Chats and AI agents

Kaizendex's /chat surface is more than a single feed. Your conversations live in the left sidebar, in two sections — Log views (your log, plus any view you pin) and Chats (the AI agents you can talk to). The one you pick fills the middle of the screen, and opening an agent replaces the sidebar list with that agent's conversations.

The sidebar

Blue reply dots and orange input-needed dots clear when you open that chat. A new reply can show a dot again. Opening a chat does not answer its questions or approve its plans, and a working indicator stays visible while the AI runs.

The circular arrow beside a closed chat reopens it in the log view. It does not refresh messages or ask the AI to reply again.

When an agent's picture is shown full width, the name on it is white by default. In your theme JSON, set overrides["agent-portrait.foreground"] to change its color; use overrides["agent-portrait.background"] for the fade behind it.

When you open a conversation link on a new browser, the first data download keeps the conversation you selected and any message anchor in the link.

Down the left edge of the sidebar is a narrow strip of icons, the rail. Search sits on top — click it (or press ⌘K / Ctrl+K) to search everything you have. Under it are the places you go: Chats, Dashboard, Calendar, Collections, Kaizendex Library, Automations and AI, with Inbox at the bottom. Point at an icon to see its name. Beside the rail is the home panel, the same on every page: Log views, Chats, Favorites and Folders. Click a section's name to fold or unfold it. Only Calendar swaps in its own panel (the mini calendar and your calendars); every other page keeps its tabs and filters in a bar at its own top.

On a computer, click the app logo at the top-left to fold the sidebar into the round logo. Click the logo to open navigation again; your page stays in place. You can keep the rail visible instead, and choose which icons and sections show, in Settings → Appearance → Sidebar.

Log views and Chats sections

Each section header has a + button:

• + next to Log views pins a view here. Search for any Collection or global collection and pick one of its views — feeds and lists only, since a table doesn't belong in a chat. The same dialog can make a brand-new global collection if you want one first. A pinned view that covers a single Collection logs straight into it when you type.
• + next to Chats opens the agent gallery so you can add a new AI agent. An agent you've already added shows "✓ added" — you can't end up with two chats for the same one.

An agent's row shows a small coloured dot when one of its chats wants you — it's thinking, or it asked you something and is waiting.

You can also drag rows to reorder them within each section, and hover a row to reveal a ⋯ menu in its top-right corner: change that chat's background, edit an agent's picture and persona, or edit/delete a view. On a computer, use the view name’s ⋯ menu to open View settings, including filters, sorting and layout, in the full-height panel on the right. On your phone, there's no hover or drag — long-press a row instead, and use Move up / Move down to reorder it.

In the browser or desktop app, choose Open full view from a log view's ⋯ menu to open its full collection page with that same view selected. This also works from the chat header and in the browser on your phone.

The last row under Chats is All chats, every agent — see Every chat in one place below.

The agent's conversations

On the web and desktop app, opening an agent puts its picture and conversations in the left sidebar, in place of the home panel. Along the top runs a row of small pictures: the chats you used last — agents are round, log views are rounded squares. Click one to switch straight to it; an orange dot means that agent needs you. When they don’t all fit, the last bubble is ⋯ — click it for every agent and log view, with search. The ← arrow at the start of the row takes you back to your log views and agents.

The agent's picture shows as a circle with its name below. Click it to see the picture full width, and again to go back. The ⋯ in the picture's corner holds agent options.

• Global chat shows every conversation with this agent.
• Open in log view lists chats currently available in a log’s chat dropdown, even from previous days.
• All chats contains the other conversations, including ones you closed from the log dock.
• + beside All chats starts a new conversation. A new conversation starts out named by its date and number — 09-21 chat 1, 09-21 chat 2, and so on — and the AI gives it a proper name once its first reply is done.

Click a row to read and reply to that conversation. Use Find a chat to filter by name. Both groups collapse, and the list scrolls without moving the agent picture. Collapsed groups still show when a chat is working or needs your input.

A spinning indicator means working, an orange dot means needs your input, and a blue dot means completed. Empty chats have no dot. These colours can be changed in your theme JSON.

The log chat dock

The dock header starts with the agent's picture — click it to talk to a different agent — then your open conversations, + for a new chat, and a rightmost Hide chat button. On a computer the open conversations are tabs: click one to switch, hover it for an × that closes it, and when there are more tabs than fit, a +N button lists the rest. On a phone the dock shows one conversation at a time: tap its name to pick another (long lists offer search), and the × right beside the name closes it. Hover or keyboard-focus the icons to see their tooltips.

+ starts a chat. × closes the selected chat without deleting its messages or stopping its work; it moves to All chats in the agent sidebar. Its reopen button returns it to the log. Hide chat hides the transcript while keeping the message box and your draft. Show chat brings it back.

