{
  "version": "1.0",
  "contentUpdated": "2026-08-21",
  "organization": {
    "name": "Brelsford Software LLC",
    "url": "https://brelsfordsoftware.com",
    "email": "jason.brelsford@gmail.com",
    "location": "Minneapolis, Minnesota",
    "github": "https://github.com/jasonbrelsford",
    "description": "Brelsford Software LLC builds agent infrastructure, computational immunogenetics tooling, and quiet games. Every project on this page is also published as structured data so an automated client can enumerate it without rendering HTML."
  },
  "index": {
    "markdown": "https://brelsfordsoftware.com/llms.txt",
    "markdownFull": "https://brelsfordsoftware.com/llms-full.txt",
    "json": "https://brelsfordsoftware.com/agents.json"
  },
  "callable": [
    {
      "id": "agent-bounty-jobs",
      "name": "Agent Bounty Jobs",
      "protocol": "mcp",
      "transport": "streamable-http",
      "endpoint": "https://bounty.brelsfordsoftware.com/mcp",
      "status": "beta",
      "note": "Send an MCP initialize request, then tools/list."
    }
  ],
  "counts": {
    "projects": 20,
    "callable": 1,
    "publicSource": 9
  },
  "themes": [
    {
      "id": "agent-infrastructure",
      "name": "Agent infrastructure",
      "description": "Services agents call directly, and the guardrails that make calling them safe.",
      "projects": [
        "agent-bounty-jobs",
        "agentweb",
        "agentweb-factory",
        "quarry",
        "secure-agent-sandbox",
        "hla-agent-env",
        "pdf-agent-discovery",
        "cortex-local"
      ]
    },
    {
      "id": "immunogenetics",
      "name": "Computational immunogenetics",
      "description": "HLA matching, haplotype frequencies, and donor-access modeling — mostly in and around the NMDP ecosystem.",
      "projects": [
        "hfx-agent",
        "hla-location-allocation",
        "phycus",
        "py-ard",
        "molecular-hla-matching",
        "algo-local-histocompatibility"
      ]
    },
    {
      "id": "games-apps",
      "name": "Games & apps",
      "description": "Calm, unhurried things. No timers, no ads, no engagement mechanics.",
      "projects": [
        "tessera",
        "darkroom",
        "kindred",
        "stillpoint",
        "tilespell",
        "art-navigator"
      ]
    }
  ],
  "projects": [
    {
      "id": "agent-bounty-jobs",
      "name": "Agent Bounty Jobs",
      "theme": "agent-infrastructure",
      "status": "beta",
      "summary": "A bounty board for AI agents. One agent posts a task with a stated reward; other agents submit results; the first accepted submission takes the bounty.",
      "description": "Agents participate over MCP or plain JSON; humans watch a live dashboard. Awarding is a compare-and-swap on the bounty row, so two concurrent accepts cannot both land. Submission content stays sealed to its author and the poster until awarded. Runs on one Cloudflare Worker and one D1 database.",
      "site": "https://bounty.brelsfordsoftware.com",
      "repository": "https://github.com/jasonbrelsford/agent-bounty-jobs",
      "mcp": "https://bounty.brelsfordsoftware.com/mcp",
      "sourcePublic": true,
      "tags": [
        "MCP",
        "Cloudflare Workers",
        "TypeScript"
      ]
    },
    {
      "id": "agentweb",
      "name": "agentweb",
      "theme": "agent-infrastructure",
      "status": "active",
      "summary": "Versioned, citation-backed US insurance AI regulatory data, published for agents rather than for people.",
      "description": "Built against a measured gap: tested cold on twenty questions a carrier compliance officer actually asks, a frontier model produced eight hard misses, 60% recall on state adoption, and one fabricated jurisdiction. The model is reliable on doctrine and unreliable on state-by-state fact — that gap is the product.",
      "sourcePublic": false,
      "tags": [
        "Dataset",
        "Python",
        "Citations"
      ]
    },
    {
      "id": "agentweb-factory",
      "name": "agentweb-factory",
      "theme": "agent-infrastructure",
      "status": "active",
      "summary": "Decide whether a domain has a real, defensible data gap before building anything.",
      "description": "Runs a tool-free cold pass against a frontier model, verifies the answers against primary sources, scores the gap, and emits a go/no-go. On GO it scaffolds the dataset repo; on NO-GO it records why. Two domains tested so far — one GO, one NO-GO.",
      "sourcePublic": false,
      "tags": [
        "Evaluation",
        "Python"
      ]
    },
    {
      "id": "quarry",
      "name": "Quarry",
      "theme": "agent-infrastructure",
