The AI in your LifeOps

What it doesFive jobs with switches, and Chat without one

Settings → AI lists five jobs, one switch each, and each row names what it sends when it runs. Area naming: A background sweep runs every two minutes, takes entries you wrote in the last two days that have no life area, and files each one under one of your areas, reading the entry’s text and your areas with their descriptions. An area filed this way looks like any other; refiling the entry yourself replaces the app's pick for good. Each entry gets one verdict: if no area clearly fits, the entry stays unfiled and is never asked about again. Document filing: the same sweep, for documents. Every few minutes it takes a document with no life area and files it under one of yours, reading the document’s name, kind, description and text. One verdict per document, and a document it cannot place stays unfiled until the next time you add, rename or remove a life area, which asks about it again: “none of these” is an answer about the areas that existed when it was given. Filing a document yourself, from under its title, replaces the app’s pick for good. Task filing: the same sweep again, for your task list. A task with no life area is filed under one of yours, reading the task’s words, its kind and its note, and only while it is still open. A task it cannot place is asked about again when your area list changes, the same as a document. A task under a stream skips the model entirely and takes its stream’s area, which is why that half keeps working with AI switched off. Filing a task yourself, from its detail sheet, replaces the app’s pick for good, and clearing the area yourself keeps it clear. Transcribe a recording: a video or a voice memo you post is transcribed, and the transcript becomes a title, a summary and a cleaned-up read-through. It runs only when Transcribed recordings is switched on in Settings → Features and a key is stored for it. It is the one job that sends a file rather than text. See Your videos & photos. What changed in a version: when a document or a stream is sealed as a new version, this writes the line saying what moved between it and the one before, and for a stream, whether the target itself moved and from what to what. It sends the two versions’ text.

Every job on that screen also names the model it runs on and the prompt it runs with, and both are yours to change. The Model row on each job offers Opus, the accurate default, and Haiku, the cheap and fast one. Prompt beside it opens the words the job sends, which start as the ones this repository ships. Your edit replaces them until you clear it, and clearing it puts the shipped prompt back.

Chat is the sixth thing that runs on your Anthropic key, and it is not on that screen. The speech bubble in the top corner opens one conversation. Asking a question sends the recent thread, up to the last twenty-four messages, to Anthropic, along with a token that lets the model call back into your own backend. It can WRITE a journal entry, create a task, change a task's status and defer one. It can only READ your day, your areas, your streams and your documents. It can also search the web and fetch pages, through tools that run on Anthropic’s side. Chat has no switch in Settings → AI or anywhere else: the master switch does not stop it and neither does any per-job switch, and removing your Anthropic key in Settings → Keys is the only thing that does, after which it refuses instead of answering. It runs on claude-opus-5 rather than the Opus or Haiku the rows above offer, set on the backend by LIFEOPS_ASSISTANT_MODEL, and the words it is sent with are written into the code. You cannot change either from the app. The user’s guide’s Chat page covers using it.

Three more jobs are planned and not built: document summaries, task status lines, and travel time for calendar events. They are tracked as issues #261 to #263, and each one appears in Settings as its own switch in the release that builds it.

One thing the switches do not govern. Entries tagged Agent in your journal come from a different actor, your Claude, writing on your behalf. The user's guide's journal page covers them.

What it seesWhere the text goes once it leaves

Area naming sends the entry's text and your life areas with their descriptions, and gets back one area name, or the word NONE. Document filing sends a document's name, kind, description and text with that same list of areas, and task filing sends a task's words, its kind and its note. All three get back the same one-word answer. Transcribing a recording sends the audio to ElevenLabs and then the transcript it returns to Anthropic, and the version line sends the two versions' text. Chat sends more than any of them: your question, the recent thread, and whatever its tools then read out of your own data, which is your areas, your journal, your day, your tasks, your streams and your documents. Your photographs never leave, Chat included. Each switch in Settings states what its job sends, exactly, and Chat says what it reaches on its own guide page.

Turning it offYour switches in Settings, and the key underneath them

Your switch is under Settings → AI: a master switch, and under it a switch per job. Flipping one takes effect on the next sweep, within a couple of minutes, with nothing to restart.

Underneath the switches is the key everything runs on, and it is yours, in the app. Settings → Keys holds one slot per outside service: Anthropic, ElevenLabs and Resend. A key is stored once and never shown again, and saving an empty field removes it. The five jobs and Chat all run on the Anthropic key stored there and bill it, so removing that key stops all six, and it is the only thing that stops Chat.

One thing on the AI screen is stale. It hides its switches and says AI is off for this whole instance when the backend has no ANTHROPIC_API_KEY environment variable, which was true before keys moved into the app and is the one check that still reads that variable. The jobs themselves read the stored key, so an instance whose environment variable was removed keeps running every job on the stored key with the switches hidden. Your Claude can tell you which state yours is in.

Turning AI back on reaches back a little: the sweep classifies entries from the last two days that never got a verdict, so recent writing gets filed. Anything older stays as you left it.

With it offThe app keeps working, with less filled in for you

Every control still does what it says, and every screen still finishes loading.

• Entries, documents and tasks save immediately with no area filled in. You can set an area yourself, or leave it blank.
• A task under a stream still takes its stream's area. That half is arithmetic on your own data, not a model call, so it keeps working.
• A recording still uploads, plays and is kept. It gets no transcript, no title and no summary, and the audio is not sent anywhere.
• A document or a stream still seals as a new version. The line saying what changed is not written.
• Chat still answers, and still sends your question and its reading of your data to Anthropic. The master switch does not reach it. Removing the Anthropic key in Settings → Keys is what stops it.

Everything you type is still stored, searched, and shown exactly as before.