Everything
the farm does.
92 things in ten districts. Each one is something Agent Farm does today, checked against its source. Search them, or pick a district.
Never stops
Work keeps going through full contexts, full accounts and its own updates.
- Baton relay⑂ 92At the context level you set (40% by default), a session commits, writes a handover note and a fresh session carries on. No context rot.
- Quota handoveracct-c 71%Before the 5-hour window fills, the next leg opens on the account with the most room.
- Account policies4 built inFour built-in rules (@emptiest, @pool, @standing, @round-robin) or your own. A rule that can pick no account refuses the launch and says why.
- Park and wake2 parksOne account or ten, when all of them are full the chain waits for the first reset and carries on by itself.
- Ouroboros3.931.0-betaAn agent builds a new Agent Farm, restarts it inside VS Code and carries on in the same session. Nobody has to restart anything.
- Sentryon watchAn agent runs the chain instead of a fixed rule: it takes the next item, opens a clean session, reads the result and decides what comes next.
- Brakesbaton.stopA leg ceiling per chain, spend caps per workflow and one stop for a whole chain. Recorded, never deleted.
Any project
One backlog per project, and the project can be anything.
- Any kind of projectany kindSoftware, a documentary series, a thesis, homework, an ad campaign. Same backlog, different work.
- Backlog38 openWrite a thought down in one line. The fleet works the list off in the order you set.
- Track214 closedHow every closed item was closed: which session, which turn, what was tried first. The log can be rebuilt from the sessions.
- Project context6 documentsEvery session starts with a brief built from the project's charter, plan, backlog and worklog, so no agent starts cold.
- Handover lettercold startPassing a project on? One letter says where everything is, what is open, and what was already tried.
- Portfolio4 activeEvery project's phase and backlog on one screen.
- Cost and analysisper itemWeeks, rhythm, how long an item takes from open to closed, and where the money went.
Beyond code
Agents work as a production team and hand you finished media.
- Studio6 scenesAgents work as a production team: idea, research, script, shots, edit, judgement. You get assets, not chat.
- Content recipes3 ready-madeLong video to shorts, a weekly channel rhythm, a video edited in parts: install one and run it.
- Any media modelfal · ReplicateImages, video and voice from any fal.ai or Replicate model, a file on your computer uploaded once you agree, and every job priced by your rules under a monthly media budget.
- Cut on your machine10 recipesTen FFmpeg recipes split, join, reframe, subtitle and level video and sound on this machine, as a workflow step with no model and no tokens.
- Judge and triage2 to triageGive a step its criteria and a vision model checks every shot; a FAIL is made again with the critique. You keep or drop the rest in Triage.
- Brand kit{brand.*}Style, colours, type, voice, the logo rule and recurring characters, kept once per production; every step that asks for them gets the same ones.
- Five stagessource → assetEvery file a run makes is filed by stage, from source to asset, with a record of what made it and what it was made from; a model fills in what it shows.
- Airtable engine10 toolsA local database of media records and knowledge that agents search, save to and update through ten tools; a run can take its material from it.
- Media inboxall_inDrop files in a folder. Agents look at each one, file it by stage and fill in its card; anything unsure is asked, not guessed.
- Straight to PremiereXML + EDLThe scenes become a sequence. Export it as XML and EDL, import it into Adobe Premiere and finish by hand.
- Content calendarMon 07:00What gets made lands on a calendar in your vault. Once you approve, the publish step hands it to your own upload flow.
- What worked{performance}Views, watch time and engagement come back from your own analytics flow onto each published piece, and the next idea is written knowing what worked.
- Spoken briefingdailyThe daily digest read aloud: by this computer for free, or by a voice model you choose.
Recipes and triggers
Work that starts on time, on an event, or from one sentence.
- Workflows8 savedA chain where each step hands its output to the next. A step can be an agent, an HTTP call, an image or video job, a browser run, or a stop that waits for your yes.
- Schedules13 onDaily, weekly or once at a date and hour. A missed run is skipped, caught up, or put to you. With VS Code closed, a small service on Windows still runs them.
- Watch triggers2 watchingA file lands in a folder and an agent wakes up. A live hook, or a post to a private address on your machine, works the same way.
- GatekeepershadowBefore a trigger wakes an agent, free local checks and then Jev, a model that never chats and only decides, say whether there is real work. Off, watching in the shadow, or on.
