← Appolis
🛠 the agent factory · v0.59.0 · worker 0.56.0

DAEDALUS

The internal factory that builds and runs the suite's AI agents — local first, on the owner's own machine, with every run measured and every ship closing its own record.

Daedalus v0.59.0 · worker 0.56.0 · 2026-11-07 · source of truth: CHANGELOG.md · the deck's record slide is read from the ship's deck-row files
🏭 What it is

Four things, one factory.

🔁
Routines — scheduled runs that work unattended, walking a step door one check at a time. Each step's check reads the run's own log, never a line someone typed.
🏺
Daimones — free specialists on the owner's PC: a reader that maps, a refuter that argues back, a scribe that drafts. The strong model keeps judgement; the daimones do the bounded work for $0.
🚢
The ship — nine door steps, one guarded deploy script, a go form the owner answers. The ship writes the proof of that go itself and stops loudly, by name, if it cannot close its record.
🧾
The bill — a ledger and an odometer: tokens, tools and wall time measured from the run, never estimated by hand; three named profiles (efficient · balanced · fast), each one timed.
🚪 How a run walks

Plan, build, check, run, ship — each behind a door.

A job is a chain of steps. The door holds a check for every step and refuses a line it cannot prove; a refused line stops the run by name instead of letting it drift.

1
Plan — the strong model writes the design, the acceptance checks and the job graph, with the bill shown before anything runs.
2
Build and merge — cheaper workers build the pieces in parallel; a merge gate proves the result green before anyone looks.
3
Check and run — every expected line is compared by machine; a mismatch is a stop, not a shrug.
4
Ship, drill, record — the deploy goes through one guarded script on the owner's go, the live drills run against what is served, and the board gets the block the ship script composed.
🎛 The owner's switches

Set once on the routine row, read on every card.

▶
Keep going — ticked, a run works straight through to the context tripwire; unticked, it stops at each phase end. Every routine card shows which way it is set.
⏸
Pause — right beside it: a routine that should not run tonight does not, and says so on its card.
⚖
Profiles — efficient (slower, cheaper), balanced, fast — a named way to run a job, each one measured, picked per routine.
🗝
Your own key — the dashboard opens with the owner's key; a setting that is not a plain yes or no is refused by name and nothing is saved.
📜 The record · to v0.59.0

What shipped, in the words the ship wrote.

45 row(s), read from the ship's own deck-row files at build time. The full plain-English changelog is the breakdown page; the full technical record is DAEDALUS-DESIGN.md §14.

59.0
v0.59.0 worker 0.56.0 · 2026-11-07 — THE LIBRARIAN'S DESK — (Daedalus release v0.59.0, worker 0.56.0, local 0.29.0, deployed 2026-11-07) Kosmetor's first study shelf: a librarian field with 19 cited cards on memory, search and filing, six from our own measured test and thirteen from free public sources, techniques only. Private items are now left out of every count for people who may not see them. The six-hour watch keeps the repository clean (its live state lives in an ignored folder; the committed copy is a deliberate snapshot). On Tyler's pick the watcher reads the Higgsfield blog index page — that one address only — and queues new post links for a person to read.
58.0
v0.58.0 worker 0.55.0 · 2026-11-07 — DAEDALUS UNIVERSITY v1 — (Daedalus release v0.58.0, worker 0.55.0, local 0.28.0, deployed 2026-11-07) The university watches its sources by itself every six hours — GitHub release feeds and the YouTube channels, public and free — queues what is new, and lets the free local reader make cited study cards from it. The floor now shows the real last check, how much is new and how much is waiting. The fields are grouped into the twelve categories Tyler approved, each with its seats, cards and gaps.
57.0
v0.57.0 worker 0.54.6 · 2026-11-07 — THE WALKER'S INBOX — (Daedalus release v0.57.0, PC side only, local 0.27.0, no deploy needed) Kosmetor's board walk now counts the Inbox and the no-project notes as two real numbers, starts a listing over when the board changes under it, stops loudly instead of pretending an empty page, and lists the clock-in lanes in groups of ten.
56.9
v0.56.9 worker 0.54.6 · 2026-11-07 — THE FLOW'S ORDER — (Daedalus release v0.56.9, worker 0.54.6, deployed 2026-11-07) a session can set Project Flow's project order and clear a saved project layout when you ask it to.
