ModelEval
Operational benchmark and public evidence system. The live release has 77 cards and 318 runs, load-excluded median test time, both required paired benchmark modes, and the gold thinking-on badge.
Status rule: Active projects have current accepted work and activity within 24 hours. Idle projects include paused work, waiting work, and running automation with no current build lane. Ledger pages keep their official ledger snapshot for reference.
Active
9 projectsProjects with current human-directed build work.
Agent Evaluation System
Nightly cross-model eval foundry with author/verifier separation, policy and schema gates, mutation checks, local examinee runs, and gated promotion. The July 26 cycle ran; its output quality, model labels, and dirty generated state need review.
Dig
Local-first, evidence-aware knowledge tutor with exact provenance, source erasure, provider controls, lifecycle tooling, packaging, updates, and diagnostics. The installed app is healthy and its profile scan is complete; the browser UI currently uses a labeled demo explainer until an approved real model is connected.
Dig Browser
Free explicit-gesture browser capture surface for Dig with review-before-send, native messaging, durable local capture, source erasure, and current browser acceptance. Local gates pass; legal, Store, update, signing, and beta authority remain open.
Sales Workbench
Deterministic offline host for enterprise customer sales tools with manifest discovery, isolated state, backup, and broker contracts. Industry Leadership and Field Atlas are merged; 296/296 current Node checks and 347/347 staged transfer checks pass while managed acceptance and the dated ZIP remain open.
Field Atlas App
Offline territory, nearby-account, and route-planning app now merged into Sales Workbench. Its local product, browser, and package gates pass; managed Windows, second-host, national-fixture, topology, and salesperson acceptance remain open.
Industrial Knowledge App
Deterministic Industry Intelligence compiler for 23 cited industry packs plus a separate approved-leadership publisher. Three sanitized releases are integrated; the formal real-content, steward, managed-machine, and enterprise acceptance gate remains open.
Retail Sample Website
Private sample plant storefront with catalog, exact and batch sale types, cart, test-only pickup checkout, and business pages. The concept is live and tested; reviewer approval and all real owner, account, inventory, payment, and launch gates remain open.
Operations Manager Pilot
File-first Hermes operations package for a pilot sponsor and industrial client. The operations manager package and its 17-check tool suite exist; the revised pre-pitch plan still requires transport, Windows, egress, agreement, disclosure, and customer compliance work.
Idle
28 projectsPaused, operational, partial, stale, migrated, or waiting-for-next-decision work.
Agent Operations Console
Live operations console at /cockpit for company fleet observability, with the internet-facing surface kept read-only. The v2 lane adds LAN-only model control and inference statistics behind propose/confirm, dry-run, allowlist, and hash-chained audit gates; control remains off the public route.
Work Control Plane
Governed task-dispatch and control-plane package for company agents. The supervised local-pinned executor=hermes track has passed 3/3 packages and residual HMAC/scanner gates are closed; broad unattended Hermes execution remains a separate owner governance decision.
Agent Architecture
Define a centralized, wall-guarded, garrison-driven AI agent architecture.
ColdOutreachWriter
Hook-anchored cold-outreach system with SSR scoring, anti-AI-tell penalties, enterprise customer/data-center domain pack, optimizer, CLI, and salesperson web app. It is inactive after the 24-hour ledger check; the next proof point is real send/reply calibration.
ConcurrentWork
Coordinate multi-machine concurrent work across local coordinator, a worker agent, a worker agent, and a worker agent, where everyone (local coordinator included, as worker-01) shares the burden.
Municipal Business Directory
A meeting-ready municipal operations platform business-directory demo that makes the pilot municipality council see a useful, attractive, repeatable, free baseline web profile system for every identified pilot municipality business.
WebsiteBuilder
A reusable company website-generation product that can rapidly produce polished, conversion-focused websites for local and regional businesses, beginning with five high-quality single-file HTML + CSS framework homepage templates and an interactive style-switching demo.
WiHART Gateway Local Display
A standalone field-station hardware/single-board computer local-display field station for enterprise customer 1410S WirelessHART gateways: plug into a gateway, auto-discover joined devices by HART tag via read-only HART-IP, show PV/SV/TV/QV on the 800×480 LCD with auto-scroll, log SQLite + per-tag CSV, export CSVs through USB or local download, survive power cycles/off-grid use, and be cloneable across the remaining field-station hardware boxes without per-customer manual Modbus mapping.
Frontier Token Capture Machine
A harvest-first measurement-and-capture machine that turns existing and future frontier sessions plus allowlisted local corpus files into reusable artifact candidates, eval seeds, schemas, workflows, decisions, failure reports, and local-first improvement assets for Hermes/legacy platform and company.
Intelligence Agent Prototype
Build analyst, an AI-powered intelligence analyst for enterprise customer's data-center team.
Enterprise 2nd Brain
Decision-infrastructure prototype for decision-capture prototype: durable DecisionRecord schema, SQLite ledger, context assembler, MCP-shaped server, drafter/review loop, session ingestion, and validation harness. The open question is whether real decision owners feel the leverage.
MEP Office Contact Intelligence
Data-cleaning and contact-intelligence lane for MEP and integration-specialist offices. It has normalizers, validators, research packets, templates, and a professional directory assisted-browsing protocol; it needs a real office-list pilot.