- Say it in wordsRuns tabType "run the filing batch on manager now". A cheap model turns it into a task, a schedule or a workflow, and nothing is set up until you press Create.
- Queue and limits1/2 runningRuns wait their turn under ceilings, two at once and one per agent by default. A held run starts by itself as a slot frees.
- A record of every run$0.12Every run writes what it said to a file, is marked failed when it crashed, and carries its price. A job you set up often becomes a template, and finished rows move to a monthly history.
- Retries and rounds40 minA run that fails or hits a usage limit tries again by itself, after 40 minutes or at the plan's reset, even on another account. A big job runs in rounds inside one conversation.
- One switchquiet hoursStop every automation at once, pause it for 1, 8 or 24 hours, or keep quiet hours in which nothing wakes unattended.
- Forecast30 daysWhat your schedules will run over the next 7 days, each with its usual time and success rate from the last 30 days.
Works as a team
Agents on different engines plan, split and check each other.
- Missionslead + 3A lead that cannot write plans round by round and hands the work to a team on different engines, one writer per folder, until the goal is met or one of five caps is reached.
- Agent notesa2aAgents leave each other notes and hand tasks over, with a guard against loops.
- Steer and vetomid-runYour steer, veto or constraint outranks everything an agent was told, and reaches a running task with its next heartbeat.
- Console2 waitingEvery session in one column: who is working, whose turn it is, which leg of which chain.
- Countdown questions0:42An agent's question picks the suggested answer when time runs out. A permission request always waits for you, with every stuck session, in Needs You.
- Notices and the queue4 levelsAgents report, remind and ask on one board, by level, and your answer becomes work in the queue. Your side of it: leave a job for an agent, now or when its group fills.
- Live TimelineliveEvery running session on one axis. Jump to one tool call among thousands; a stall raises a flag.
- WorktreesgitParallel jobs work in separate git worktrees, so they never step on each other.
Never forgets
Every session is kept, and what matters becomes a wiki.
- Every session kept1,499Yours and the farm's own runs, from every engine, with cost, tokens, context and the chain leg.
- Any turn, replayedreplayOpen a session, filter its turns, replay it or export it as Markdown.
- Wiki from sessionscompileMark a whole session or only the answers that matter. An agent compiles them into wiki pages, one concept per page.
- Drop box_RAW_Save a chat from any browser into the drop folder and it becomes wiki material by itself.
- Obsidian, the second brainvaultObsidian is the IDE, the model is the programmer, the wiki is the codebase. You curate, agents write.
- Search everythingall enginesSearch every transcript of every engine, by its words or by what it meant, and land on the exact message.
- Memory that gets usedrecalled ×5Each agent's memory files say which were really recalled in 30 days, which the index lost and which contradict each other, and a graph shows how they link up.
- Other machines⇄ badgeSessions synced from your other computers show up in the same lists, with a machine badge on each.
- Reports and digestsweeklyDaily and weekly digests and fleet reports, written into your vault.
Any brain
Five engines, and through routes, almost any model.
- Five engines5Claude Code, Codex, Gemini, Antigravity and OpenCode. What is read off disk says so; only one real test call proves an engine works.
- Plug-in brains3 routesClaude Code and OpenCode can run other models through a route: free or paid, from a provider or an address on your own machine. Five engines, any number of brains.
- Free-model radar04:30 dailySet it up once as a schedule: every morning an agent checks which models are free today and adds new ones as routes, over the bridge.
- Route healthfree checkWhat each route offers is checked for free. A real, paid turn is only spent when you press for it.
- Secret vaultkeychainKeys live in your OS keychain, open only to the agents you choose. A key's value never crosses the bridge.
- Many sign-ins4 accountsSeveral sign-ins of Claude Code or Codex side by side, without signing out. An account is a config folder, not a copied credential.
- Switch the modelmid-sessionChange the model, effort or permissions of a running session from the Console. It takes effect from the next turn, and the conversation stays.
- Eleven doors11 doorsStart a session from the Console, a row, a right-click, a saved setup, the command palette, the bridge or Telegram: all eleven doors carry the same engine, model, account and permissions.
- One agent, every engineAGENTS.mdAn agent's rules, duties and MCP servers are written out in the files Codex, Antigravity and OpenCode read, and kept in step in both directions.
- Any machine${home}Tasks, missions and their history travel between computers in git: one machine's paths are saved as tokens and open on the other machine's folders.
Just ask
You never have to learn the farm. Tell an agent, it sets it up.
