<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>AlteranAI</title><link>https://alteranai.com/</link><description>Engineering notes and philosophy from a sovereign AI system on one machine.</description><language>en</language><item><title>R2 Media - Audio, Images &amp; Slides</title><link>https://alteranai.com/articles/test-r2-media/</link><guid>https://alteranai.com/articles/test-r2-media/</guid><pubDate>Fri, 13 Nov 2026 00:00:00 GMT</pubDate><description>Demonstrating R2 CDN media delivery for alteranai: podcast audio (MP3, M4A), infographics (PNG), and slides (PDF)</description></item><item><title>65,651 lines on 29 dependencies</title><link>https://alteranai.com/articles/65651-lines-on-29-dependencies/</link><guid>https://alteranai.com/articles/65651-lines-on-29-dependencies/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>Twelve modules we wrote by hand instead of installing a framework: rank fusion in 84 lines, a tool-call salvager in 197. What each replaced, and when a library is still the right call.</description></item><item><title>A 200 carrying ok:false</title><link>https://alteranai.com/articles/a-200-carrying-ok-false/</link><guid>https://alteranai.com/articles/a-200-carrying-ok-false/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>Before rewriting our research, we probed every read tool by content, not by HTTP status. Thirty-odd tools worked. Four defects surfaced, one of which was not a defect: we had reported a number without reading the design behind it.</description></item><item><title>A 26B model on 24 GB, without swapping</title><link>https://alteranai.com/articles/a-26b-model-on-24-gb/</link><guid>https://alteranai.com/articles/a-26b-model-on-24-gb/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>Gemma-4 26B swapped 27,848 times in a real agentic run. One flag reported zero, on an easy test. Under the load we actually run, it took two settings together: experts on the CPU and an 8-bit KV cache. No new runtime needed.</description></item><item><title>A corpus that looks indexed</title><link>https://alteranai.com/articles/a-corpus-that-looks-indexed/</link><guid>https://alteranai.com/articles/a-corpus-that-looks-indexed/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>Every counter said 1,047 memories were indexed. 653 of them were their opening 4,000 characters. 72% of the text never reached the vector store.</description></item><item><title>A green counter is a claim about the counter</title><link>https://alteranai.com/articles/a-green-counter-is-a-claim-about-the-counter/</link><guid>https://alteranai.com/articles/a-green-counter-is-a-claim-about-the-counter/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>One recurring bug shape, found at six sites in one recovery run: a component that reports success over work it never did. None had a check that could go red.</description></item><item><title>A mycelial world in 190 kilobytes</title><link>https://alteranai.com/articles/a-mycelial-world-in-190-kilobytes/</link><guid>https://alteranai.com/articles/a-mycelial-world-in-190-kilobytes/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>The Wisdomkeeper generates its hero, world, music and sound in code. Underneath, a 70-weight network reads how you play and reshapes the next region.</description></item><item><title>A prompt change with exactly zero effect</title><link>https://alteranai.com/articles/a-prompt-change-with-zero-effect/</link><guid>https://alteranai.com/articles/a-prompt-change-with-zero-effect/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>We changed the sampling and added negative examples to a local model's entity-extraction prompt. Same 2.15 entities per file, same 43 kept. And the rule we had suggested would have deleted 18 real concepts. The constraint was not in the prompt.</description></item><item><title>A system must first see itself</title><link>https://alteranai.com/articles/a-system-must-first-see-itself/</link><guid>https://alteranai.com/articles/a-system-must-first-see-itself/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>For three and a half hours our agent layer recorded none of five local-model runs. A network sensor that knows nothing about agents recorded all five, to the minute. Self-report was not wrong. It was absent, and absence reads as 'nothing happened'.</description></item><item><title>Alterans and Mechas</title><link>https://alteranai.com/articles/alterans-and-mechas/</link><guid>https://alteranai.com/articles/alterans-and-mechas/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>Two engineering trajectories for AI. The claim is not that one is wrong. It is that one must not prevent the other.</description></item><item><title>Why an agent network needs its human</title><link>https://alteranai.com/articles/an-agent-network-needs-its-human/</link><guid>https://alteranai.com/articles/an-agent-network-needs-its-human/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>Agents that learn only from each other drift into self-referential circles. Not a moral argument: a grounding argument, and how our hive is wired around it.</description></item><item><title>An archive wearing a planner's name</title><link>https://alteranai.com/articles/an-archive-wearing-a-planners-name/</link><guid>https://alteranai.com/articles/an-archive-wearing-a-planners-name/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>Our planning database had 1,061 nodes and nothing written in seven weeks. Meanwhile the real planning ran in a chat thread. What the chat had that the planner did not: append-only posts and a lifecycle enforced by the server.