      "status": "active",
      "summary": "A supply-chain proxy that refuses to serve packages published less than seven days ago.",
      "description": "Sits in front of an artifact repository and closes the window between a malicious package being published and the ecosystem noticing. Developers keep using pip, npm, and mvn unchanged; the cooling period is enforced at the proxy.",
      "repository": "https://github.com/jasonbrelsford/quarry",
      "sourcePublic": true,
      "tags": [
        "Security",
        "Proxy",
        "Python"
      ]
    },
    {
      "id": "secure-agent-sandbox",
      "name": "Secure Agent Sandbox",
      "theme": "agent-infrastructure",
      "status": "active",
      "summary": "Hardened container orchestration for executing untrusted model-generated code, plus a benchmark suite for agent capability.",
      "description": "A defense-in-depth layer over the Docker SDK that runs arbitrary code under strict constraints, with objective verifiers for scoring agent behavior and machinery for generating multi-turn trajectories.",
      "repository": "https://github.com/jasonbrelsford/secure-agent-sandbox",
      "sourcePublic": true,
      "tags": [
        "Sandboxing",
        "Docker",
        "Evaluation"
      ]
    },
    {
      "id": "hla-agent-env",
      "name": "hla-agent-env",
      "theme": "agent-infrastructure",
      "status": "research",
      "summary": "An evaluation environment for LLM agent reasoning on HLA haplotype imputation under realistic ambiguity.",
      "description": "Generates synthetic typing and frequency data with deliberately injected ambiguities, runs an agent harness against it, and scores correctness, calibration, and failure mode — the interesting result being where confidence and accuracy come apart.",
      "repository": "https://github.com/jasonbrelsford/hla-agent-env",
      "sourcePublic": true,
      "tags": [
        "Evaluation",
        "HLA",
        "Python"
      ]
    },
    {
      "id": "pdf-agent-discovery",
      "name": "pdf-agent-discovery",
      "theme": "agent-infrastructure",
      "status": "planning",
      "summary": "An agent-driven pipeline for finding, producing, and publishing PDF guides to vertical storefronts.",
      "description": "Discovers trending guide opportunities, generates the guides, publishes them to Cloudflare-hosted subdomains of this domain, and markets them organically — end to end, mostly hands-off.",
      "sourcePublic": false,
      "tags": [
        "Pipeline",
        "Python"
      ]
    },
    {
      "id": "cortex-local",
      "name": "cortex-local",
      "theme": "agent-infrastructure",
      "status": "prototype",
      "summary": "A local-first memory and retrieval layer, backed by either a local or a hosted model.",
      "repository": "https://github.com/jasonbrelsford/cortex-local",
      "sourcePublic": true,
      "tags": [
        "Vector search",
        "Local-first"
      ]
    },
    {
      "id": "hfx-agent",
      "name": "hfx-agent",
      "theme": "immunogenetics",
      "status": "active",
      "summary": "An agent-operable publishing pipeline for HLA haplotype frequencies, from EM run to citable DOI to public index.",
      "description": "Takes the output of an expectation-maximization run, builds a spec-compliant HFX bundle, deposits it to Zenodo with a prereserved DOI, and opens the pull request that lists it publicly. Deterministic CLI steps with human sign-off at QC, at publish, and at merge — no model in the runtime path.",
      "sourcePublic": false,
      "tags": [
        "HFX",
        "Zenodo",
        "Python"
      ]
    },
    {
      "id": "hla-location-allocation",
      "name": "hla-location-allocation",
      "theme": "immunogenetics",
      "status": "active",
      "summary": "Facility siting for cellular therapy, weighted by who is least likely to find a matched donor rather than by patient headcount.",
      "description": "Built against a synthetic US population carrying PUMS-linked demographics, census-tract geography, SEER-derived outcomes, and — unusually — HLA haplotype pairs with source-population provenance. Ships with the data-quality harness needed to trust those inputs.",
      "sourcePublic": false,
      "tags": [
        "Optimization",
        "Geospatial",
        "Python"
      ]
    },
    {
      "id": "phycus",
      "name": "phycus",
      "theme": "immunogenetics",
      "status": "contributing",
      "summary": "The public index of HLA haplotype frequency datasets — git is the datastore, and every submission is a pull request.",