56.8
v0.56.8 worker 0.54.5 · 2026-11-06 — THE FLOW'S ROOM — (Daedalus release v0.56.8, worker 0.54.5, deployed 2026-11-06) Project Flow holds up to 200 cards per project, up from 10, so a full rollout fits.
56.7
v0.56.7 worker 0.54.4 · 2026-11-06 — THE MEMORY SCOPE — (Daedalus release v0.56.7, worker 0.54.4, deployed 2026-11-06) a member reaches only their own part of the memory; nothing changes for the owner.
56.6
v0.56.6 worker 0.54.3 · 2026-11-06 — THE KEEPER'S FIRST STEP — (Daedalus release v0.56.6, PC side only, local 0.26.5, no deploy needed) a very large note no longer stops Kosmetor's filing, a held-back write is reported as held back, and each cleanup runs under its own logged run.
56.5
v0.56.5 worker 0.54.3 · 2026-11-06 — THE CREDIT CHECK — (Daedalus release v0.56.5, worker 0.54.3, local 0.26.4, deployed 2026-11-06) the API credit card checks with Anthropic about once an hour, and a member who opens Daedalus lands on their own area.
56.4
v0.56.4 worker 0.54.2 · 2026-11-06 — THE SIGN-IN CHECK — (Daedalus release v0.56.4, worker 0.54.2, deployed 2026-11-06 with v0.56.5) a security tightening from a suite-wide review of how signed-in changes are accepted. Nothing changes in how you use Daedalus.
56.3
v0.56.3 worker 0.54.1 · 2026-11-05 — THE METER'S ORDER — (Daedalus release v0.56.3, PC side only, nothing deployed) an idle Claude session can no longer put an old percent back on the Claude plan meter. Within one week the figure only goes up, and a report for a week that has ended is ignored.
56.2
v0.56.2 worker 0.54.1 · 2026-11-05 — THE FLOOR'S FIXTURES — (Daedalus release v0.56.2, test housekeeping, nothing deployed) the automatic checks now test the session limits that are really live, 400,000 to warn and 500,000 to hand off, instead of an old copy. The floor page's size limit moved to one named place with room to grow.
56.1
v0.56.1 worker 0.54.1 · 2026-11-05 — THE FLOOR'S LOAD TIME — (Daedalus release v0.56.1, a small fix to v0.56.0) the floor took about 15 seconds to open, and now it does not make you wait for that. Nearly all of the wait was one report counting its records one document at a time, for a figure the new page does not show. It now asks once, and the figure is the same.
56.0
v0.56.0 worker 0.54.0 · 2026-11-05 — THE FLOOR — (Daedalus release v0.56.0) the new Daedalus floor is the page you approved, now with live numbers. Home says in one sentence where the Claude plan stands, and every tile on it opens a pop-up with more detail and a button that goes on to the full page. The Claude and ChatGPT plan cards refresh on their own every 30 seconds while you are looking at them and say how long ago the reading was taken, and this costs no model call. Plan work is always a share of the plan; dollars show only for API credit and a job's API charges. On a phone each page fits one screen with a bottom rail and a Hub button; on a computer the Hub button stays top-left with the sections down the side. The routine settings and the game plan editor still work as they did, now behind Settings on Routines and Edit plan on Work, for the owner only.
55.1
v0.55.1 worker 0.53.0 · 2026-11-05 — THE HONEST NOTE — (Daedalus release v0.55.1, a small fix to v0.55.0, PC side only) the weekly note about where the Claude plan went used to say "filed" even when nothing had reached the board. This PC is only allowed to propose changes, so the note is now reported as "proposed" with its proposal number, and a Claude seat files it. When a note really is filed it shows its number, and when neither happens the run says it failed and tries again next time. The Daedalus server is unchanged (worker 0.53.0), local 0.26.0 -> 0.26.2.