Your message goes to the selected conversation. The attachment control is on the left; the microphone is on the right when the draft is empty, replaced by Send when you type or attach a file. The input grows with text; a longer draft also has an expand/collapse control.

Every chat in one place

All chats, every agent — the last row under Chats — shows every conversation you've had with anyone, oldest first, each one introduced by a line with that agent's picture. Click a line to jump into that conversation on the agent's own page.

It's for reading, not writing: since there's no such thing as sending a message to everyone, the box at the bottom asks you to pick an agent first, and picking one takes you to their page.

The chats

Your log — what you already know. The activity feed you've been using all along: every entry you logged, in chronological order. It's simply the default pinned view, so it always exists and can't be removed. Logging works exactly the same — type a Collection name into the box at the bottom and it lands at the top. That box is also where you talk to your AI, and it's the same conversation as the one on the agent's own page. See Logging by chat.

The Navy Seal — a no-nonsense accountability voice, modelled on the energy of David Goggins. Open this chat when you need a kick. Type anything; you'll get a short, direct reply.

Coach Jack — a calmer, aspirational life-coach voice, modelled on Robin Sharma and The 5 AM Club. Open this chat when you want gentle encouragement and morning-discipline framings.

Each agent's voice isn't fixed — you can rewrite its personality yourself under AI → Soul.

And these two coaches are just the start: under AI → Agents you can add more agents from a marketplace, give each one its own picture and chat background, rename them, or remove the ones you don't use. See Managing your AI agents.

How it works

Click a row in the sidebar and the main view in the middle swaps to that conversation. Either way there's one box at the bottom: under your log it doubles as the Collection search you already use, and on an agent's page it's a plain message box. Press Enter to send. Both boxes reach the same AI and the same conversation — they're two doors into one chat, not two chats.

While the agent "thinks," a small typing indicator appears below their avatar. The reply lands a moment later. Messages stick around across reloads and across devices — they're stored the same way your entries are.

Where a conversation starts and stops

An agent doesn't carry every message you've ever exchanged into every reply — and now the chat tells you where it drew the line instead of quietly forgetting.

A chat is one long scroll with a labelled line across it wherever the agent's memory restarts, saying why:

• "New chat · you tapped New chat" — you asked for a fresh start. New chat in the agent's panel (and Start fresh under your log) draws this line right where you are: nothing is deleted, nothing disappears, and you keep typing in the same place.
• "New chat started · you were away 4 hours" — you came back after a long break, so the next thing you send starts a new conversation rather than replying to yesterday's.

Everything above a line stays on screen forever. A line only changes what the agent reads, never what you can see.

Picking an old conversation back up

Every line has Continue this ↩. Tap it and a new part of the chat starts at the bottom, carrying the older conversation's memory with it. Its line reads "Continued from: …" and shows a small Remembers: note — a short summary of what's being carried across. Tap that note to edit it: whatever it says is exactly what the agent is told, so you can correct it or narrow it down ("just the injury, not the whole plan").

If the older conversation is too long to replay word-for-word, the note says so — you'll never be shown "remembered in full" when only the summary made it across.

Finding an old conversation

The magnifying glass at the top of the chat opens your past conversations — the same conversations available under All chats, with search on top. It opens on the list itself — every conversation with that agent, newest first, with its title, its date and a line about what it was about. You don't have to type anything to see them.

Typing narrows that same list: titles, summaries and message text, filtered as you type. Flip on All chats to look across every agent instead.

Click a conversation to read it. You get the whole thing back, exactly as it happened, with a Remembers: note at the bottom showing what would carry across if you picked it up — including whether it would ride in full or as a summary. Continue this ↩ is right there, and Back returns to the list having changed nothing. Every row in the list also has its own Continue button for a conversation you already recognise.

Chat with several agents at once

You don't have to wait for a reply. Ask one agent a question, switch to another chat and ask something else, then go check your calendar — every agent keeps working in the background, and each reply lands in its chat whether or not you're looking at it.

While agents are working, the app keeps you posted:

• In the sidebar, an agent mid-thought shows a pulsing dot on its row, and a spinning indicator marks that conversation in the agent sidebar. Open it and you'll see the reply being written live.
• A reply that arrived while you were elsewhere shows an unread dot on its row, and a small toast pops up anywhere in the app ("Benjamin Franklin replied") with a View button that jumps straight there.
• A pulsing dot at the top of the sidebar, beside the Kaizendex name, means at least one agent is thinking somewhere — click it to jump to that chat.

The Stop button (it replaces Send while an agent is thinking) genuinely stops the agent — the reply is discarded, even if you've navigated away in the meantime.

