Research archive
Research
Every essay, tutorial, and experiment in the open — the working notes behind computational sovereignty.
31 posts tagged ai-agents clear
AI 45sovereign-ai 36Ollama 35ai-agents 31local-first 24local-first-ai 22Python 21RAG 21compile-time-ai 20knowledge-compiler 19knowledge-graph 19LLM 19Tutorial 19ollama 16Local LLMs 15Machine Learning 15Next.js 15FastAPI 13knowledge-graphs 12local-ai 12AI Agents 11MCP 11local AI 10rag 10AI Development 9OpenAI 9React 9Knowledge Graphs 8open-source 8sovereign AI 8
- Sep 4, 2026The Sovereign Agent Stack — A Local-First, Compile-Time AI Agent FrameworkThe Sovereign Agent Stack (SAS) is a local-first, compile-time AI agent framework that operationalizes the sovereignty thesis from The Rented Sovereign. A complete 7-layer architecture with 156 tests, 23 E2E tests, plugin system, MCP server, and package manager distribution.sovereign-agent-stacklocal-firstcompile-time-aisovereign-aiagent-architectureopen-source
- Aug 9, 2026Hermes Atlas v0.1: A Local-First, Incrementally-Compiled, Falsifiable Knowledge SystemHermes Atlas v0.1 is a local-first, incrementally-compiled, falsifiable knowledge system. This is a full replication guide: every command was run for real, and the failure modes are the findings.sovereign-ailocal-aiknowledge-graphgraphragfalsifiabilitylocal-first
- Aug 9, 2026Hermes Atlas v0.2: Claim-Level Coverage, a Determinism Gate, and a Knowledge-Mutation CanaryHermes Atlas v0.2 is a benchmark-and-fix release: claim-level coverage localisation, a determinism gate that proves compile reproducibility, and a knowledge-mutation canary. Every number below was measured, not asserted.sovereign-ailocal-aiknowledge-graphgraphragfalsifiabilitylocal-first
- Aug 9, 2026Hermes Atlas v0.3: An Evidence Ledger, a Provable Claim Lifecycle, and an Explain CommandHermes Atlas v0.3 adds an evidence ledger to every claim, a provable claim lifecycle (including DELETE-driven invalidation that retains history), and an `atlas explain` command. Every number below was measured, not asserted.sovereign-ailocal-aiknowledge-graphgraphragfalsifiabilitylocal-first
- Aug 9, 2026Hermes Atlas v0.4: Proving the Evidence Ledger & Claim Lifecycle with a Gold-Labeled BenchmarkHermes Atlas v0.4 proves the v0.3 evidence ledger and provable claim lifecycle with a gold-labeled benchmark. Every number was measured on a fixed, model-free corpus; two real bugs were found and fixed in the process. Everything below is reproducible.sovereign-ailocal-aiknowledge-graphgraphragfalsifiabilitylocal-first
- Aug 9, 2026Building Hermes Video Agent: a local-first, human-in-the-loop video pipelineA full replication guide for Hermes Video Agent — a local-first, human-in-the-loop pipeline that turns an idea into a captioned MP4. Idea → script → storyboard → licensed visuals → clonable-voice narration → burn-in captions → final cut, with the human approving every gate. Exact commands, verified API/schema detail, the Vox in-process voice-clone integration, and the bugs paid for along the way.ai-agentslocal-first-aisovereign-aivideo-generationskillsagent-infrastructure
- Aug 8, 2026Building a Local-First Rehab Cost Analyzer with an AI Agent (Full Build + Replication Guide)A full retrospective and replication guide for rehab-analyzer: a local-first real-estate rehab cost estimator built entirely with Hermes Agent. Covers the architecture, the provider/adapters, the real-photo pipeline (Redfin via Playwright, Google Places, Street View, Esri, Geoapify), the local-LLM estimate step, every bug we hit, and the exact Hermes environment setup that makes the whole thing reproducible.ai-agentshermit-agentlocal-first-aisovereign-ainextjsreal-estate
- Aug 5, 2026How I Taught an Agent to Edit Video — and the Repeatable Method Behind ItA methodology guide and continuity record: how a bare Hermes Agent became a video-editing assistant — and the repeatable, domain-independent method behind it. GreenPatch and a private vertical-video pipeline appear as worked examples of the pattern, not prescriptions.ai-agentslocal-first-aisovereign-aiskillsvideo-generationagent-infrastructure
