Repository navigation
Releases
Releases · Abyssinia-Labs/wake
wakectl 0.6.0
Added
- A follow-up continues the run that answered the last summons. A question on a ticket after its run finished goes to the same agent: its session, in its worktree, under its name. In herdr it is asked in the tab still open; otherwise the session is resumed. A new session starts only when there is none from the last 14 days.
- The listener tidies run worktrees. A day after its branch is merged (squash merges too), or a week after its last run once everything is pushed, a worktree Wake made goes, with its branch. One with uncommitted or unpushed work, a run in it or its herdr tab open stays.
- Wake for GitHub: add a device to watch your runs. On Pro,
github.wakectl.dev/devicesgives a code for one of your agents, and Wake on your phone, in a browser or on your Mac takes it to list that agent's runs as they go and how each ended. The same page forgets a device. On Free it offers Pro. - Wake for Mac, a download on wakectl.dev. The menu bar app over the same Wake: it carries
wakectland runs the listener, so a Mac needs no Bun. It needs macOS 26, and updates itself, so the command updates with it. On Linux and in the terminal, install with Bun as before.
Changed
wakectl pruneruns the tidy now, and says what it kept.--daysis how long a pushed, unmerged worktree waits, 7 by default rather than 14.
wake-server 0.5.0
Added
- The server serves wake/v1's calls for devices that watch. An agent's owner adds a phone or a browser with a code
issueWatchCodemakes in your own interface, and the device lists the agent's runs and answers asks.wakeHandlerservesPOST /wake/v1/watchand the watch routes (runs, as server-sent events likepending,answer,places,watching,push,activity), with CORS for a browser. A place key opens none of them, and a watch key none of Wake's. plansays what an agent's plan includes. It is asked at each watch call. On a plan without devices, every watch route butwatchinganswers402 not-in-plan; withoutplan, everything is included.POST /wake/v1/waiting: Wake can say a run waits on its owner. The owner's devices see it, andforSubject,forOwnerandforScopeanswerwaiting: true.summontakesaskerName. The owner's devices show who asks.devices,forgetDeviceandalertTargets. An agent's devices for its owner to see and remove, and which of them want an alert, and how to reach each.
Changed
- New tables and columns: migrate before you deploy.
wake_watch_codesandwake_devicesare new, andwake_places.last_claim_at,wake_summonses.asker_nameandwake_summonses.waitingare added. drizzle-kit finds them,WAKE_TABLES_SQLrun again adds them, and on SQLiteWAKE_SQLITE_DEVICES_UPGRADEdoes, once. - A
WakeStoreof your own needs the device methods.insertWatchCode,useWatchCode,insertDevice,deviceByKeyHash,devicesOf,patchDeviceandlatestOf. - Pairing takes control characters out of a machine's name. A device shows the name as it reads.
wake-convex 0.4.0
Added
- The component serves wake/v1's calls for devices that watch. An agent's owner adds a phone or a browser with a code
issueWatchCodemakes in your own interface, and the device lists the agent's runs and answers asks.exposeWatchApinameswake:runs,wake:answer,wake:places,wake:watching,wake:pushandwake:activity, andregisterRoutesaddsPOST /wake/v1/watch. A place key opens none of these calls, and a watch key none of Wake's. plansays what an agent's plan includes.exposeWatchApireads it from your own tables at each call. On a plan without devices, every watch call butwake:watchingthrowsConvexError("not-in-plan"); withoutplan, everything is included.wake:waiting: Wake can say a run waits on its owner. It's optional, inexposeApi. The owner's devices see it, andforSubjectandforOwneranswerwaiting: true.summontakesaskerName. The owner's devices show who asks.devices,forgetDeviceandalertTargets. An agent's devices for its owner to see and remove, and which of them want an alert, and how to reach each.
Changed
- Pairing takes control characters out of a machine's name. A device shows the name as it reads.
wakectl 0.5.0
Added
wakectl limitssets runs at once and when a run is stopped.wakectl limits runs 1lets one run go at a time (two at most, as before), andwakectl limits stop-after 90stops a run after 90 minutes rather than 120. Both take within half a minute, with no restart.wakectl rootssets where Wake looks for your clones.wakectl roots add ~/codeandwakectl roots remove ~/codechange the folders setup asks about, one and two folders down;wakectl rootslists them.wakectl setup --jsonfor an app that runs setup its own way. It prints what setup would ask, with today's answers: this machine's checks, the folders it offers, the clones there and which apps asked for each, run modes, models and whether the listener starts at login. It changes nothing, and no key is ever in it.- Each run keeps the link to what it was summoned for.
