<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: Ryan Sharma</title>
    <description>The latest articles on DEV Community by Ryan Sharma (@rrhinos).</description>
    <link>https://dev.to/rrhinos</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4245008%2Fe1edb865-b71c-48e3-817b-392374b4b805.png</url>
      <title>DEV Community: Ryan Sharma</title>
      <link>https://dev.to/rrhinos</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/rrhinos"/>
    <language>en</language>
    <item>
      <title>I got a RAG app running on a 512 MB free server. The trick was not running the embedding model at all.</title>
      <dc:creator>Ryan Sharma</dc:creator>
      <pubDate>Sat, 14 Nov 2026 19:58:45 +0000</pubDate>
      <link>https://dev.to/rrhinos/i-got-a-rag-app-running-on-a-512-mb-free-server-the-trick-was-not-running-the-embedding-model-at-1m2h</link>
      <guid>https://dev.to/rrhinos/i-got-a-rag-app-running-on-a-512-mb-free-server-the-trick-was-not-running-the-embedding-model-at-1m2h</guid>
      <description>&lt;p&gt;Most "build a RAG app" tutorials work fine in a notebook. Then you try to put one online on a free host, and it falls over.&lt;/p&gt;

&lt;p&gt;I wanted a document Q&amp;amp;A app that answers from my own files, cites its sources, refuses when the answer isn't there, and runs &lt;strong&gt;entirely on free tiers with no credit card&lt;/strong&gt;. Here's what broke along the way, and the setup that finally worked.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Live demo:&lt;/strong&gt; &lt;a href="https://rag-0.onrender.com" rel="noopener noreferrer"&gt;https://rag-0.onrender.com&lt;/a&gt; (free tier, so the first request after it's been idle takes about a minute to wake up)&lt;br&gt;
&lt;strong&gt;Code (free, 50 lines):&lt;/strong&gt; &lt;a href="https://github.com/Ryan140-cloud/zero-cost-rag" rel="noopener noreferrer"&gt;https://github.com/Ryan140-cloud/zero-cost-rag&lt;/a&gt;&lt;/p&gt;
&lt;h2&gt;
  
  
  Problem 1: the embedding model doesn't fit
&lt;/h2&gt;

&lt;p&gt;Render's free web service gives you &lt;strong&gt;512 MB of RAM and 0.1 CPU&lt;/strong&gt;. The usual tutorial setup embeds text with &lt;code&gt;sentence-transformers&lt;/code&gt;, which pulls in &lt;strong&gt;PyTorch&lt;/strong&gt;. PyTorch plus a model is a lot to fit in 512 MB, and on a host that small it typically either runs out of memory or crawls.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;The fix: don't run an embedding model on your server at all.&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;Qdrant Cloud has a free cluster, and its &lt;strong&gt;Cloud Inference&lt;/strong&gt; feature embeds text on Qdrant's side. &lt;code&gt;sentence-transformers/all-minilm-l6-v2&lt;/code&gt; (384 dimensions) is listed as free. Your server just sends plain text:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight python"&gt;&lt;code&gt;&lt;span class="k"&gt;from&lt;/span&gt; &lt;span class="n"&gt;qdrant_client&lt;/span&gt; &lt;span class="k"&gt;import&lt;/span&gt; &lt;span class="nc"&gt;QdrantClient&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;models&lt;/span&gt;

&lt;span class="n"&gt;client&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="nc"&gt;QdrantClient&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;url&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="no"&gt;QDRANT_URL&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;api_key&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="no"&gt;QDRANT_API_KEY&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
                      &lt;span class="n"&gt;cloud_inference&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="kc"&gt;True&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;   &lt;span class="c1"&gt;# Qdrant embeds, not your server&lt;/span&gt;

&lt;span class="no"&gt;MODEL&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="s2"&gt;"sentence-transformers/all-minilm-l6-v2"&lt;/span&gt;

&lt;span class="c1"&gt;# storing a chunk: send text, not a vector&lt;/span&gt;
&lt;span class="n"&gt;client&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;upsert&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s2"&gt;"docs"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;points&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="n"&gt;models&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nc"&gt;PointStruct&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;
    &lt;span class="nb"&gt;id&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="n"&gt;chunk_id&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
    &lt;span class="n"&gt;vector&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="n"&gt;models&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nc"&gt;Document&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;text&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="n"&gt;chunk_text&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;model&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="no"&gt;MODEL&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
    &lt;span class="n"&gt;payload&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="s2"&gt;"text"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="n"&gt;chunk_text&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="s2"&gt;"source"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"guide.pdf"&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;