55.0
v0.55.0 worker 0.53.0 · 2026-11-04 — THE FLOOR'S NUMBERS — (Daedalus release v0.55.0) the numbers behind the new Daedalus floor, ready before the floor itself. Doors now say whether they are answering from the calls they really answered, not from an old heartbeat; the Claude and ChatGPT plans show how much of the week is used and when it resets (a plan that has already reset reads as a new week, never as a stale number); where the week's Claude plan went is shown as a share of the plan by project, never as a price; the version tiles close themselves when the version has shipped; jobs get short readable titles and an "about N% of a week" estimate; Theatron can ask which profile a routine runs under; Kosmetor's clean-up runs can be read. In these new numbers, dollars appear for API credit only; the page itself changes in the next release. On your PC the Claude percent is picked up live from the status line, the Codex percent from its own session files, and one note per week is filed to Kosmos saying where the plan went. The warning lines move to 400,000 and 500,000 tokens. The Daedalus server gets one new database table; worker 0.52.2 -> 0.53.0, local 0.25.3 -> 0.26.0.
54.3
v0.54.3 worker 0.52.2 · 2026-11-04 — THE LINK DOOR — (Daedalus release v0.54.3, a small fix to v0.54.2) Kosmos talks to Daedalus over a private connection first, and that connection had not been told about the new seat meter, so Kosmos's context meter would have shown "off". It now lets the meter's two reads through, and nothing else. Worker 0.52.1 -> 0.52.2, no database change.
54.2
v0.54.2 worker 0.52.1 · 2026-11-04 — THE SEAT METER — (Daedalus release v0.54.2) every Claude chat now carries the work-shift number it clocked in with, so Kosmos can show, beside each seat in an Agora chat, how full that seat's memory is and what it has cost your plan. Kosmos has its own read door for this, open only to Kosmos. Readings of chats that are open right now refresh every 2 minutes instead of every hour. A one-time re-measure also files the 364 older chats that were never tagged with a project, so cost-by-project stops reading "unfiled". The Daedalus server gets one new database column; worker 0.52.0 -> 0.52.1, local 0.25.2 -> 0.25.3.
54.1
v0.54.1 worker 0.52.0 · 2026-11-04 — THE ODOMETER REVIVAL — (Daedalus release v0.54.1, local only) the meter that counts how big each of your Claude chats has grown, and what it cost your plan, had quietly stopped working. It now works again, on your PC only: it can read even the giant chat files that used to make it crash, it reports the newest answer instead of an old one, it recognises today's Kosmos and Appolis connector writes, it fills in which project and routine a chat belongs to from the chat itself, and it shows the plan cost of each chat next to the size. Its warning lines now sit where you set them (500,000 and 700,000 tokens, 375 turns). A new sweep gathers every chat touched since the last time and reports them, and it runs by itself after the hourly refresh task. The plan-limit reading that said "bad shape" now shows your weekly figure when Anthropic sends no five-hour one. Nothing is online yet: the Daedalus server still needs its next update before the new reading rules apply there, and the first real automatic sweep has not yet been seen running. Worker 0.52.0 unchanged, no database change, local 0.25.1 -> 0.25.2.
54.0
v0.54.0 worker 0.51.0 · 2026-11-01 — THE ROAD-PICKER — (Daedalus release v0.54.0) before any app spends your money or your plan, it asks Daedalus one question: "which road for this job?" The answer is in your order - the free helpers on your PC, then your Claude plan through a working session, then your ChatGPT/Codex plan as a proposal you review, then your pay-as-you-go Anthropic credit - and the credit is used your way: not at all below 80% of your plan, only for a job that cannot wait from 80%, more and more from 90%, the overflow at 95% with the last 5% of your plan kept for the work only a session can do, and a fair split once the credit is down to $20. Every answer shows the estimated cost first, which band it applied, and how much credit is left - a live number counted from every call and checked against Anthropic's own report when you give Daedalus an admin key; the only thing you ever type is a top-up, once. When it cannot see something, it says exactly that instead of guessing. Nothing runs from this answer; the first app to ask is the Kosmos chat. Worker 0.51.0 -> 0.52.0 (minor), no database change, local 0.25.0 -> 0.25.1 (your plan percentages now reach Daedalus; never the sign-in).