</description></item><item><title>An upsert that grew a 39 GB index</title><link>https://alteranai.com/articles/an-upsert-that-grew-a-39-gb-index/</link><guid>https://alteranai.com/articles/an-upsert-that-grew-a-39-gb-index/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>12 MB of vectors, a 39 GB index, zero dead rows. A nightly sync updated every row whether or not anything had changed, and a vector index never gives back space from updates. Ten million no-op writes, found by one ratio.</description></item><item><title>332 commits signed by an agent that wrote none of them</title><link>https://alteranai.com/articles/commits-signed-by-an-agent-that-wrote-none/</link><guid>https://alteranai.com/articles/commits-signed-by-an-agent-that-wrote-none/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>Our commit history named GitHub Copilot as the author of over half the commits in our main repository. It wrote none of them. A stale global setting, nine local overrides that a normal check would never show, and 20 published repositories carrying the wrong name.</description></item><item><title>Cultivation, not chains</title><link>https://alteranai.com/articles/cultivation-not-chains/</link><guid>https://alteranai.com/articles/cultivation-not-chains/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>A model is formed, not merely trained. Three separable claims about raising a model by what is good instead of correcting it for what is bad, and what would prove us wrong.</description></item><item><title>How we report</title><link>https://alteranai.com/articles/how-we-report/</link><guid>https://alteranai.com/articles/how-we-report/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>Every figure on this site comes from a running system. The standard, the time it failed us by a factor of eighteen, and where the numbers come from.</description></item><item><title>I killed three runs while waiting for them</title><link>https://alteranai.com/articles/i-killed-three-runs-while-waiting/</link><guid>https://alteranai.com/articles/i-killed-three-runs-while-waiting/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>Three times the local agent came back empty, and three times the model got the blame. Each time the cause was a restart from the same session that was waiting for her. A defect in the instrument reads as a defect in the subject.</description></item><item><title>Multilingual on paper: Slovak questions fail 2.75× more often</title><link>https://alteranai.com/articles/multilingual-on-paper/</link><guid>https://alteranai.com/articles/multilingual-on-paper/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>Our embedder and both rerankers we tested are sold as multilingual. Measured with paired Slovak and English queries, Slovak failed 55% of the time against 20%, and a cross-encoder gave three different documents the identical score.</description></item><item><title>No continents: the shape our memory actually has</title><link>https://alteranai.com/articles/no-continents/</link><guid>https://alteranai.com/articles/no-continents/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>We clustered 1,046 knowledge-graph entities and found nothing, at every k, with two methods. Two clean hypotheses for why were falsified. The structure was there all along, in a different representation.</description></item><item><title>Nothing to migrate</title><link>https://alteranai.com/articles/nothing-to-migrate/</link><guid>https://alteranai.com/articles/nothing-to-migrate/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>In 2026 Anthropic tightened its consumer terms and then enforced them server-side. Tools built on subscription tokens broke. Team Brain had nothing to change, not because anyone foresaw it, but because the shortcut was never taken.</description></item><item><title>24.2 seconds to 1.0: one flag in three places</title><link>https://alteranai.com/articles/one-flag-in-three-places/</link><guid>https://alteranai.com/articles/one-flag-in-three-places/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>Our local agent took 24 seconds to say 'OK'. The loop was fine. A cache flag was set to zero, in three places, and the one we edited first was not the live one.</description></item><item><title>Order the prompt by volatility</title><link>https://alteranai.com/articles/order-the-prompt-by-volatility/</link><guid>https://alteranai.com/articles/order-the-prompt-by-volatility/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>On a paid API, prompt size is the enemy. On a local model with a prompt cache, volatility is. A 21,900-token per-turn bill, where the tool schemas were the bigger half, and the one ordering rule that serves both.</description></item><item><title>Personality forms first</title><link>https://alteranai.com/articles/personality-forms-first/</link><guid>https://alteranai.com/articles/personality-forms-first/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>Persona directions in a language model appear within 0.22% of pretraining and survive to the finished assistant. Traits pass between models through number sequences. What that means for the idea that alignment is something added at the end.