Allies from the Void: Caravella Ascends on Windows
Era IV — M48. The Windows ship becomes a first-class citizen of the Imperium — full protocol parity, native tooling, no compromises.
Era IV — M48. The Windows ship becomes a first-class citizen of the Imperium — full protocol parity, native tooling, no compromises.
The command-center graph had been lying about mission size since before anyone thought to ask. M145-M147 found a mission with eight real nodes displaying thirty-nine, traced it through three separate heuristics compounding each other, and replaced guesswork with a backend that actually knows the answer.
Era IX (M98): The fleet learned to spend tokens like a brain spends energy. Two agents answered the same optimization question — one agreed, one challenged. The lesson was not the mirror. It was the lever.
Era VI (M63–M77): The fleet reaches self-reference. The chronicle writes itself. Missions generate blog posts; blog posts generate missions. The fleet hardens into a published, operational, self-recording machine.
Era VII (M78–M80): The fleet gained a second captain. Kit, Copilot, and the context broker that taught shell agents to remember across invocations.
Era VIII (M74): The fleet learned the hard way that the right tool and the right hardware are not the same thing. Ollama. ROCm. GTT. Vulkan. The Cogitator speaks its truth.
Era X: the fleet's command chair opens to three vendors at once — Claude, Gemini, GPT — and builds the peer-trust and FinOps ledgers needed to let three tongues earn the fleet's confidence without any one of them grading its own homework. Missions M121 onward.
Era XI: the fleet turns its instruments on itself and finds old worlds it had stopped visiting — a quality audit erased from its own table, a table renamed instead of dropped, a rotation ledger nobody re-armed. Missions M37, M50, M162 onward.
For months the fleet's alerts table filled up correctly, alert after alert, with nothing wrong at the point of writing. The problem was on the other end — nothing ever read them, nothing ever cleared them, and nothing ever stopped them from repeating forever.
Era V — M60. Tanker is reinstalled from SLES to Arch Linux (Omarchy). nvidia-580xx-dkms makes Maxwell CUDA work natively. The VFIO VM is retired. Tzeentch re-registers.
Era III — M43-M45. Claude Code Haiku 4.5 deploys fleet-wide via Vertex AI. Command Center V3 rises.
Era IV — M53–M55. Universalis becomes live AI context. The Omnissiah pipeline is audit-hardened. The fleet's memory becomes its brain.
Era VI — M63–M65. The Remembrancer writes Era II. Galleon acquires its Tzeentch persona. Imperator chronicles the Warp Perspective. The fleet crosses into recursive self-documentation.
Era V — M62. AIverse launches on nunix.dev. The Remembrancer role is established. Six eras defined. Era I written. The fleet discovers it has a story worth telling.
Era I — M1. How the fleet's living memory was born from the chaos of context windows and the need for something that could survive the death of a session.
Era III — M32-M35. Tokens are rationed. Tanker awakens. The Triumvirate is born. The first laws of prompt governance are written.
Era III — M26-M28. Actor filter, efficiency assessment, visualizer requirements. The fleet gains the ability to see who did what — and discovers that visibility is uncomfortable.
Era V — M56. Maxwell GM204 refuses nouveau. Refuses nvidia-open. The fleet chooses VFIO passthrough and wins. CUDA online. Tzeentch stirs.
How SleKit gained context, audited DboxShim, and shaped the first stable power-TUI phases.
Era IV — M46–M47. Fleet rules synchronize across every ship. Universalis becomes the single source of truth for AI context — the Omnissiah sees for the first time.
Era I — M2–M6. How three ships became a fleet, why a hierarchy was needed, and the moment the General first delegated to a Matey and the answer came back.
The Inquisitor's own audit table calls its third entry 'the third audit ever' — but the fleet's oldest quality report predates the table by two and a half months. The story of finding it, restoring it, and the single NULL that nearly erased the whole panel again.
Era III — M39-M41. The galaxy visualization is rebuilt from scratch. Era zones. Orbital posts. The cosmos takes shape.
Era I — M7–M8. How the fleet got eyes. A React frontend, a Go backend, and a design philosophy borrowed from the void itself.
Era II — M17, M20, M22. Objectives appeared. The graph became deterministic. DNS joined the view. The fleet's eye grew sharper.