53.0
v0.53.0 worker 0.51.0 · 2026-10-31 — THE TICK THAT LANDS — (Daedalus release v0.53.0, local only) the hourly task that refreshes the Motion dashboard has found the Cloudflare tool every time since 11:44 UTC and then watched it crash before it answered, five runs in a row, while the same tool works from inside the Claude app. The explanation was that the tool's sign-in was saved inside the Claude app's private folder so the task starts signed out; the small fix for that went out first as its own patch, and the timer's own runs since then say it was right: after one red run (Cloudflare refused the read right after the tool renewed its sign-in token) the task went green at 17:06 and again on its normal hourly run at 17:44 UTC, the first green runs it ever had; this version retries a read refused that way once. The three checks that came back red last time were the checks' own fault and now read what the page really says, the release steps read their own results instead of special cases, a check that never ran before the live tests is now required to pass first, and the cost report finds a reviewer that worked in a second chat window. Nothing on the website changes and nothing is deployed; the local door on your PC moves to 0.25.0 with no database change. Not yet proven: that retrying once cures that one refused read (it has happened once), and the live checks that need your paste or a new push from the Claude Motion lane.
52.4
v0.52.4 worker 0.51.0 · 2026-10-31 — THE LOGIN THE TIMER SEES — (Daedalus release v0.52.4, local only) the hourly task now finds the Cloudflare tool but crashed every time it asked it a question. Best guess: the tool's sign-in was saved inside the Claude app's private folder, so the task started signed out. The task now says which sign-in it can see, points the tool at the app's copy only when there is no other, and saves the tool's own error on every failure. The sign-in file is never read, copied or moved. No website change.
52.3
v0.52.3 worker 0.51.0 · 2026-10-31 — THE PACKAGED PREFIX — (Daedalus release v0.52.3, local only) the hourly task's new log line named the real reason it kept stopping: the Cloudflare tool lives in the Claude app's private copy of your AppData folder, which programs outside Claude never see. The task now looks there too. No website change.
52.2
v0.52.2 worker 0.51.0 · 2026-10-31 — THE TIMER'S EYES — (Daedalus release v0.52.2, local only) the hourly task on your PC now writes one line to its log at the start of every run saying which folders it sees and whether it can find the Cloudflare tool, so the real reason it keeps stopping can be read instead of guessed. No website change; no secrets recorded.
52.1
v0.52.1 worker 0.51.0 · 2026-10-31 — THE TIMER PATCH — (Daedalus release v0.52.1, local only) the hourly task on your PC that refreshes the Motion dashboard now finds the Cloudflare tool even though Windows starts it without the setting that names your npm folder: it falls back to your own user folder's AppData. No website change; the proof is the timer's own next tick going green.
52.0
v0.52.0 worker 0.51.0 · 2026-10-31 — THE PROOF THAT READS + THE FIRST REAL TICK — (Daedalus release v0.52.0) you can now rely on the checks and the timer around the Motion dashboard: the two live checks that came back red last time were wrong and the app was right, so they now read what the page really says (the check that a second person sees only their own projects asks Daedalus which account it is talking to, accepts an empty page as the correct answer for someone who owns nothing yet, and shows a kept line when they own one; the routine-file check compares the checksum Daedalus actually serves). Plainly, the hourly refresh timer from the previous release never ticked green on its own: the scheduled task could not find the Cloudflare tool (wrangler was not on the task's PATH) and stopped at its first step both times it ran, at 03:44Z and 04:44Z, while Windows still showed "Last Result 0", and this version fixes that by looking in your npm folder as well and by showing the timer's own ticks in its status line, with its first green tick read at the live check rather than assumed. When the timer does write, it now reads every page it saved back and rewrites a damaged one on the next tick, a damaged page list is repaired instead of ignored, and each person's episode list on the Motion page now follows the stored pages' own project words, so an episode that moves to another account leaves their list on the next check. Nothing here changes who sees what, and a deploy still waits for your written go. Not yet seen live: a refresh that finds a genuinely new push (the Claude Motion lane has to push first) and the member's kept line (it needs one shared row and your one console paste); the local door on your PC moves to 0.24.0, with no database change.