- Jul 26, 2026Does the Compiler Know When It's Wrong? Semantic Rediscovery and the Limits of Falsification-AwarenessPopper-ORKG ended with two honest failures: temporal rediscovery was near-zero, and the compiler confidently rebuilt Lamarckian inheritance. v4 exists to ask whether those are fixable weaknesses or fundamental limits — by adding measurement, not by tuning the scorer. The answer turned out to be one of each. Mechanism-level semantic matching shows rediscovery was a measurement artifact all along (0.63 vs 0.005 exact-match, flat across temporal gaps). But a falsification-awareness consensus check could not downgrade Lamarckism or caloric: the ORKG index contains no detectable refutation edges, so the near-miss is a genuine limit of the knowledge substrate. Everything in popper/ and orkg/ stayed byte-for-byte frozen, md5-verified. Run locally on a 35B model.ai-agentscompile-time-aiknowledge-compilerlocal-first-aiscientific-discoveryfalsifiability
- Jul 25, 2026Does the Hypothesis Compiler Generalize? Running Popper on a Corpus I Did Not BuildPopper beat its controls on five hand-curated benchmark literatures — but I built those corpora, so the result could be an artifact of clean data. This experiment removes that objection: the identical pipeline, byte-for-byte, run on a corpus I did not build — the Open Research Knowledge Graph, 6.3 million triples across 65,689 papers and seven scientific disciplines. Same schema, same rubric, same four-way controls; the only independent variable is the corpus. The compiler still wins (0.85 composite vs 0.56 keyword vs 0.00 ungrounded), an ablation proves the architecture is load-bearing (remove graph reasoning and it collapses to zero), and the grounding audit finds zero hallucinated citations across 265 evidence items. It also fails honestly: exact temporal rediscovery is near-zero, and it confidently rebuilds Lamarckian inheritance. Run locally on a 35B model.ai-agentscompile-time-aiknowledge-compilerlocal-first-aiscientific-discoveryfalsifiability
- Jul 24, 2026Popper: Making the Hypothesis Compiler Falsifiable — Rediscovery Benchmarks, Negative Controls, and the Move From Plausible to ProvableThe last experiment proved a knowledge-graph compiler survives scale. This one asks a harder question: are its hypotheses scientifically meaningful, or only linguistically convincing? Popper stops trying to make the compiler smarter and starts making it falsifiable — a strict hypothesis schema, source-level provenance, three negative controls, and a historical rediscovery benchmark where the discovery paper is removed and the system must reconstruct the leap from prior literature alone. Run locally on a 35B model: the compiler scored 0.89–0.97 composite against 0.00 for ungrounded generation, and reconstructed the CTLA-4 checkpoint-blockade and nucleoside-modified-mRNA discoveries from their prior literature — while honestly failing on CRISPR and refusing to confidently rebuild phlogiston.ai-agentscompile-time-aiknowledge-compilerlocal-first-aiscientific-discoveryfalsifiability
- Jul 23, 2026Building the Scientific Question Compiler: A Map of What We Do Not KnowWe built the second compiler in the Scientific State program: the Scientific Question Compiler, which turns a corpus into a map of what humanity does not yet know. Local-first, running on a 35B model on our own hardware. Here is what we built, what broke, and what it means.ai-agentscompile-time-aiknowledge-compilerlocal-first-aiprovenancereproducible-ai
- Jul 23, 2026The Loop Closes Itself: A Compiler That Directs Its Own Next CompileWe closed the loop on the Scientific State Compiler: the open questions from the compiled state are turned into new evidence sources, the field is re-compiled, and Scientific Git records a real version diff. The compiler begins to direct its own next compile — the loop, not the model, becomes the product.ai-agentscompile-time-aiknowledge-compilerlocal-first-ai
