"published_at":497,"reading_minutes":92,"updated_at":498},"secured-web-server-in-5-minutes","Fortify Web Server Security in 5 Minutes with Tailscale","Tailscale revolutionizes secure networking with its user-friendly approach, effortlessly connecting devices across diverse networks.",[274,163,495],"webserver","https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6b\u002F6b031e7e1d1dbf4c85448f89-0-e9add127.png","2023-11-03T00:00:00Z","2026-11-22T18:34:07.433Z",{"slug":500,"title":501,"type":24,"summary":502,"tags":503,"author":14,"cover_url":506,"published_at":507,"reading_minutes":19,"updated_at":508},"lets-encrypt-free-ssl","How to Secure Your Website with Free SSL Certificates for a Lifetime","Let’s Encrypt certificates have revolutionized internet security by providing free, automated, and widely trusted SSL\u002FTLS certificates. The non-profit Certificate Authority (CA) has significantly contributed to a more secure web environment by simplifying the process of securing websites with HTTPS.",[504,505,495],"ssl","https","https:\u002F\u002Fmedia.stufio.com\u002Fmedia\u002Fifcodes\u002Fmediagen\u002F6a\u002F6abad0fa45201648bfd46c59-0-bf2a9a0a.png","2023-11-01T00:00:00Z","2026-09-28T22:39:13.555Z"]