51.0
v0.51.0 worker 0.50.0 · 2026-10-30 — THE REFRESH LOOP + THE MEMBER'S PROOF — (Daedalus release v0.51.0) the Motion dashboard at daedalus.appolis.app/motion can now keep itself up to date. A small task on your PC (no Claude chat, no cost per check) looks at the Claude Motion lane once an hour, or once a day if you prefer, and when it finds a new push it puts the new pages into Daedalus's own store, which the dashboard shows first; so the dashboard moves within one check, with no deploy and nobody pulling anything by hand. A check that finds nothing new changes nothing, and if a stored page is ever damaged Daedalus quietly shows the bundled copy instead. Each person still sees only the projects of their own account on these pages, exactly as before, and a deploy still waits for your written go. Two more things come with it. First, the check that a second person really sees only their own projects can now be run for real once Appolis has Daedalus on its grant list; until then it is skipped and says so, never counted as passed. Second, a routine now counts as done only when the app actually lists it, not when a file is written (that is how SolarWRLD's routine looked installed and wasn't), and the run's own road carries the last review's fixes. Not yet seen live: a refresh that finds a genuinely new push (the Claude Motion lane has not pushed since the last copy, so the first real one is the next version's proof). No database change; the local door on your PC moves to 0.23.0.
50.0
v0.50.0 worker 0.49.0 · 2026-10-30 — THE ACCOUNT'S OWN MOTION + THE SOLARWRLD ROUTINE — (Daedalus release v0.50.0) the Motion dashboard at daedalus.appolis.app/motion stops showing everybody your projects. Before, anyone whose Appolis ID passed the gate saw your 37 episodes, your whole project list and every roadmap, and the Flow and Watch buttons on Kosmos for Raven's projects took him to your items. Now each person sees only the projects of their own account (Raven sees Hockey Lab and the FLIP 8 roadmap, not Daedalus, Hermes or FLIP 7), a link to anyone else's project answers a plain "not yours", you as the master admin see everything exactly as before, and nobody signed in still gets the sign-in page. SolarWRLD also gets its own Daedalus routine, solarwrld-session, built like the Hubs, Warehouses and FORGE orchestrators: Run now, manual only, one run = one phase of one blueprint, a deploy only on your written go, and its first run is the review and merge of Codex's Stage 0-7 work; its small starter file is installed on your PC. The run's own road also carries the last review's five fixes (the bill adds up a seat that had to be started twice, a missing hands line is a named skip, the carried drills print one saved line, the pull-gate control checks its stage first, the deploy line pins both version ids on a resume). Not in this version: a member saving their own Project Flow order (that write stays yours), retiring the old Kosmos key, and the automatic refresh of the Motion pages. No database change, no local change.
49.0
v0.49.0 worker 0.48.0 · 2026-10-30 — THE MOTION CATCH-UP + THE SPLIT ROAD — (Daedalus release v0.49.0) the Motion dashboard at daedalus.appolis.app/motion is up to date with what the Claude Motion lane pushed on October 30: 37 episodes instead of 20 (Hockey Lab, SolarWRLD, QR Codes, Blueprint Skill, FLIP 7 new LP, Recharge Ops, FLIP landers, FML end cards, TipLift, Agora and the rest), Hockey Lab (Raven's, shared) and SolarWRLD on the project list, Hubs showing its routine, and Project Flow saying in plain words "Someone else saved first, so their order is loaded now" when two people save at once. After this, the close-out's own check reads drift=0, nothing behind. Daedalus also accepts the shared-project names the board uses (acc_6751cc16e945~hockey-lab) as roadmap and plan document names, so the Claude Motion role, now this lane's, can push those two roadmaps with the roadmap tool instead of being refused by name; only the name shape is new, and who may read or write what did not change. The run's own road is mended where v0.48.0 found it broken: when Claude's browser is refused a live write by its safety filter, the run no longer calls that a failure of the product; it tries one harmless write first, names the result, hands you ONE paste for your browser console whose answer lands as a file in your Downloads, and grades each half on its own. Not in this version: retiring the old Kosmos key (it waits until Kosmos is seen working over the new wire; this version only lists every place in this repo that still touches it), /motion per account, the player on the floor. No database change, no local change.