When a turn doesn't finish cleanly

If a turn stops early — you pressed Stop, the connection dropped, or the agent ran out of time — the chat tells you per action what happened rather than guessing. Under the warning line you'll see one line for each change the agent had started:

• "… is unconfirmed" — the agent began this change but couldn't confirm it landed. It may or may not have applied: check the Collection before asking again, so you don't end up with a duplicate.
• "… failed: …" — this change did not happen, with the reason.

A reply interrupted by a provider error stays marked as failed, even if the connection then closes normally. Work that already completed keeps its own result.

Changes that completed normally aren't listed; their pills show under the message as usual. Every change also keeps its own time limit, so a slow website or a stuck tool can't hold the whole reply hostage — the agent stops waiting, tells you what it couldn't confirm, and answers with what it has.

Watching an agent work

While an agent is on a longer job, its chat bubble narrates what's happening instead of just showing dots:

• A status line tells you what the agent is doing right now — "Reading a skill…", "Logging Entry…", "Thinking…" — with a timer showing how long the turn has been running.
• Tap Thinking to watch the agent's actual reasoning stream in live, like reading over its shoulder. It's there while the agent works and disappears once the reply is written.
• After a few seconds the bubble reminds you that it's safe to leave — the agent keeps working either way. Tap Notify me when done to get a notification when the reply is ready: if the tab is just in the background you'll get a heads-up right away, and if you closed the app entirely the reply arrives as a push notification (and in your Inbox) that jumps straight to the chat.

Longer jobs are fine too. An agent turn can run for several minutes when the work needs it — deep web research, a many-step request — and it runs in the cloud, not in your browser tab. If your phone locks, your signal drops, or you switch apps mid-turn, nothing is lost: when you come back, the chat rejoins the running turn and catches up on what you missed. When time does run out, the agent wraps up with what it found and tells you what's left — say "continue" to let it pick up from there. If a turn fails outright, the chat says so with a notice that stays until you close it.

On mobile

Tapping Chats in the bottom bar opens your log — it's your home base on the phone, the same screen you land on when you open the app. The bottom menu stays put while you're in any chat, so the rest of the app is always one tap away. To see all your chats (your log plus the AI agents), tap the back arrow at the top of the chat; from that list, tap any row to open it. The agent panel is a desktop thing — a phone screen has no room beside the conversation for it.

What they can do

The agents are real now — they think, act, and remember:

• Take action on your data. Ask one to set up a tracker, log something, or summarise your week and it does it — the new Collections and entries appear across your devices, the same as if you'd tapped them in yourself. Deletions go to Trash, so nothing is ever lost.
• Show you what it made. Every action leaves a small pill under the reply, naming the thing with its own emoji — "✅ Logged Work for 7 hours productively", "✅ Created Tasks". When that thing still exists, the pill is a link with a ↗ arrow: click it and the entry or Collection opens. Open one from the assistant sidebar and the chat stays right where it is, so you can read the entry and keep talking. (Pills for deletions, failures, and actions you've undone stay plain — there's nothing left to open.)
• Draw you a picture. Ask for an illustration and — if image generation is turned on — it appears right in the chat. See Image generation below.
• Remember you. As you talk, the agent quietly remembers durable facts about you — your goals and preferences — and uses them next time. You stay in control: see, edit, or forget any of them in AI → Facts.
• Follow your skills. Teach an agent a repeatable way of working under AI → Skills.
• Look back over your chats. Past conversations are part of its memory too — an agent can search and summarise what you've talked about before. It searches by topic, not just exact words, so it can resurface a relevant earlier chat even if you phrase it differently this time. Your conversations are listed under AI → Conversations, each one named automatically.

If a reply ever stalls, it times out and tells you rather than hanging — just send again.

Big changes come with a plan to approve

For anything large or hard to undo — moving a whole Collection's entries somewhere else, or cleaning up a big batch at once — your AI doesn't just do it. It first shows you a plan: a card right in the chat with a short title, the numbered steps it's about to take, and any warnings about what could be lost.

The card has two buttons:

• "Yes, run this plan" — the AI goes ahead exactly as described.
• "No — let me explain" — nothing happens; the message box focuses so you can say what you'd rather it do.

You're never limited to the buttons — you can just reply in your own words, and your AI reads that instead. Once you've answered (or moved on to something else), the card settles into a quiet record of what was decided.

Answer with a tap

Plans aren't the only place your AI offers buttons. When it asks a quick, clarifying question — "Which one, Running or Gym?" — it can show a row of small pills under its message instead of making you type. Tap one and it's sent as your reply, just as if you'd typed it yourself.

The same message can also include a link button — a shortcut that jumps straight to a page in the app, like "See your agents" or "Open your new chat." Link buttons stay tappable forever; quick-reply pills quiet down (but stay visible) once you've answered.