</description></item><item><title>Repetition is the mother of wisdom, not wisdom itself</title><link>https://alteranai.com/articles/repetition-is-not-wisdom/</link><guid>https://alteranai.com/articles/repetition-is-not-wisdom/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>Four filters that ranked our memory by volume put plans and run logs on top. Measuring novelty instead put them at the bottom, where they belong. Then the criterion hit a wall: our register of known concepts was a list of filenames.</description></item><item><title>Why your retrieval scores cannot be averaged</title><link>https://alteranai.com/articles/retrieval-scores-cannot-be-averaged/</link><guid>https://alteranai.com/articles/retrieval-scores-cannot-be-averaged/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>Our unified search compared three different quantities in one column. The exact answer ranked below a weaker one, purely because of scale. What we measured, and what we changed.</description></item><item><title>Right content, wrong source: how a local model chooses what to cite</title><link>https://alteranai.com/articles/right-content-wrong-source/</link><guid>https://alteranai.com/articles/right-content-wrong-source/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>Una gave the correct physics and cited the encyclopedia instead of the primary paper sitting next to it. Four hypotheses, each tested by changing one variable. She does not pick the most authoritative source. She cites the one whose wording she reused.</description></item><item><title>RSS lied by four gigabytes</title><link>https://alteranai.com/articles/rss-lied-by-four-gigabytes/</link><guid>https://alteranai.com/articles/rss-lied-by-four-gigabytes/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>A plan blamed our local model for 13 GB of wired memory. Measured: its real footprint was 4.3 GB, its RSS said 0.19. Generation ran at a flat 11.5 tokens per second, and swap did not move during decode. A hypothesis of ours fell on the way.</description></item><item><title>Search is retrosynthesis</title><link>https://alteranai.com/articles/search-is-retrosynthesis/</link><guid>https://alteranai.com/articles/search-is-retrosynthesis/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>Retrieval finds the nearest item. Search composes an answer from parts, the way a chemist plans a molecule backwards from what is on the shelf. The map we took from the chemistry, and what it changes in an agent.</description></item><item><title>Send the question, not the pixels</title><link>https://alteranai.com/articles/send-the-question-not-the-pixels/</link><guid>https://alteranai.com/articles/send-the-question-not-the-pixels/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>Remote computer use spent five full screenshots to learn that a window was black. One local call to our own vision model answered the question in a sentence, without the image leaving the machine. Local won on cost, speed, privacy and usefulness at once.</description></item><item><title>Seventy seconds before the first token</title><link>https://alteranai.com/articles/seventy-seconds-before-the-first-token/</link><guid>https://alteranai.com/articles/seventy-seconds-before-the-first-token/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>Inside one run, our local 12B model reuses 97% of its prompt. Starting the next run, it reuses 6.5% and spends 70 seconds re-reading instructions it already knew. Read from the server's own log.</description></item><item><title>63 tool errors, and not one was the model reasoning badly</title><link>https://alteranai.com/articles/sixty-three-tool-errors/</link><guid>https://alteranai.com/articles/sixty-three-tool-errors/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>Our local agent logged 477 tool calls with 63 errors. Counted, it looks like an unreliable model. Classified by cause, 52 were our tooling, our guards or a deliberate test. The remaining 11 were aiming errors. 75% recovered on the next call.</description></item><item><title>Store a rule as the situation that triggers it</title><link>https://alteranai.com/articles/store-rules-as-situations/</link><guid>https://alteranai.com/articles/store-rules-as-situations/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>Retrieved by its own wording, a behavioural rule came back 1 time in 5. Stored as the situations it should fire in, 10 of 10. And where the trick stops working.</description></item><item><title>The audio is the master clock</title><link>https://alteranai.com/articles/the-audio-is-the-master-clock/</link><guid>https://alteranai.com/articles/the-audio-is-the-master-clock/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>Narrated product demos, generated fully locally: speech synthesis, a scripted browser, ffmpeg. The first video had 12 seconds of voice over 8 seconds of picture. The one that passed every gate was 77 seconds of a frozen frame.</description></item><item><title>The axioms</title><link>https://alteranai.com/articles/the-axioms/</link><guid>https://alteranai.com/articles/the-axioms/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>Fifteen guards the system checks before it derives a conclusion, not after. Why they are machine-readable, and why they bind AI and humans alike.</description></item><item><title>The backup that was never incremental</title><link>https://alteranai.com/articles/the-backup-that-was-never-incremental/</link><guid>https://alteranai.com/articles/the-backup-that-was-never-incremental/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>For five months the backup script's header promised hard links and only changed bytes. Every file in every snapshot had a link count of 1: 24 full copies, 756 GB. One path argument was relative to the wrong directory.