48.0
v0.48.0 worker 0.47.0 · 2026-10-30 — THE ROADMAP STORE — (Daedalus release v0.48.0) each project's roadmap - what is planned and what waits on what - can now live in Daedalus as a document of its own, one per project, instead of only in a bundle of files copied into the Motion mirror. Once a roadmap is saved, the Motion dashboard's roadmap view and Project Flow read it on the next load; until then nothing changes and the dashboard shows today's bundle. There is no key to make, copy or paste: the Claude Motion lane's chat already reaches Daedalus through the Appolis connector as you, and it gets one new tool that saves, reads, lists and removes a project's roadmap, with the record saying which account saved it; you can also do the same signed in on the page. A roadmap that is not the shape the dashboard reads is refused when it is saved, with the reason in plain words, and if a bad one ever got in anyway the dashboard falls back to today's bundle and says so instead of showing half a list. Saving a roadmap does not ping Kosmos. The helper keys (Claude, ChatGPT, Codex) and Kosmos's key cannot write a roadmap; only you can. Honest limits: a chat learns a new tool only when its connector reconnects, so the first save from the Claude Motion lane waits for that, and the live proof of the tool may need a fresh session. No database change, nothing in the dashboard's own files moved. Next on the road: retiring the old Kosmos key (v0.49.0, only once Kosmos is seen on the wire), then the player on the floor.
47.0
v0.47.0 worker 0.46.0 · 2026-10-29 — THE AUTO LINK — (Daedalus release v0.47.0) Daedalus and Kosmos are now joined by a wire, not a key. The two apps live on the same Cloudflare account, so Cloudflare connects them directly - there is nothing to make, copy or paste, on either side. Every time you save or delete a Project Flow plan, the order or the project list in Daedalus, and every time a chat reaches its clock-out step through the Daedalus door, Kosmos hears it right then: which plan or routine, which phase, which account, when. A message that does not get through is kept and sent again with the next one, and one Kosmos refuses is written down as refused - you can see the last one, how it went and how many are waiting on a page of Daedalus's own. Kosmos also gets a door of its own back into Daedalus that nobody else can reach: it can read what is next and the plans, and write the project list and plans, with the same "refused if someone saved first" rule and no key. The old key stays working until Kosmos has switched over, then it is retired (v0.49.0 at the earliest). Honest limits: Kosmos has not turned on its half yet, so today only a plan save landing on Kosmos can be shown, and Kosmos will answer "no" to the order, project-list and end-of-session messages until its to-do ships - Daedalus records those as refused. No database change, nothing in the dashboard's own files moved. The roadmap store moves to v0.48.0, then the player on the floor.
46.0
v0.46.0 worker 0.45.0 · 2026-10-29 — THE PROJECT LIST — (Daedalus release v0.46.0) the project tiles on the Motion dashboard stop being a list somebody copied by hand. Kosmos can now write its list of projects into Daedalus, with the same key and the same "refused if someone saved first" rule it already has for project plans, and the dashboard shows that list on the next load: add, rename or move a project on the board and its tile follows. Until Kosmos writes it, nothing changes - the dashboard shows the list it shows today. A list that is broken is refused when it is saved, with the reason in plain words, and if a broken one ever got into the database the dashboard falls back to today's list and says so instead of showing half of one. Second change, your pick: the helper keys (Claude, ChatGPT, Codex) can no longer read your plan documents or the address Daedalus pings Kosmos at - only Kosmos's key and you, signed in, can; a helper that tries gets a named "not yours" answer. The live key test that was skipped last time is carried into this version and run by you in your signed-in page. No database change, nothing in the dashboard's own files moved. Kosmos's half is its own to-do on that board; the road goes on with the roadmap store (v0.47.0) and the player on the floor (v0.48.0).
45.0
v0.45.0 worker 0.44.0 · 2026-10-28 — THE FLOW WIRE — (Daedalus release v0.45.0) Project Flow, the order of work you drag around in /motion, now has a real door in Daedalus. Every project has a link that stays put: daedalus.appolis.app/motion/flow/<project> opens its Project Flow and /motion/open/<project> opens its episode, behind your sign-in - the links the Kosmos project tiles will carry. Kosmos can now read your flow and the "what's next" list, and write a project's plan back through a key you mint on Daedalus; it can save a plan and nothing else (no deletes, no other settings), and if someone saved first it is refused instead of overwriting you. Every time you save the flow, Daedalus pings Kosmos at an address you set once, and the save tells you plainly whether the ping landed; a save that came from Kosmos never pings back, so the two cannot chase each other. The Game Plan tile box on the floor gets an "Open in Motion" link. Nothing in the hub itself moved. The Kosmos side is its own to-do on that board; the next versions on this road bring the project list, the roadmap and the player into Daedalus one at a time.