&lt;span class="c1"&gt;# searching: send the question as text too&lt;/span&gt;
&lt;span class="n"&gt;hits&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="n"&gt;client&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;query_points&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s2"&gt;"docs"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
    &lt;span class="n"&gt;query&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="n"&gt;models&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nc"&gt;Document&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;text&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="n"&gt;question&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;model&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="no"&gt;MODEL&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
    &lt;span class="n"&gt;limit&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="mi"&gt;3&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;with_payload&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="kc"&gt;True&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;points&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;With no PyTorch in the dependencies, my full Flask app used about &lt;strong&gt;141 MB of RAM&lt;/strong&gt; when I measured it locally. Plenty of headroom.&lt;/p&gt;

&lt;p&gt;One catch: LangChain's &lt;code&gt;QdrantVectorStore&lt;/code&gt; expects a local embeddings object, which defeats the point. So I wrote a ~25-line custom LangChain retriever that passes raw text to &lt;code&gt;query_points&lt;/code&gt; instead.&lt;/p&gt;

&lt;h2&gt;
  
  
  Problem 2: the Groq model in every tutorial is gone
&lt;/h2&gt;

&lt;p&gt;Groq is the go-to free LLM API for these projects, and most tutorials hard-code &lt;code&gt;llama-3.1-8b-instant&lt;/code&gt;. Groq &lt;strong&gt;shut that model down for free and developer-tier accounts on August 16, 2026&lt;/strong&gt;. Their recommended replacement is &lt;code&gt;openai/gpt-oss-20b&lt;/code&gt;. Old code now just returns &lt;code&gt;404 model_not_found&lt;/code&gt;.&lt;/p&gt;

&lt;p&gt;Two lessons:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Never hard-code the model name.&lt;/strong&gt; Read it from an environment variable, so the next retirement is a config change, not a code change.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;gpt-oss is a reasoning model.&lt;/strong&gt; It "thinks" before answering, and those thinking tokens count toward &lt;code&gt;max_tokens&lt;/code&gt;. A low cap like 512 can give you cut-off or empty answers. What worked for me:
&lt;/li&gt;
&lt;/ol&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight python"&gt;&lt;code&gt;&lt;span class="k"&gt;from&lt;/span&gt; &lt;span class="n"&gt;langchain_groq&lt;/span&gt; &lt;span class="k"&gt;import&lt;/span&gt; &lt;span class="nc"&gt;ChatGroq&lt;/span&gt;

&lt;span class="n"&gt;llm&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="nc"&gt;ChatGroq&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;
    &lt;span class="n"&gt;model&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="n"&gt;os&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;getenv&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s2"&gt;"GROQ_MODEL"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="s2"&gt;"openai/gpt-oss-20b"&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
    &lt;span class="n"&gt;temperature&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="mi"&gt;0&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
    &lt;span class="n"&gt;reasoning_effort&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="s2"&gt;"low"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;   &lt;span class="c1"&gt;# faster and cheaper; plenty for answering from context&lt;/span&gt;
    &lt;span class="n"&gt;max_tokens&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="mi"&gt;1024&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;          &lt;span class="c1"&gt;# thinking tokens count here too&lt;/span&gt;
    &lt;span class="n"&gt;max_retries&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="mi"&gt;4&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;            &lt;span class="c1"&gt;# backs off and retries on HTTP 429&lt;/span&gt;
&lt;span class="p"&gt;)&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h2&gt;
  
  
  Problem 3: the limit that actually runs out
&lt;/h2&gt;

&lt;p&gt;Groq's free tier for &lt;code&gt;openai/gpt-oss-20b&lt;/code&gt; (check their rate-limits page, these change):&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Limit&lt;/th&gt;
&lt;th&gt;Value&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Requests / minute&lt;/td&gt;
&lt;td&gt;30&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Requests / day&lt;/td&gt;
&lt;td&gt;1,000&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Tokens / minute&lt;/td&gt;
&lt;td&gt;8,000&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Tokens / day&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;&lt;strong&gt;200,000&lt;/strong&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;In RAG, every question carries the retrieved chunks with it. In my real test runs (3 chunks of ~800 characters, short answers), one question cost &lt;strong&gt;about 600–700 tokens&lt;/strong&gt;. That's roughly &lt;strong&gt;280 questions a day&lt;/strong&gt; before the daily token cap, well before the 1,000-request cap.&lt;/p&gt;

&lt;p&gt;That's plenty for a portfolio project or demo. It's not enough for real public traffic, and it's worth being honest about that.&lt;/p&gt;

&lt;h2&gt;
  
  
  Problem 4: the two sleep timers
&lt;/h2&gt;