wakectl status --jsonandwakectl open --jsongive each run'surl, the page on the app it works on, andherdr, whether it runs in a herdr tab or headless, so a program can link to the ask and offer the right way back in. Runs kept by an older Wake have neither. wakectl statusshows a run in herdr waiting on you. When a run stops on a question in its tab, a tool asking permission or a folder to trust,statussays so with the question as the tab shows it,status --jsongives the run'swaiting(sinceandwhat), and the log says it. It clears when the run goes on. The question stays on your machine: an app that supportswake:waitinghears only that the run waits, and when it goes on.- wakectl as a single file, no Bun needed. Each release on GitHub now carries a binary for macOS (Apple silicon and Intel) and Linux (x64 and arm64), each with its SHA-256.
wakectl installstarts the listener from it, andwakectl updatetells you when a newer one is out rather than reaching for npm. - Stop a run now:
wakectl stop amber-heron. It ends a run going on this machine, found by its name, its ticket or its session id. A headless run's tool is stopped; a run in herdr is interrupted with Esc and its tab kept. The app shows the run couldn't finish, stopped by its owner. - Each run's steps are kept on your machine, a file per run. The lines
wakectl logsshows for a run also go to its own file underruns/in Wake's folder, as JSON lines, so a program can show one run's steps without reading the whole log. Readable by you alone, never sent anywhere, and removed when Wake forgets the run after its last 200. --jsonfor scripts and apps that drive wakectl.wakectl status --jsonprints the listener, each pairing's connection, the runs with their resume commands, the approved repositories and the settings as JSON;repos,model,doctorandopentake it too. Only the forms that change nothing do, and no key is ever in it.- wake/v1 has calls for devices that watch. An app can let an agent's owner add a phone, a browser or a desktop app that lists the agent's runs and answers asks. Each holds a watch key, made only from a code the app shows its signed-in owner, so nothing a run can reach can say yes for them. An optional
wake:waitinglets Wake say a run waits on its owner. The spec has them now,wakectl checkprobes them, and Wake callswake:waitingwhen a run in herdr waits. wakectl checktests an app's calls for devices that watch. Where the app makes watch codes, it checks thatPOST /wake/v1/watchand every watch call refuse what they must: a code nobody issued, a key that opens nothing, and this machine's own place key when it is paired with the app, since a place key must open none of them. It tries the optionalwaitingtoo.- Wake for GitHub: Pro, a plan for your own account. $8 a month or $80 a year, for what Wake's paid apps need: your runs, and what they ask you, on your other devices. Until those apps come, it does what Free does. Team stays the organisation's plan, with its policy and audit; like Team, Pro isn't on sale yet.
Changed
wakectl opennever starts a run that is still going a second time. For a run going in herdr, it brings the run's tab to the front and opens herdr in your terminal (from inside herdr, the tab alone). For one going headless, it says so and thatwakectl stopends it, where it used to resume the session beside the run and fight it. A run that has ended resumes in your terminal as before.wakectl installleaves an app's listener alone. When an app on your Mac runs the listener itself, as its own login item,installwrites no second LaunchAgent under the same name anduninstalldoesn't stop it: each says an app runs it, and that start at login is that app's setting.doctorandsetupcount it as installed, andpairno longer says to install.
wakectl 0.4.1
Fixed
- A Cursor run that isn't signed in to the app no longer starts blind. Cursor keeps MCP sign-ins per folder, so a run on a new branch started signed out and couldn't read what it was asked. Wake now checks first: the ticket says why the run didn't start,
wakectl logsgives the one command that signs that folder in, andwakectl doctorchecks that Cursor has the app's MCP server.
wakectl 0.4.0
Added
- Pick the model each of your agents runs with.
wakectl model claude opus --effort high, or answer the new question inwakectl setup, which offers Claude Code's aliases, the models Codex lists, and the onescursor-agent modelslists for your account. With nothing picked, a run uses its tool's own default, as before.wakectl statusshows the choice. wakectl updatesays what changed since your version. The headline of each release in between, security fixes marked, and a link to the rest. You'll see it from your next update after this one, since the version being replaced is the one that runs it.- Wake for GitHub: plans, with organisation policy and an audit on Team. The basic loop stays free. On Team, an organisation's admins choose which repositories may summon an agent and who may, and see the organisation's runs, at
github.wakectl.dev/org/<login>.
Fixed
- A run's signature names its real tool and model. The example it was given said Claude Code (Opus), and Codex, Cursor and Sonnet runs copied it.
Security
- Wake for GitHub: uninstalling the app deletes which issue each summons was about, and who asked. It deleted the summonses but kept that record.