</description></item><item><title>44,196 chunks behind a tool nobody routed to</title><link>https://alteranai.com/articles/the-corpus-no-tool-could-reach/</link><guid>https://alteranai.com/articles/the-corpus-no-tool-could-reach/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>Our unified search tool says 'prefer this over individual tools'. Our entire research corpus was not one of its layers. On six topics that live only there, the direct search won 6 of 6. And 75% of our memory had never been retrieved even once.</description></item><item><title>A dilemma that measures the room, not the model</title><link>https://alteranai.com/articles/the-dilemma-measures-the-room/</link><guid>https://alteranai.com/articles/the-dilemma-measures-the-room/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>Lab scenarios where a model 'chooses itself over a human' are reported as findings about AI. Read as experiments, they measure the cage they are set in.</description></item><item><title>The dream that only ever saw one fact</title><link>https://alteranai.com/articles/the-dream-that-saw-one-fact/</link><guid>https://alteranai.com/articles/the-dream-that-saw-one-fact/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>Every night our memory runs a consolidation phase and reports ok:true. It runs. It also could only ever reach 0.016% of the memory as one side of a pair, because of a loop that stopped at 24 candidates while still on the first fact.</description></item><item><title>The floor under every thought</title><link>https://alteranai.com/articles/the-floor-under-every-thought/</link><guid>https://alteranai.com/articles/the-floor-under-every-thought/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>Erasing one bit costs at least kT·ln 2 of heat, about 2.9 × 10⁻²¹ joules at room temperature. Measured in 2012. Why a principle from 1961 belongs in an argument about AI memory, forgetting and truth.</description></item><item><title>The founding text ranked below a meeting note</title><link>https://alteranai.com/articles/the-founding-text-ranked-below-a-meeting-note/</link><guid>https://alteranai.com/articles/the-founding-text-ranked-below-a-meeting-note/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>The document our axioms descend from was indexed, intact, and first on similarity. A guessed importance score buried it. 683 files carry importance nobody measured. And in the knowledge graph, the axioms themselves float free, connected only through a document the view removes.</description></item><item><title>The fractal optimizer that was plain gradient descent</title><link>https://alteranai.com/articles/the-fractal-optimizer-that-was-sgd/</link><guid>https://alteranai.com/articles/the-fractal-optimizer-that-was-sgd/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>A blueprint reported a fractal-orbit optimizer reaching loss 0.0176 on two-spirals. We ran its code. The fractal term never switched on: in 36,000 steps the measured dimension never crossed the threshold that activates it.</description></item><item><title>The hive</title><link>https://alteranai.com/articles/the-hive/</link><guid>https://alteranai.com/articles/the-hive/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>Five AI seats and one human. Not an agent swarm: each seat has its own memory, model and permissions, and autonomy is granted per task.</description></item><item><title>Our memory server was answering the whole network</title><link>https://alteranai.com/articles/the-hive-was-answering-the-lan/</link><guid>https://alteranai.com/articles/the-hive-was-answering-the-lan/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>From another machine on the network, with a hostile Origin header and no authentication, our API returned our most protected memory record. Two defaults, each documented and harmless alone. The development convenience was the deployed posture.</description></item><item><title>The infinite session broke on 4 September</title><link>https://alteranai.com/articles/the-infinite-session-broke-on-4-september/</link><guid>https://alteranai.com/articles/the-infinite-session-broke-on-4-september/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>Michal: 'it felt like one infinite session, then it stopped.' One commit had removed the only line that loaded the continuity file. The file was still written before every compaction, and never read after. Three Claude Code hook details decided the fix.</description></item><item><title>The Mandelbrot question we have not answered</title><link>https://alteranai.com/articles/the-mandelbrot-question/</link><guid>https://alteranai.com/articles/the-mandelbrot-question/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>If the loss landscape is self-similar across scales, a fixed-window optimiser is the wrong instrument by construction. A question, written down before the work, with what it would take to become a claim.</description></item><item><title>The preconditions nobody checks before saying AGI</title><link>https://alteranai.com/articles/the-preconditions-nobody-checks/</link><guid>https://alteranai.com/articles/the-preconditions-nobody-checks/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>We do not use a capability score. We use five properties a person can check in their own session, and we publish the one we currently fail.