A team, not a tab
Bring complementary specialties together so work can move from idea to finished product without constant manual handoffs.
01 / PROJECT OVERVIEW • SYSTEMS LABONE VISION.
ONE VISION.
MANY MINDS.
REAL WORK.
We're building a self-hosted, multi-agent digital production team. Specialized systems coordinate research, engineering, security, design, and testing—with real workflows, explicit approvals, and verified results.
EXPLORE BUILD PROGRESSREGRESSION TESTS RUN↗—LATEST SUPPLIED EVIDENCE
REVIEW CHECKS◇—INDEPENDENT VALIDATION
SKIPPED TESTS≠—TRANSPARENT RESULTS
TRACKED WORKSTREAMS▤—STATUS-DRIVEN, NOT GUESSWORK
Build progress PUBLIC TRACKER
Read-only, evidence-driven reportingSystem readiness
LIVE FETCH / VERIFIED CONTENTCURRENT RELEASE POSITION
CHECKINGLoading project evidence…
This page reports the age and verification level of its source.
Reading public milestones…
Build dispatches
PUBLIC JOURNALLoading verified dispatches…
Why we're building it
02 / The missionBeyond the chatbot / beyond the demo
Most AI tools can answer a question. Our experiment asks a harder one: can the same technology reliably manage work, coordinate roles, preserve context and produce results people can independently verify?
Context that lasts
Maintain durable Kanban states, decisions and evidence across restarts, disconnected sessions and model changes.
Own the machinery
Make local models and self-hosted infrastructure meaningful contributors to complex technical work.
Evidence over hype
Governed actions, security gates, scoped approval, regression tests and honest blocked states.
Designed for accountable execution
03 / System architectureFrom request to evidencePROPOSED OPERATING FLOW
01ResearchDISCOVER
02PlanREADINESS
03ApproveHUMAN GATE
04ExecuteDELEGATE
05VerifyQA & RELEASE
The intended lifecycle combines native Hermes Kanban with a separate authorization controller. Public telemetry does not confer access to private tasks or approvals.
Safety isn't a checkbox
A system that knows when to stop.
Good automation needs boundaries. We test what happens when models refuse, providers fail, cards change, or a task needs fresh consent.
01 Human-readable approval requests.
02 Scope-bound, explicit authorization.
03 Durable native Kanban records.
04 Audit evidence, no fabricated green lights.
The stack underneath
04 / Actual technologiesSYS 01
Agent intelligence
Hermes Agent + native Kanban
Agent organizations, delegated tasks, lifecycle states and integrated work coordination.
OpenAI Codex CLI
Code generation, technical review, staged fixes and automated regression verification.
Ollama + Qwen
Locally served conversation and coding profiles with configured 65,536-token context windows.
SYS 02
Runtime & infrastructure
Windows + WSL2 Ubuntu
A Linux operations environment backed by the local Windows workstation.
Docker + Python + systemd
Containerized services, runtime controllers, testing, monitoring and scheduled status exports.
NVIDIA RTX + local GPU inference
Running and optimizing local model workloads on consumer graphics hardware.
SYS 03
Control & delivery
Discord + approval workflows
Owner-facing updates, actionable review requests and feedback in plain language.
GitHub + Cloudflare Pages
Versioned code, static publishing, and a restricted public status projection.
Tailscale + SSH + tmux
Remote operation, session continuity and secure terminal access.
THIS IS AN ACTIVE ENGINEERING EXPERIMENTBUILDING THE TOOLS
BACK TO TRACKER ↑BUILDING THE TOOLS
TO BUILD WHAT'S NEXT.
Follow along as working software, tests and verified integrations replace prototypes. Every milestone is reported with the evidence available—not with optimism disguised as production readiness.