Moon-Beam 8B-A1B preview out; terrabeam one-command section

README.md

@@ -103,7 +103,7 @@ Every number on our cards is held out, paired and graded by a fixed outside judg
 <p class="ds">The Moon shines with borrowed light. Moon-Beam models are trained to read the Tablinum library: answer when a passage states it, decline when it doesn't.</p>
 <div class="rung"><span><b>16B-A3B</b> GGUF IQ2–IQ4</span><span class="g">EngramBench +73.3</span></div>
 <div class="rung"><span><b>16B-A3B</b> MLX · BF16</span><span><a href="https://huggingface.co/tablinum/Moon-Beam-16B-A3B-MLX-4bit">Mac</a> · <a href="https://huggingface.co/tablinum/Moon-Beam-16B-A3B">weights</a></span></div>
-<div class="rung"><span><b>8B-A1B</b> thinking, tools, 256K</span><span class="s">in training</span></div>
+<div class="rung"><span><b>8B-A1B</b> GGUF · MLX · FP8</span><span class="s"><a href="https://huggingface.co/tablinum/Moon-Beam-8B-A1B-GGUF-Preview" style="color:var(--sun)">preview</a></span></div>
 </div>
 <div class="line sun">
 <div class="orb sun"></div>
@@ -140,12 +140,24 @@ model with one instruction: use a passage only if it states the answer. Two real
 <section>
 <h2>On the bench now</h2>
 <div class="bench">
-<div class="card"><p class="st">In training</p><h3>Moon-Beam 8B-A1B</h3>Ling-3.0-tiny (1.3B active, MIT) taught to use the library while keeping everything it already does: thinking on or off, preserved thinking, tool calls, 256K context. Practice data from its own checked reasoning, tool calls, web pages and customer-service tasks.</div>
+<div class="card"><p class="st">Preview out</p><h3><a href="https://huggingface.co/tablinum/Moon-Beam-8B-A1B-GGUF-Preview" style="color:var(--white)">Moon-Beam 8B-A1B</a></h3>Ling-3.0-tiny (1.3B active, MIT) taught to use the library while keeping thinking on or off, preserved thinking, tool calls and 128K context. Out now as a preview in <a href="https://huggingface.co/tablinum/Moon-Beam-8B-A1B-GGUF-Preview">GGUF</a>, <a href="https://huggingface.co/tablinum/Moon-Beam-8B-A1B-MLX-Preview">MLX</a>, <a href="https://huggingface.co/tablinum/Moon-Beam-8B-A1B-FP8-Preview">FP8</a> and <a href="https://huggingface.co/tablinum/Moon-Beam-8B-A1B-Preview">BF16</a>, reading Tablinum-2 until Tablinum-3 is tested.</div>
 <div class="card"><p class="st">Building</p><h3>Tablinum-3</h3>About twice the library. Humanities first (primary texts, full-text papers), plus an <b style="color:var(--white)">agent</b> domain for models that use computers: sysadmin and desktop Q&amp;A, command-line references, developer tooling, app manuals, web accessibility.</div>
 <div class="card"><p class="st">Pilot</p><h3>Sun-Beam</h3>Can a 3B model carry the library inside itself, trained on one workstation? First test: one subject's facts trained into Nanbeige's hashed n-gram memory, recalled with no search at all.</div>
 </div>
-<p style="margin:14px 0 0;max-width:74ch">Coming with the next release: <b style="color:var(--white)">one command</b> that installs the model and its library together and gives you a normal
-OpenAI-style endpoint on your machine. No setup to learn.</p>
+</section>
+
+<section>
+<h2>One command: terrabeam</h2>
+<p style="margin:0 0 12px;max-width:74ch"><a href="https://github.com/chadstthomas-svg/terrabeam"><b>terrabeam</b></a> installs a model and its library together and gives you a normal OpenAI-style
+endpoint on your machine. It works with our models, <b style="color:var(--white)">any GGUF on Hugging Face</b>, or a server you already run.</p>
+<div class="card" style="font-family:var(--mono);font-size:12.5px;line-height:1.7;color:var(--white);overflow-x:auto;white-space:pre">curl -fsSL https://raw.githubusercontent.com/chadstthomas-svg/terrabeam/master/install.sh | sh
+terrabeam run tablinum/Moon-Beam-8B-A1B-GGUF-Preview      <span style="color:var(--lav2)"># → http://127.0.0.1:8080/v1</span></div>
+<div class="ex" style="margin-top:12px">
+<div class="card"><p class="st" style="font-family:var(--mono);font-size:10.5px;letter-spacing:.14em;text-transform:uppercase;color:var(--teal);font-weight:700;margin:0">Any app</p>
+Add the library to Ollama, LM Studio or vLLM with <code>terrabeam proxy</code>, or give Claude Code, Cursor and other agents a <code>search_library</code> tool with <code>terrabeam mcp</code>. <a href="https://github.com/chadstthomas-svg/terrabeam/blob/master/docs/harnesses.md">Setup for Open WebUI and others</a>.</div>
+<div class="card"><p class="st" style="font-family:var(--mono);font-size:10.5px;letter-spacing:.14em;text-transform:uppercase;color:var(--teal);font-weight:700;margin:0">Memory · MIT</p>
+<code>--memory long-term</code> keeps notes across conversations, with secrets redacted. Open source under MIT: build on it freely, and a credit or thank-you is appreciated.</div>
+</div>
 </section>
 