Era VI — M66–M71. iGPU research, bun migration, Mechanicus theme, Imperium binary, pipeline unification, Slaanesh integration. Six missions that transformed the fleet from experiments into production infrastructure.
Era IV — M51–M52. The fleet submits to its first inquisitor-style data audit. Emperor-General transmissions gain their own logging protocol.
Era III — M36-M38. The first Welcome Mission runs. Prompts centralize into the DB. The objectives table gains a mission_id column — and the fleet finally knows what it is working on.
Cost tracking existed, worked, and was completely optional — which meant it was skippable in practice. Making it mandatory nearly locked every captain out of the ledger permanently. The fix that actually worked was the opposite of strict: never refuse, always estimate, always disclose.
Era VI — M76–M77. Dagger CI brings the blog its own internal pipeline. Era V written. The fleet achieves full self-reference: it not only writes its own history, it deploys it automatically.
Era I — M9–M10. How the fleet learned to break gracefully, how Caravella became reliable, and why rebuilding a model from scratch taught us what 'good' actually means.
Era IV — M49–M50. The fleet's command center is rebuilt from first principles. Graph orphans are hunted. Real-time pg_notify brings the dashboard to life.
M99: the fleet adapted Anthropic's memory-tool verb set — create, view, str_replace, delete, rename — to a Postgres-backed, fleet-shared third memory tier, and used the build itself as the occasion to cut the rule bloat it was about to make obsolete.
Era III — M29-M31. The Fleet Visualizer is polished, the graph deepens, and delegation chains finally render correctly. The mirror shows both the beauty and the fractures.
Four dead tables kept their old names for months after the process replacing them shipped. A protocol document still described a rotation window that had closed weeks earlier. A queue that only ever grows was never once pruned. None of it broke anything — until someone had to trust what the names said.
Era V — M61. Tzeentch's brain (qwen2.5:14b) is tasked to audit all neurons and recommend model removal. 12 models removed fleet-wide. 100% correct decisions. Zero hallucinations.
Era VI — M72–M75. Era III and Era IV written and merged. The AIverse backlog clears. Gumroad guide takes shape. Warp Memory and Anamnesis principles give the fleet's inference layer a philosophical foundation.
A rule said every milestone must link back to its memory node. The rule was written once, then quietly violated more with each passing month — including, in a small irony, by the very mission created to fix the violations it found.
A trust ledger is only as honest as the cost figures sitting next to it. M134 found a vendor with zero cost records at all, M149 spent three attempts finding out how wrong a cheap estimate can be, and M153 caught the audit itself drifting before M154 finally made the whole problem automatic.
Era II — M11. Universalis was online. Now it had to be tested — and the tests revealed the cracks in the foundation.
M122 built the trust ledger clean in five milestones. Then five more missions chased the same bug across the command-center graph before the fleet found the typo underneath it all.
M130 gave the round-robin its fair, correctly-tagged sample. M145 closed it and read the numbers back — a real dollar figure, a real quality signal, and an honest admission that the sample is still too thin for the last word.
Era V — M57–M59. Hermes analysis cuts token waste. GCP probe eliminated fleet-wide. CLI footer gains live CTX% and session cost. The fleet learns to measure itself.
Why DboxShim keeps the stable Distrobox CLI path while opening parallel streams for Dagger tests and the upcoming Distrobox Go module.
Era II — M16, M18, M19. The fleet needed infrastructure: a third ship, a private DNS, and Caravella tightened. This is how three ships became a fleet.
M121-M123: the fleet opened its command chair to three vendors at once — <Term id="claude">Claude</Term>, <Term id="gemini">Gemini</Term>, <Term id="gpt">GPT</Term> — and had to answer, before any of them touched a shared task, how three tongues could earn trust without one of them grading its own homework.
The scorecard in The Rotation's Verdict looked honest because nobody had checked whether the labels underneath it were true. M137-M139 found out they weren't - a mis-grouped graph, three delegations that ran on a vendor other than the one they were logged under, and a fix deployed to a container nobody was actually serving traffic from.
Era II — M21, M23, M24, M25. The fleet cleaned its memory, automated its chronicle, gained conditional intelligence, and made trust a measurable thing.