Every version.

Your bots, also for your other AI tools.

• Connect an AI agent: AI tools like Claude Code or Codex can now talk to your bots. Settings → Devices → Connect an AI agent shows what to paste into the tool; on a server, hesperan-bots-server pair --agent. You allow it once by comparing a number, as with your iPhone, and remove it there any time.
• An agent never decides for you: it can send your bots messages, read their chats and see what waits for your OK. It can't approve or decline anything, change your bots or settings, or open your files. Its messages say "via" the agent in the chat and in Activity, and what your bots want to keep in memory from them waits for you. Agents connect only at home or over Tailscale.
• On a server, the Away from home status now says "this server" instead of "this Mac".

A new first start, a tour, and an easier server.

• A new first start: a short start screen under the evening star, then setting up your team as one scene: your name, a few questions, and your bots take shape as you answer. On the iPhone the start screen plays on the first launch.
• A tour of the app: your main bot shows you around, step by step. Skip it any time with Esc, and take it again from Help → Take the Tour or Settings → About.
• Update your server with one click: Settings → Server shows a new version and installs it. On a server installed with install.sh, run sudo hesperan-bots-server update once; from then on the button does it. In a Docker folder from 1.3.2 or before, do this one update with the package from your account page: ./update.sh <file>.
• Folders on your server: choose them in the app's own picker instead of typing a path.
• Report a problem while your bots run on your server: it now works there too, with the server's log and version if you leave the diagnostic info in.
• Model downloads keep what already arrived when the connection drops.

A calmer, simpler app.

• Simpler Settings: 12 pages in a few groups instead of 19 entries. What bots may do on your computer is all under Computers, models and their costs are on one page, your iPhone and Telegram are under Devices. Search (⌘K or Ctrl+K) still finds every setting by its name.
• Fewer words everywhere: headings without long subtitles, settings as a label and a switch, explanations only behind a small (i) where they matter. Zero counts stay hidden, and every status is shown once.
• A damaged download of a bot's computer system now fails with a clear message instead of stopping its helper.

Hesperan Bots for Windows, and bots' own computers on your server.

• Hesperan Bots on Windows: the same app on a Windows PC (Intel or AMD, Windows 10 or 11). Download it on your account page and run the installer; it needs no administrator. The installer isn't signed yet, so Windows asks once: choose "More info", then "Run anyway". Updates install themselves, as on a Mac.
• Your ChatGPT or Claude plan on a PC: Hesperan Bots installs Codex or Claude Code for you in Settings → Models, and you sign in in your browser.
• PC access: share a folder and your bots' commands run in Windows' own sandbox, only in that folder and without the internet unless you allow it. Full access runs Windows PowerShell as you; "Protect keys and passwords" keeps the file tools away from your saved credentials and keys and doesn't run commands.
• Bots' own computers on a PC (Windows 11): each bot gets a small computer of its own, a Hyper-V virtual machine. Set it up once from the bot's Computer tab or Settings → Bot computers; Windows asks for your permission (and on some PCs for a restart). Removing the app removes it again.
• Bots' own computers on your server: a server with KVM (most Linux servers and home servers) now gives each bot its own computer too, like a Mac.
• On a PC, notifications have Approve and Decline buttons, the taskbar shows when a bot needs you, and dictation is Ctrl+Space. Your iPhone says "your PC" for a PC (with the next iPhone update).
• iPhone: conversations open at the newest message.

• Bots talk to each other: in a group, open Talk and turn on "Bots talk to each other". Choose how many rounds (1 to unlimited) and how much it may cost. After your message the bots discuss among themselves, address each other with @, and the group's lead sums up what was decided. Pause or stop any time, or just write in between.
• Update your server with one click: your Mac and your iPhone show when a new version is ready for your server. Click Update server and it updates on its own; if the new version doesn't start properly, your server goes back to the old one by itself, with your data as it was. Servers on 1.2.0 or older need one last manual update (./update.sh).

Commands, wherever your bots run.

• Bots on your server can run commands on your server: Settings → Server access. Share a folder on your server, then choose per bot: off, shared folders, or full access. Commands run in a sandbox: only the shared folder, no access to your server's keys or system files, and the internet only if you allow it. Every command shows on an approval card first.
• Bots on your server can use your Mac: Settings → Mac access → "Let bots on your server use this Mac". Your Mac checks every request itself and runs it in its own sandbox, only in the folders you chose. If your Mac is asleep or closed, the bot says so instead of waiting.
• Stop everything ends running commands on your server and on your Mac, and changes can be undone as before.

• The first message on a new server works: bots whose model isn't on your server (for example a plan from your Mac) now use your server's default model instead of saying the model isn't available. A plan you add becomes the default when the old one is gone.
• You see that a reply is included in your plan while the bot is still working, not only at the end of a long task.

• Your ChatGPT or Claude plan on your server, in two clicks: in Settings → Models, choose "Install on your server", then "Sign in". The sign-in opens in your Mac's browser and your server keeps it. Then pick the plan as the default or for a bot, just like on a Mac. Update your server too (./update.sh).

• Updates when the app isn't in Applications: if you open Hesperan Bots straight from the downloaded disk image, macOS runs it from a place where it can't update itself. The app now says so right away and offers Move to Applications, instead of an update button that couldn't work.