 <section>

index.html

@@ -101,7 +101,7 @@ Every number on our cards is held out, paired and graded by a fixed outside judg
 <p class="ds">The Moon shines with borrowed light. Moon-Beam models are trained to read the Tablinum library: answer when a passage states it, decline when it doesn't.</p>
 <div class="rung"><span><b>16B-A3B</b> GGUF IQ2–IQ4</span><span class="g">EngramBench +73.3</span></div>
 <div class="rung"><span><b>16B-A3B</b> MLX · BF16</span><span><a href="https://huggingface.co/tablinum/Moon-Beam-16B-A3B-MLX-4bit">Mac</a> · <a href="https://huggingface.co/tablinum/Moon-Beam-16B-A3B">weights</a></span></div>
-<div class="rung"><span><b>8B-A1B</b> thinking, tools, 256K</span><span class="s">in training</span></div>
+<div class="rung"><span><b>8B-A1B</b> GGUF · MLX · FP8</span><span class="s"><a href="https://huggingface.co/tablinum/Moon-Beam-8B-A1B-GGUF-Preview" style="color:var(--sun)">preview</a></span></div>
 </div>
 <div class="line sun">
 <div class="orb sun"></div>
@@ -138,12 +138,24 @@ model with one instruction: use a passage only if it states the answer. Two real
 <section>
 <h2>On the bench now</h2>
 <div class="bench">
-<div class="card"><p class="st">In training</p><h3>Moon-Beam 8B-A1B</h3>Ling-3.0-tiny (1.3B active, MIT) taught to use the library while keeping everything it already does: thinking on or off, preserved thinking, tool calls, 256K context. Practice data from its own checked reasoning, tool calls, web pages and customer-service tasks.</div>
+<div class="card"><p class="st">Preview out</p><h3><a href="https://huggingface.co/tablinum/Moon-Beam-8B-A1B-GGUF-Preview" style="color:var(--white)">Moon-Beam 8B-A1B</a></h3>Ling-3.0-tiny (1.3B active, MIT) taught to use the library while keeping thinking on or off, preserved thinking, tool calls and 128K context. Out now as a preview in <a href="https://huggingface.co/tablinum/Moon-Beam-8B-A1B-GGUF-Preview">GGUF</a>, <a href="https://huggingface.co/tablinum/Moon-Beam-8B-A1B-MLX-Preview">MLX</a>, <a href="https://huggingface.co/tablinum/Moon-Beam-8B-A1B-FP8-Preview">FP8</a> and <a href="https://huggingface.co/tablinum/Moon-Beam-8B-A1B-Preview">BF16</a>, reading Tablinum-2 until Tablinum-3 is tested.</div>
 <div class="card"><p class="st">Building</p><h3>Tablinum-3</h3>About twice the library. Humanities first (primary texts, full-text papers), plus an <b style="color:var(--white)">agent</b> domain for models that use computers: sysadmin and desktop Q&amp;A, command-line references, developer tooling, app manuals, web accessibility.</div>
 <div class="card"><p class="st">Pilot</p><h3>Sun-Beam</h3>Can a 3B model carry the library inside itself, trained on one workstation? First test: one subject's facts trained into Nanbeige's hashed n-gram memory, recalled with no search at all.</div>
 </div>
-<p style="margin:14px 0 0;max-width:74ch">Coming with the next release: <b style="color:var(--white)">one command</b> that installs the model and its library together and gives you a normal
-OpenAI-style endpoint on your machine. No setup to learn.</p>
+</section>
+
+<section>
+<h2>One command: terrabeam</h2>
+<p style="margin:0 0 12px;max-width:74ch"><a href="https://github.com/chadstthomas-svg/terrabeam"><b>terrabeam</b></a> installs a model and its library together and gives you a normal OpenAI-style
+endpoint on your machine. It works with our models, <b style="color:var(--white)">any GGUF on Hugging Face</b>, or a server you already run.</p>
+<div class="card" style="font-family:var(--mono);font-size:12.5px;line-height:1.7;color:var(--white);overflow-x:auto;white-space:pre">curl -fsSL https://raw.githubusercontent.com/chadstthomas-svg/terrabeam/master/install.sh | sh
+terrabeam run tablinum/Moon-Beam-8B-A1B-GGUF-Preview      <span style="color:var(--lav2)"># → http://127.0.0.1:8080/v1</span></div>
+<div class="ex" style="margin-top:12px">
+<div class="card"><p class="st" style="font-family:var(--mono);font-size:10.5px;letter-spacing:.14em;text-transform:uppercase;color:var(--teal);font-weight:700;margin:0">Any app</p>
+Add the library to Ollama, LM Studio or vLLM with <code>terrabeam proxy</code>, or give Claude Code, Cursor and other agents a <code>search_library</code> tool with <code>terrabeam mcp</code>. <a href="https://github.com/chadstthomas-svg/terrabeam/blob/master/docs/harnesses.md">Setup for Open WebUI and others</a>.</div>
+<div class="card"><p class="st" style="font-family:var(--mono);font-size:10.5px;letter-spacing:.14em;text-transform:uppercase;color:var(--teal);font-weight:700;margin:0">Memory · MIT</p>
+<code>--memory long-term</code> keeps notes across conversations, with secrets redacted. Open source under MIT: build on it freely, and a credit or thank-you is appreciated.</div>
+</div>
 </section>
 
 <section>