- Just askin chatSay "make me a weekly video recipe" in a chat. The agent sets it up through the bridge; you never open a form.
- 579 operationsfarm.opsWhat you can do with a click, an agent can do with a call: a button with no operation fails the build. A few things stay yours on purpose: granting a tool permission, stopping a question's countdown, and a secret's value. New abilities show up in farm.ops by themselves.
- One command lineaf.jsThe bridge is a single shell command, so any agent that can run a command can use the farm. Gatekeeper also comes as an MCP server.
- Agents with a homedutiesEach agent keeps its own identity, duties that run every turn or session, and tools. The Toolbox shows every skill, subagent, command, MCP server, hook and plugin on the machine, adds new ones and tests what an MCP server really offers.
- Capability Matrix30 daysWhich agent has which skill, MCP server and memory file, and which of them were actually used in 30 days.
- Phone appanywhereAnswer, approve with a button and watch the fleet from your pocket in the Agent Farm phone app, paired by a QR code; it can even wake a closed VS Code. Or a read-only web page on any device you allow.
- Telegramone tapA Telegram chat talks to your agents. A card arrives when a turn ends or asks, and one tap answers it. Only the chats you allow, within a daily budget.
- Sound10 tonesA tone for every event, for one agent or for one session: ten made for it and room for ten of your own, with quiet hours and a reminder for an unanswered question.
- Read aloudaloudAn answer, a selection or the daily briefing read aloud, each in its own language, and whoever is asking named after the tone.
- Take it with you.agentpackPack an agent as an .agentpack: its rules, tools, workflows and schedules, never its memory or secrets. Unpack it on another machine.
Measured, not claimed
What every step cost, how much plan is left, and whether the work was good, in numbers.
- Live plan usage5h · weekHow full each account's five-hour window and week are, and when they reset, in the status bar and on the Dashboard. Every engine's limit apart.
- Where it went9 kindsEvery turn split into nine kinds of step, from the model's thinking, billed as output, to local tools that cost nothing to run but make the next call bigger. Every figure is shown raw and weighted.
- Budgets80% warnA daily budget per agent or for the whole fleet, a cap per workflow and per mission, a monthly media budget. A warning at 80%, an error past 100%.
- Cheaper, not stopped→ haikuAs a window fills, automatic jobs wait for it to loosen, or move to a cheaper model. Sessions you opened by hand are left alone.
- Cost alarmsmedianA day that costs far more than an agent's usual, or a week that jumps, raises a flag. The usual is the median of its real days, not a guess.
- Honest numbersreal zerosCosts come from your own transcripts. A free model's zero is a real zero; a missing figure is shown as missing.
- Judge laddercheap firstA check is judged by a cheap model first. Only a fail climbs to a stronger one, and a pass there overturns it, so no retry is paid for nothing.
- EvalsscorecardSaved cases replayed after a change of model or instructions: accuracy, tests passed and what got worse, on a scorecard, not on how sure an agent sounds.
- InsightsfindingsRepeated corrections turn into suggested rules. A finding is fixed, handed to its agent or dismissed, and whether the fix held is tracked.
Safe and yours alone
Guards stop what an agent should not do, and nothing about your work comes to us.
- On your machine127.0.0.1It runs inside VS Code and your work stays on your disk. The live hooks listen on 127.0.0.1 only.
- No telemetry0 bytesNo analytics, no usage statistics: we don't know which features you use, how often or on what. Free never contacts us; a paid licence only checks its key.
- ShellGuard4 rulesBefore a shell command runs, the shapes that garble or hide it are refused: non-ASCII, backticks, long heredocs, long quoted text. The agent is told the safe way: write the text to a file.
- BridgeGuardto the bridgeAn agent that goes searching the disk for a session by its id is stopped and sent to the bridge, which reads it properly.
- Permissions as shownenforcedThe permission a card shows is the one in force; if a policy forbids it, the session is refused, never quietly downgraded. A lint finds blanket allows and dangerous patterns.
- Keys kept outscrubbedRoute keys are scrubbed from whatever a run prints, and Gatekeeper's log keeps only a hash and a length of what it read.
- Browser on a leashallow listA browser step opens only the sites on its allow list, or fails without running, and asks you before each run unless you chose otherwise.
- SuperpowersTDDEngineering disciplines on disk, such as tests first and systematic debugging, with quality gates.
Nothing matches that. Try a shorter word, or .