Monthly Earner
Generate $300+/month recurring revenue through discrete operational tools for small businesses.
Municipal Operations Intelligence
Municipal-ops product starter kit focused on small-town accountability: council dashboards, WWTP/pilot municipality source framing, discovery packet, data model, risk register, and service tiers. Next step is field validation with real records.
Agent Governance Archive
Provide agent architecture's quad governance brain, resolver rules, skills, memory notes, and historical operating doctrine.
Agent-Led Marketing
Make Hermes useful for small-business marketing, social account operations, and customer outreach across the United States.
Hermes Model Council
Model-council harness for testing whether complementary model panels can beat solo baselines on defined task classes. It preserves disagreement, frozen rubrics, evidence trails, cost, latency, failures, and promotion gates; mock and local experiments currently prove harness flow, not frontier-level lift.
Enterprise Intelligence Pitch Materials
Package an informal executive brief for an enterprise customer data-center intelligence platform conversation.
security-review
Review and harden the Mac Studio local coordinator host and its three-machine fleet.
storytime
Provide private family story generation at storytime.company.
Artifact-Aware Router and Fine-Tune Pipeline
Designs an explicit retrieval, ranking, and controller policy over structured artifact and trace cards. The planning package is idle while it is revised away from folder-first routing; a local fine-tune remains a later evidence-gated option.
Hermes MoA Local-Frontier Capability Lab
Links ModelEval evidence and Hermes Model Council recipes to test when local and frontier model mixtures improve real company work. The next proof is a small held-out coding and refactoring sweep, not a broad routing or training claim.
Self-Improvement Outreach Corpus
Builds a permitted, recipient-first corpus for more natural outreach. A 30-work, roughly 1.7-million-word open corpus exists; trace, concept-note, preference-pair, leakage, provenance, and human-voice evaluation contracts remain to be designed.
Hermes Status Hub
Plans a cockpit-gated page and progress feed for every ledger project. Inventory, page templates, locked decisions, dossier generation, and initial stubs exist; the gated host and redaction-safe stack view remain pending.
Critical Pursuits App
Deterministic local pursuit-planning system with immutable reports, SQLite state, publishing, and an accountable action loop. The clean standalone baseline is retained; current integration work belongs to the Sales Workbench lane.
Opportunity Miner
Private federal-opportunity mining and decision workflow for consulting client. The search and collaboration surface exist, but responses still use browser-local storage and key business decisions must be signed before the setup lane proceeds.
Social Agent Intelligence Feed
Daily read-only AI-agent intelligence report built through Hermes, with 15 enabled social network accounts, priority and focus controls, state tracking, raw capture, and HTML, JSON, and Markdown outputs. Tests pass 8/8; the first three live reports need review.
Agent Efficiency Review
Evidence-based review lane for moving deterministic agent work into measured scripts and repeatable judgment into validated skills. Bounded search is accepted; the patch-control candidate remains idle until one realistic validation proves value.
Closed
11 projectsComplete, finished, or retained historical project records.
Operations Cockpit MVP
Static daily operating dashboard that pulls company project, agent, cron, knowledge, eval, revenue, and system-health signals into one local HTML surface. It is complete historical infrastructure, now succeeded by live-console and control-plane ideas.
Operations Cockpit Audit
Package the Cockpit MVP for review and audit.
Evaluation Gauntlet Expansion
Adds company-real evaluation families around governance, tool discipline, wiki boundaries, small-business, municipal reasoning, and tool-output comprehension. The package is staged with seed cases and rubrics; production scoring and live-model runs are still pending.
Knowledge Atlas
Turns the company wiki into a navigable, evidence-aware static knowledge site with curated article tiers, source trails, and graph views. The MVP is complete; future work is polish and regeneration as the underlying wiki changes.
Frontier-Hardened Local Stack
Planning and audit package for making the local Hermes/inference stack more reliable: compression rollout, route schema, synthetic router data, eval integration, and local/frontier operating policy. No runtime profiles were changed.
Graphify Local Knowledge-Map Engine
Deterministic parser that converts wiki and Atlas markdown into nodes, edges, and reviewable graph data. Level 1 is complete and feeds Atlas; typed entities and workplace-corpus generalization remain successor scope.
Open-Source Release Discipline
Executable release gate for company-built tools: intake, checklist, scanner, validator, templates, code host procedure, security/privacy review, and announcement materials. It is ready to run on the next real release candidate.
Small Business Playbook
Concrete HVAC-first automation offer with discovery questions, demo data, launch-assistant workflows, one-pager, and reusable trades checklist. It is ready for prospect validation rather than more internal packaging.
Stateful Project Tracking Discipline
Governance toolkit behind the official ledger: parser, validator, status rules, review runbook, cross-harness activity audit, SQLite ledger, and weekly refresh pattern. Recurring work is keeping rows aligned with live reality.
Storytime Polish and Hardening
Hardening package for the existing private Storytime app: parent controls, saved-story library, continuity rules, print/export, quality/safety rubric, and eval loop. It deliberately avoided mutating production.
Wiki Library Governance and Knowledge Base Growth
Governance kit for keeping the wiki/library/Atlas ecosystem navigable and evidence-disciplined. It defines boundary rules, triage surfaces, Atlas-ready frontmatter, source trails, and follow-up maintenance cadence.