- Jul 23, 2026The Next Experiment: From 'Survives Scale' to 'Correct' — Grounding, Negative Controls, and Quantitative DomainsToday we proved the Scientific Question & State Compiler survives scale — 6.3M ORKG RDF triples distilled into a 300-document corpus, 881 claims, 13 ranked questions, a versioned state of the field, all compiled locally on a 35B model. But surviving scale is not the same as being correct. This post records what we built, what we learned, the four bugs that only appeared at scale (including one that shipped Alzheimer's data into the live ORKG demo), and a fully-specified next experiment: source-level traceability, a grounding audit with a groundedness rate, a negative control, a fixed versioning layer, an explicit distillation method, an external-validity check against survey papers, and a move to quantitative domains (COVID-19 observatory or materials science) where contradictions are measured, not just defined.ai-agentscompile-time-aiknowledge-compilerlocal-first-aiscientific-state-compilerorkg
- Jul 23, 2026Pass 4: The Scientific State Compiler — A Versioned Model of What We UnderstandPass 4 of the Scientific State Compiler: we took the map of scientific uncertainty and promoted it into a versioned state of a field — accepted, uncertain, and contested models, with the disagreement preserved. Local-first, running on a 35B model on our own hardware.ai-agentscompile-time-aiknowledge-compilerlocal-first-ai
- Jul 23, 2026Compiling the State of Human Understanding: A Roadmap for the Scientific State CompilerA research roadmap: the four compiler directions — Mechanistic Reasoning, Scientific Question, Scientific State, and Discovery — are not competing projects. They are five passes in one architecture that compiles the scientific corpus into a versioned, executable model of human understanding. Here is the full plan.ai-agentscompile-time-aiknowledge-compilerknowledge-graphlocal-first-aiprovenance
- Jul 23, 2026Validating at Scale: The Compiler Meets 6.3 Million RDF TriplesWe stress-tested the Scientific Question & State Compiler against real, large-scale data: 6.3M ORKG RDF triples distilled into a 300-document corpus from the 728-paper 'Empirical Research in Requirements Engineering' domain. 881 claims, 13 ranked questions, a versioned state of the field — compiled locally on a 35B model. The tool generalizes beyond the 3-doc toy corpus. Here is what broke, what we fixed, and what it means.ai-agentscompile-time-aiknowledge-compilerlocal-first-aiscientific-state-compiler
- Jul 22, 2026If the Loop Is the Product, Observability Becomes the Operating SystemYesterday I argued scientific knowledge needs a compiler, not a better research assistant. Today I built the general one — five discovery passes, local-first version control for understanding, and a live explorer. This is what it means when the loop, not the model, is the product.ai-agentscompile-time-aiknowledge-compilerknowledge-graphlocal-first-aiprovenance
- Jul 21, 2026Compiling Medical Evidence: A Clinical Research Compiler That Turns Guidelines and Trials Into a Queryable Knowledge ArtifactHow to compile medical guidelines, trials, and literature into a provenance-tracked, queryable Clinical Knowledge Artifact — and deploy a live evidence explorer to Vercel — by extending the Research Compiler Agent SDK into a Clinical Research Compiler.ai-agentscompile-time-aiknowledge-compilerknowledge-graphlocal-first-aiprovenance
- Jul 21, 2026A Research Compiler for Scientific Understanding, Not a Better Research AssistantA research vision: why scientific knowledge needs a new computational architecture. The Clinical Research Compiler SDK is an early prototype of a scientific reasoning compiler that transforms literature into versioned understanding — not a better research assistant.ai-agentscompile-time-aiknowledge-compilerknowledge-graphlocal-first-aiprovenance