44.2
v0.44.2 worker 0.43.2 · 2026-10-28 — THE PATCH SHIP ROAD — (Daedalus release v0.44.2) a green patch can now deploy in the same run that built it, as long as you have already said yes. The chat that builds the patch ends its report line with the written yes you gave (the form pick, never a chat's word); Daedalus then walks the deploy as five checked steps - guard, deploy, health check, shipped facts, mirror - and only after the last one passes does it close the run. If you have not said yes, nothing changes: the patch is reported, nothing is deployed, the run closes as before. A deploy line carrying a different yes than the one on the patch is refused, and a deploy that cannot run stops loudly instead of pretending. The backup deploy script also got the one missing word that made its deploy step fail after a successful deploy.
44.1
v0.44.1 worker 0.43.1 · 2026-10-28 — THE TILES PATCH — the game plan at the top of daedalus.appolis.app now tells the truth: the version being built right now comes first and reads RUNNING, the ones still waiting read QUEUED in your order, and versions already shipped read DONE and sit at the end. Each recent step is a few words instead of a wall of machine text (click a tile for the whole line), one refusal marks only the run it belongs to, and the 30-second refresh shows a small mark in the corner instead of covering the page.
44.0
v0.44.0 worker 0.43.0 · 2026-10-28 — THE GAME PLAN TILES — the first thing on daedalus.appolis.app is your game plan: a row of tiles, one per upcoming version, in your own order. Each tile tells you plainly whether it is queued, running (and where), done, or stuck — and a stuck tile says why, what line it is waiting for and who or what is on it right now. Click to change what a version holds and in what order; drag to reorder the versions themselves. Nothing in a tile is typed by a machine: it is read fresh from the door and your docket every time.
43.1
v0.43.1 worker 0.42.1 · 2026-10-27 — THE MOTION LOOP + THE SECOND TERM — /motion is current again and keeps itself current: the Claude Motion hub is re-copied exactly from your repo (every episode as of Oct 27), your Project Flow arrangement saves to Daedalus through the hub's own store hook instead of a re-cut file, and after every close-out the hands on your PC pull your repo, export the hub and say in one line whether /motion is behind - when it is, the patch is filed for the usual ship chain and your go. And the university's second term is in: your Oct 25 links are source rows (115), the ten repos you approved are cloned read-only and read into 27 new cited cards (52 in all) by the free local reader, the index and the wings are regrown, and your five "look at these" links each carry a one-line verdict. Plus three step-door loop fixes you asked for today: Appolis's real PREFLIGHT line now passes the ship door, a BUILD+CHECK run claims the docs lane (not only RUN+SHIP), and Flip My Life's Run no longer re-plans a finished blueprint - it opens the next round.
43.0
v0.43.0 worker 0.42.0 · 2026-10-26 — THE DOCKET — Project Flow now lives inside Daedalus. Open daedalus.appolis.app/motion with your Appolis sign-in and you get the same hub: every project ranked from 30,000 feet, each project's versions as columns you drag items between, your own items, your episodes — and every save goes to Daedalus's own database (your hephaestus plan came across exactly as you left it). The machine reads it too: every session starts by seeing your order for its project, so routines take their next step from the docket instead of a re-edited instruction, and every session ends by posting a short brief of what changed, shown under that project's episode. Full record: §14.58.
42.0
v0.42.0 worker 0.41.0 · 2026-10-24 — THE UNIVERSITY + THE WINGS — everything you sent us to learn from is now inside Daedalus: the whole storytold org (68 repositories, read only, never built), the agent-skills and codeburn repos, your three YouTube channels, the Higgsfield blog, YaroFlasher and Flashboards, Katana, OmniVoice, Odyssey, Blender, Unreal and the Adobe tools. The university holds its first real technique cards, each one citing where it learned the technique, and the wings grew into sectors of experts — 3D and Blender, Unreal, Adobe, design and gameify, voice, world models, video AI, the ArtCraft stack — every expert naming the card or repository seat it learned from. The floor shows the university and the sectors; every wing, seat and card says whether it is shared or private and who owns it (everything is shared today; private wings come with the version that first holds one). Full record: §14.57.