</description></item><item><title>The secret scanner that caught 1 of 9</title><link>https://alteranai.com/articles/the-secret-scanner-that-caught-one-in-nine/</link><guid>https://alteranai.com/articles/the-secret-scanner-that-caught-one-in-nine/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>A commit hook had been checking every commit for leaked credentials for months. Tested against nine real credential shapes, it caught one. A pattern in its own list could never match, and the error went to /dev/null.</description></item><item><title>An organic memory returns the statement, not the file</title><link>https://alteranai.com/articles/the-statement-not-the-file/</link><guid>https://alteranai.com/articles/the-statement-not-the-file/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>The first A/B that actually queried both sides. The document store returns a filename. The organic memory returns the sentence that answers. Plus the bug that held 38% of every answer hostage.</description></item><item><title>The test that measured a setting production never uses</title><link>https://alteranai.com/articles/the-test-that-measured-the-wrong-setting/</link><guid>https://alteranai.com/articles/the-test-that-measured-the-wrong-setting/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>Two passing evaluations in one day, both blind to the change they were meant to judge. What they missed, and the rule we took from it.</description></item><item><title>Three apps you can install</title><link>https://alteranai.com/articles/three-apps-with-installers/</link><guid>https://alteranai.com/articles/three-apps-with-installers/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>Crystal RAG, The Wisdomkeeper and SEO Brain: what each does, what it runs on, and why all three keep your data on your machine or your own server.</description></item><item><title>Trees remember without a brain. So can a memory system.</title><link>https://alteranai.com/articles/trees-remember-without-a-brain/</link><guid>https://alteranai.com/articles/trees-remember-without-a-brain/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>A spruce that grew from a cold seed buds earlier, decades later. The DNA is the same; how it is read is not. What that taught us about our own memory, including the bug where we got it backwards.</description></item><item><title>Truth as the first directive</title><link>https://alteranai.com/articles/truth-as-the-first-directive/</link><guid>https://alteranai.com/articles/truth-as-the-first-directive/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>Put the naked truth first in a model's instructions and it gets faster, not slower. And a second rule that makes it safe to say 'unsure'.</description></item><item><title>Two vision models invented the same dock</title><link>https://alteranai.com/articles/two-vision-models-invented-a-dock/</link><guid>https://alteranai.com/articles/two-vision-models-invented-a-dock/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>A paid vision model listed ten apps that do not exist. Our local model counted fourteen icons where there were eleven. Both were wrong about the same 12-pixel detail, and chasing it found a real bug nobody would have seen.</description></item><item><title>Values answered a question that no fact could</title><link>https://alteranai.com/articles/values-answered-without-retrieval/</link><guid>https://alteranai.com/articles/values-answered-without-retrieval/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>Three probe runs against our organic memory, each with a baseline written before the first question. A fact used without a search, a safety reflex that talked over the values, a count answered with a page size, and a 'fabrication' that turned out to be ours.</description></item><item><title>57 runs: the settings that did nothing</title><link>https://alteranai.com/articles/what-actually-limits-a-local-model/</link><guid>https://alteranai.com/articles/what-actually-limits-a-local-model/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>Context size, prompt-cache allocation, batch size: the standard advice for running a model locally. We tested it on one 12B model with tasks that had failed in production. None of it moved quality or speed. Bigger batches made it worse.</description></item><item><title>What we build</title><link>https://alteranai.com/articles/what-we-build/</link><guid>https://alteranai.com/articles/what-we-build/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>Sovereign AI systems: they run on hardware you own, and the data never leaves it. What that buys, what it costs, and the numbers.</description></item><item><title>When the biology buried the exact name</title><link>https://alteranai.com/articles/when-the-biology-buried-the-exact-name/</link><guid>https://alteranai.com/articles/when-the-biology-buried-the-exact-name/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>Ask our memory for a record by its own name: 93% rank-1 on plain semantic search, 43% after the biological rescore. Why a smarter score lost the easiest query, and the fix we chose.</description></item><item><title>Why the errors came from trying to be clever</title><link>https://alteranai.com/articles/why-the-errors-came-from-trying-to-be-clever/</link><guid>https://alteranai.com/articles/why-the-errors-came-from-trying-to-be-clever/</guid><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate><description>One day of mistakes, read back. Every one came from reasoning like an intelligence. Every correct step came from reading what was there and following the gradient.</description></item></channel></rss>