&lt;p&gt;Free apps rarely die from traffic. They die from inactivity:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Timer&lt;/th&gt;
&lt;th&gt;What happens&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Render: 15 min idle&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;The service spins down; the next request waits ~1 minute&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Qdrant: 1 week unused&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;The free cluster is &lt;strong&gt;suspended&lt;/strong&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Qdrant: 4 weeks unused&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;The suspended cluster is &lt;strong&gt;deleted&lt;/strong&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;The Render one is annoying. The Qdrant one is dangerous: a portfolio link you shared a month ago can quietly lose all its data.&lt;/p&gt;

&lt;p&gt;My fix is a free weekly GitHub Actions job that hits a "deep" health endpoint. That endpoint does one cheap Qdrant read (a point count) and &lt;strong&gt;never calls the LLM&lt;/strong&gt;, so it can't burn your token budget:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight yaml"&gt;&lt;code&gt;&lt;span class="na"&gt;on&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
  &lt;span class="na"&gt;schedule&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
    &lt;span class="p"&gt;-&lt;/span&gt; &lt;span class="na"&gt;cron&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"17 6 * * 1"&lt;/span&gt;   &lt;span class="c1"&gt;# Mondays
&lt;/span&gt;&lt;span class="na"&gt;jobs&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
  &lt;span class="na"&gt;ping&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
    &lt;span class="na"&gt;runs-on&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s"&gt;ubuntu-latest&lt;/span&gt;
    &lt;span class="na"&gt;steps&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
      &lt;span class="p"&gt;-&lt;/span&gt; &lt;span class="na"&gt;run&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s"&gt;curl --fail --max-time 120 --retry 3 "${{ secrets.APP_URL }}/health?deep=1"&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h2&gt;
  
  
  Problem 5: the bug my own first deploy hit
&lt;/h2&gt;

&lt;p&gt;When I pasted my Qdrant API key into Render's environment settings, a &lt;strong&gt;trailing newline&lt;/strong&gt; came along with it. Every Qdrant call failed, and the error message printed the full key into my logs. I had to rotate it.&lt;/p&gt;

&lt;p&gt;Now every setting is &lt;code&gt;.strip()&lt;/code&gt;ped when it's read:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight python"&gt;&lt;code&gt;&lt;span class="k"&gt;def&lt;/span&gt; &lt;span class="nf"&gt;env&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;name&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;default&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="s2"&gt;""&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
    &lt;span class="k"&gt;return&lt;/span&gt; &lt;span class="n"&gt;os&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;getenv&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;name&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;default&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;strip&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h2&gt;
  
  
  Grounding: make it refuse
&lt;/h2&gt;

&lt;p&gt;The prompt tells the model to answer &lt;strong&gt;only&lt;/strong&gt; from the retrieved context, or reply with one exact sentence:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;Answer ONLY from the context below. If the context does not contain the
answer, reply exactly: "I can't answer that from the provided documents."
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Because the refusal is one exact string, you can &lt;strong&gt;test&lt;/strong&gt; it. I keep 10 questions (6 my documents answer, 4 they don't, including a prompt-injection attempt) and check every answer automatically before deploying.&lt;/p&gt;

&lt;h2&gt;
  
  
  The stack, summarized
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;LLM:&lt;/strong&gt; Groq &lt;code&gt;openai/gpt-oss-20b&lt;/code&gt; (free tier)&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Embeddings + vectors:&lt;/strong&gt; Qdrant Cloud free cluster with Cloud Inference&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Glue:&lt;/strong&gt; LangChain 1.x (&lt;code&gt;langchain-core&lt;/code&gt;, &lt;code&gt;langchain-groq&lt;/code&gt;)&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;API + hosting:&lt;/strong&gt; Flask + Gunicorn on Render's free web service&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Cost:&lt;/strong&gt; $0, no credit card anywhere&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The 50-line version is on GitHub: &lt;a href="https://github.com/Ryan140-cloud/zero-cost-rag" rel="noopener noreferrer"&gt;https://github.com/Ryan140-cloud/zero-cost-rag&lt;/a&gt;&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Disclosure: I also turned the full build (Flask API, Render deploy, keep-alive, tests and evaluation) into a paid guide and starter kit. If you'd rather not piece it together yourself, it's &lt;a href="https://whop.com/ryans-0fb0/ship-a-rag-app-for-0-build-deploy-test-on-free-tiers/" rel="noopener noreferrer"&gt;here&lt;/a&gt;. There's also a &lt;a href="https://whop.com/ryans-0fb0/the-0-rag-cheat-sheet-free-tier-limits-traps-fixes/" rel="noopener noreferrer"&gt;free 3-page cheat sheet&lt;/a&gt; of all the limits above.&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;Free tiers change. Everything here was checked in November 2026.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>rag</category>
      <category>python</category>
      <category>langchain</category>
      <category>ai</category>
    </item>
  </channel>
</rss>