You're never limited to the buttons here either — typing your own answer always works.

Merging two Collections

One thing your AI can do this way is merge Collections — say, folding a stray "Todo" into your main "Todos." Ask it to combine the two and it proposes a plan: move every entry across, then remove the now-empty duplicate. If some of the fields on the old Collection don't exist on the one you're keeping, the plan warns you that those values will be dropped, so you know exactly what carries over before you approve.

Your AI also tries to stop the duplicate from happening in the first place — if you ask it to create a Collection that's basically the same as one you already have (like "Todo" when "Todos" exists), it'll point you at the existing one instead of making a near-copy.

The AI page

Everything about your AI agents lives on the AI page — click AI in the sidebar's rail. Its tabs sit in a bar at the top of the page:

• Soul, Facts and Skills (Soul is the default) — your AI's personality, the facts it knows about you, and the skills you've taught it. See What your AI remembers.
• Agents — your list of AI agents, with Browse agents to add more and New agent to build your own from scratch. See Managing your AI agents and Building your own AI agent.
• Actions — a record of everything an agent did to your data, where you can undo any single change.
• Conversations — every past conversation, which you can revisit or delete. They're named for you automatically, so the list is easy to scan.
• Slash commands — your / shortcuts for the log. See Logging by chat.

Which models your AI runs on is a separate screen, in Settings → AI settings — covered below. Where your credits actually went is another one, in Settings → AI usage — see Where your AI credits go below.

Choosing your AI models

Under Settings → AI settings you pick one of three options:

• Balanced (the default) — requests like "ran 5 km" or "save this link" go to a fast model, so they land in a few seconds. Questions and photos go to a smarter model.
• Max — the deepest reasoning and the best writer, at a premium price.
• Custom — pick your own models.

How your AI picks a model

Every message goes to one of two models:

• Quick-log model — anything you ask it to do: "ate 2 eggs", "save this link and tag it", or a long instruction.
• Chat model — questions (a "?", or words like "how", "why", "explain", "research"), photos and files.

Each AI reply shows the model that wrote it, next to its time.

Custom

Choose Custom to pick the Chat model and the Quick-log model yourself. Set the quick-log model to Same as chat model to use one model for everything. Under Advanced you can also pick the models for background work (chat titles, summaries, memory) and for sorting tracked apps into categories.

Every preset can read your photo uploads and act on your data. Document uploads (PDFs and other files) work on Max; Balanced's chat model doesn't accept file uploads yet — use photos there, or pick a model that reads files under Custom.

Custom models

Prefer to choose exactly? Pick Custom to open a searchable catalog of dozens of models — from Anthropic, OpenAI, Google, Alibaba, Moonshot, MiniMax, Z.ai, and more — with two slots: the chat model for your conversations, and the background model for the busywork. Every listed model reads image uploads and supports agent actions (most also accept file uploads), and our suggestions are flagged recommended at the top.

Image generation

Also under Settings → AI settings, a separate Image generation section controls whether your AI can draw pictures at all:

• A master on/off switch. Turn it off and your AI simply won't offer to generate images — no picture requests, no prompts about it.
• Two model pickers, each showing the price per image: one for chat pictures (illustrations that appear in a conversation), and one for avatars & backgrounds (agent pictures and chat wallpapers, including ones you generate while building your own agent).

Every generated image is automatically resized before it's saved, so icons and avatars stay small and backgrounds stay a sensible size — nothing bloats your storage.

Where your AI credits go

If an answer reaches its response length limit, the chat marks it as incomplete and keeps the text already written. You can ask it to continue from where it stopped. This does not automatically start another paid response or raise your chosen budget.

Requests for explanation or comparison use your chosen chat model, including a short message such as “These sources disagree. Explain the difference.” Simple activity logs still use your logging model.

Open Settings → AI usage — it sits right under Plan & billing, and your credit balance links straight to it ("See where your credits went"). It's your whole AI spend, drawn two ways:

• By model — which model burned the tokens, with your running totals underneath.
• By purpose — chats, quick logging, and the background AI that sorts your screen-time activity.

Switch between Monthly (one bar per month, your whole history) and Daily (one bar per day, with arrows to page back through 30-day windows).

Your credit balance

Settings → Plan & billing shows one number: the credits you have left, and roughly how many messages that buys. Under it, a quieter line splits that total into the two kinds of credit, because they behave differently:

• The credits included with your plan land every month and roll over for a while — but they don't pile up forever, so if some of yours are about to lapse the screen tells you how many and on what date.
• Everything else is yours to keep: it never expires, and changing plan never takes it away.

You'll never see a dollar figure for a balance. Credits are the unit, with the message estimate beside them.