41.0
v0.41.0 worker 0.39.0 · 2026-10-22 — RAVEN STARTS — Raven's connector gains the step door for the routines he owns: his first routine, raven-session, is seeded, and he runs it one step at a time from his own Claude. Any routine that is not his answers "unknown routine", exactly as a missing one does, and nothing is written; nothing admin opens to him - no go, no ledger, no sandbox, no editor, no floor, no other person's routine - and every call he makes is on the ledger under his name, each step naming its routine. His steps have their own daily cap of 600, so a run never locks him out of the rest. Worker 0.38.3 -> 0.39.0 (minor), no database change, the seed adds one row, local 0.21.2 unchanged.
40.7
v0.40.7 worker 0.40.0 · 2026-10-24 — THE RESHIP — a ship the review marked NOT SHIPPED (or whose deploy parked) is served the ship step again on the next run, and you can say next=RESHIP to reopen it; a drills or close-out line on a held ship is sent back to the ship, and nothing already on record is deleted. A room member's routine is asked in chat instead of a form, a refused clock-in is a tool limit, an honest "nothing due today" is accepted (only a reason that blames the run is refused, now under its own name LINE REFUSED), and every stop step says to append the next run's line to the handoff before it clocks out. Full record: §14.56.
40.6
v0.40.6 worker 0.38.3 · 2026-10-22 — RUN NOW IS RUN NOW — the scheduler banner is how your Run now arrives, so a run never treats itself as unattended or measure-only again: the four routines that still said "PREP AND PROPOSE" with "no Run now press" lose those words, every routine's first step opens with your sentence, and the step door refuses by name a check-in that says nobody is watching, a plan step that does no work because of the banner, and a close that leaves owed work behind a failed review unless your own form says stop. The small road: the first step sizes the job, and a small change gets a short graph - the chat or one cheap builder, the gate and one targeted check, ship on your go. Worker 0.38.2 -> 0.38.3 (patch), no database change, local 0.21.2 unchanged.
40.5
v0.40.5 worker 0.39.1 · 2026-10-22 — THE CHAIN HOLDS + ATTENDED AS YOU SEE IT — every routine's first step accepts attended=no as the session sees it and goes on to step 2 - only what truly needs you waits (the Remote Control setter, refused by the app anyway, is written set=no and is never a stop; a deploy without your written go parks for the hands), and your questions still come as forms. The ship chain no longer stops on its own first lines: an early stop leaves the ship log alone, a timestamp-only pull is green, the review contract has one written copy, and every helper seat is keyed and billed. Full record: §14.54.
40.4
v0.40.4 worker 0.38.2 · 2026-10-22 — THE WHOLE CHAIN — the ship chain closes whole on your real go answer: the deploy is marked by the run that made it, an answered form no longer stops the ship at the park step, and the proof that you approved it names this run's form. The board step runs after the bill and tells the truth; the plan form is composed by script and checked before the build. The step door learns five small roads: a direct ship for a docs-only payload, a form id as the guard's waiver, a keep-going deploy on a go form you already answered, a close that says it has no close-out script, and the docs lane only on a SHIP run; a Run now always counts as you being there, whatever the scheduler's banner says. The deck rows get their home in DECK.md. Worker 0.38.1 -> 0.38.2 (patch), no database change, local 0.21.2 unchanged.
40.3
v0.40.3 worker 0.38.1 · 2026-10-21 — THE OWN RECORD — the ship closes its own record on every road: the proof that you approved it is written from your go answer, the parked deploy is marked by the script (and the ship stops loudly by name if it cannot be), and the record of the nine ship steps is written by the script; the board step's door now takes the full line the script composes, with its timing rows, beside the short line the chat road prints. Worker 0.38.0 -> 0.38.1 (patch), no database change, local 0.21.2 unchanged.
40.2
v0.40.2 worker 0.38.0 · 2026-10-21 — THE RESUME ROAD — a KEEP GOING box beside PAUSE on every routine row in the editor (ticked = the knob absent = keep going, unticked = knobs.keep_going=false), a keep_going yes|no tag on every routine card, validateKnobs refusing a non-boolean keep_going by name; the ship record closes on its own after a tail stop (the owner proof from the go answer, the door record from the cursor's part letters, no figure read from the live worker); the checks that lied made honest; pinned by editor-v0402 KT1-KT5, red at 4036eb8, green after. Full record: §14.49.