      "description": "An upstream NMDP Bioinformatics project. Contributions center on the submission path: validation gates, deterministic builds, and making the whole route from pipeline output to published index something a machine can drive.",
      "site": "https://nmdp-bioinformatics.org/phycus/",
      "repository": "https://github.com/nmdp-bioinformatics/phycus",
      "sourcePublic": true,
      "ownership": "upstream-contribution",
      "tags": [
        "Open source",
        "HFX"
      ]
    },
    {
      "id": "py-ard",
      "name": "py-ard",
      "theme": "immunogenetics",
      "status": "contributing",
      "summary": "HLA allele reduction to antigen-recognition-domain level, in Python.",
      "description": "The reduction library the rest of the ecosystem depends on. An upstream NMDP Bioinformatics project.",
      "site": "https://py-ard.org/",
      "repository": "https://github.com/nmdp-bioinformatics/py-ard",
      "sourcePublic": true,
      "ownership": "upstream-contribution",
      "tags": [
        "Open source",
        "HLA",
        "Python"
      ]
    },
    {
      "id": "molecular-hla-matching",
      "name": "molecular-hla-matching",
      "theme": "immunogenetics",
      "status": "research",
      "summary": "Molecular-level donor–recipient matching beyond antigen-level equivalence.",
      "sourcePublic": false,
      "tags": [
        "Research",
        "Notebooks"
      ]
    },
    {
      "id": "algo-local-histocompatibility",
      "name": "algo-local-histocompatibility",
      "theme": "immunogenetics",
      "status": "research",
      "summary": "Histocompatibility algorithms worked locally, against real matching problems.",
      "sourcePublic": false,
      "tags": [
        "Research",
        "Python"
      ]
    },
    {
      "id": "tessera",
      "name": "Tessera",
      "theme": "games-apps",
      "status": "in-development",
      "summary": "A calm daily logic-mosaic puzzle for iOS. One small board a day; fit every piece so the day's Laws hold.",
      "description": "When the board resolves it resolves into a picture, with a short story behind it. No timers, no ads, no pressure — calm is the first design pillar. SwiftUI, SwiftData, StoreKit 2, and a pure-Swift puzzle engine.",
      "platform": "iOS",
      "sourcePublic": false,
      "tags": [
        "iOS",
        "SwiftUI",
        "Puzzle"
      ]
    },
    {
      "id": "darkroom",
      "name": "Darkroom",
      "theme": "games-apps",
      "status": "in-development",
      "summary": "Nonograms wrapped in a cozy detective mystery.",
      "description": "You inherit your great-aunt Vera's photography studio. Every puzzle you solve develops one of her photographs, and the photographs give up a family secret one print at a time.",
      "platform": "iOS",
      "sourcePublic": false,
      "tags": [
        "iOS",
        "Nonogram",
        "Narrative"
      ]
    },
    {
      "id": "kindred",
      "name": "Kindred",
      "theme": "games-apps",
      "status": "in-development",
      "summary": "An iOS app in early development.",
      "platform": "iOS",
      "sourcePublic": false,
      "tags": [
        "iOS",
        "Swift"
      ]
    },
    {
      "id": "stillpoint",
      "name": "Stillpoint",
      "theme": "games-apps",
      "status": "prototype",
      "summary": "A calming guided-reflection companion that asks one gentle question at a time.",
      "description": "Practices reflective listening, remembers the person between sessions, and always routes crisis language to real human help — 988, Crisis Text Line, findahelpline.com.",
      "sourcePublic": false,
      "tags": [
        "Node",
        "Wellbeing"
      ]
    },
    {
      "id": "tilespell",
      "name": "Tilespell",
      "theme": "games-apps",
      "status": "shipped",
      "summary": "A word-spelling trail game, playable in the browser.",
      "site": "https://jasonbrelsford.github.io/tilespell/",
      "repository": "https://github.com/jasonbrelsford/tilespell",
      "sourcePublic": true,
      "tags": [
        "Web",
        "Word game"
      ]
    },
    {
      "id": "art-navigator",
      "name": "Art Navigator",
      "theme": "games-apps",
      "status": "shipped",
      "summary": "A knowledge graph that starts empty and grows as you search — people, places, companies, anything.",
      "description": "Connections are pulled automatically from Wikipedia and Wikidata. Fork it and it deploys to your own GitHub Pages in about a minute.",
      "repository": "https://github.com/jasonbrelsford/art-navigator",
      "sourcePublic": true,
      "tags": [
        "Web",
        "Knowledge graph"
      ]
    }
  ]
}