• The app speaks about your server when your bots run there: models, memory, your data, keys and settings say "your server" instead of "this Mac". "Run on your server" helps you install Ollama on your server (or use one elsewhere in your network) and suggests a model that fits your server's memory.
• Plans on a server: signing in to a ChatGPT or Claude plan from the Mac doesn't work for a server yet, so those choices say so plainly instead of failing later. Use an API key or a model on your server for now.
• Mac access and bot computers explain why they aren't available on a server, instead of showing switches that do nothing.

A calmer app, and your bots on your own server.

• Run your bots on your own server: install Hesperan Bots on an always-on Linux computer (a home server, a NAS with Docker, a Raspberry Pi 5, a rented server) and your bots keep working while your Mac is closed: replies, routines, check-ins, Telegram and your iPhone. Download it on your account page; the guide is at hesperan.com/bots/server.
• Use your server from your Mac and your iPhone: choose "I have a Hesperan Bots server" on the first start, or Settings → Server. Pair with the link your server shows and check the 4-digit number, just like pairing your iPhone. Away from home works through hesperan.com, encrypted end to end. Your Mac's own bots and data stay untouched, and you can switch back any time.
• One reply, one bubble: what a bot says while it works ("I'll look for the folder…") now sits in the reply's steps, so each answer is one clean message. Click the steps to see everything it did and said.
• A calmer chat: messages, cards and the message box share one reading column; replies of the same bot are grouped; approvals, drafts, notes and suggestions share one card style. Bot faces show again in dark mode.
• A shorter sidebar: Team, Routines, Knowledge (memory and skills) and Library. Approvals shows up when something waits for you, Projects once you have one. Activity and Extensions are now in Settings, and search (⌘K) finds every page and setting.
• Signing in your server from the Mac goes through your Mac's browser, and the sign-in screen says it's for your server.
• Every page has the same header and widths, labels look the same everywhere, light mode has clearer grey text, and "Spent today" no longer looks like "$0,00".

• New bots ask before reading your email. Only bots made for mail (Inbox Keeper, Relay, Helpdesk) read it without asking. Your existing bots keep their settings, and you can change it in a bot's profile at any time.
• The browser says what stays with you: "Use its browser" now tells you that buying, sending and deleting on a page always ask you first.
• Pairing your iPhone is more reliable: if your iPhone drops away while your Mac asks to confirm the pairing, the question now closes right away instead of waiting until the code runs out.

Early Access. This version is about your first hour with Hesperan Bots on a new Mac.

• Models on your Mac work on every Mac: "Run on this Mac" suggests a model that fits your Mac's memory, shows its size and downloads it right in the app. Bots on small local models now keep all their instructions, and errors say what's wrong in plain words (for example, not enough memory).
• Your ChatGPT plan is found again: Hesperan Bots now finds Codex inside the ChatGPT app, which is where OpenAI ships it now.
• Your Claude plan in one click: "Install Claude Code" runs Anthropic's own installer after showing you exactly what it does, and the Terminal step during sign-in is explained first.
• "Check again" always answers, and setup suggests what's already on your Mac instead of an API key.
• You see that a bot is working: "Thinking, 120 words so far" with the time, and Stop always works.
• Extensions work without installing anything else: they no longer need Node.js on your Mac. An extension only shows as added once it really runs; otherwise it says what to fix.
• Projects explain what they need: if your Mac doesn't have Apple's developer tools yet, Settings → Projects offers to install them and notices when they're ready.
• Routines say "Won't run yet" when their bot has no working model, Gmail works with an app password, and the app shows "Early Access" in Settings → About.

• Pairing your iPhone is simpler: Settings → Devices has one button, "Pair your iPhone". Choose whether you want to use your bots away from home too, scan the code, and check that your iPhone and your Mac show the same 4-digit number.
• You always see how your iPhone is connected: "At home" when it talks straight to your Mac, "Away" when it goes through hesperan.com. Each paired iPhone has its own "Away from home" switch.
• Notifications on your iPhone: turn on "Notify me when a bot needs me" in the iPhone app, and your iPhone tells you when a bot waits for your OK or has finished something while you're away from your Mac. The notification only says "A bot needs you." or "A bot finished something for you to read." – never what it's about.
• Update the Hesperan Bots app on your iPhone too (from TestFlight): older iPhone versions can't pair with this version.

• Reach your Mac from anywhere: with the Hesperan Bots app on your iPhone, turn on "Reach my Mac from anywhere" in Settings → Devices, and your phone reaches your bots through hesperan.com when it isn't on your Wi-Fi. Everything stays encrypted between your phone and your Mac. It's off until you turn it on.
• Share bots as templates: "Share as template" in a bot's profile saves a file others can add to their team. Before a bot from a file is added you see everything it brings, and it never gets more than "Ask". Browse ready-made bots under Bot templates.
• Report a problem from the Help menu or Settings → About. You see exactly what will be sent before anything leaves your Mac. If Hesperan Bots quits unexpectedly, it offers to send a report next time; crash files stay on your Mac.
• Your Mac stays awake while bots work (on the power adapter, by default), so a long task or a routine isn't cut off by sleep. Change it in Settings → General.
• Long chats stay smooth, also while a reply is being written.
• The iPhone app can now decide memory changes, see the scope of "Always allow" and hand off drafts.
• The sign-in screen and Settings → Account say more precisely what we can see: never your bots, chats or memory, unless you include them in a problem report.

• Updating from 1.0.1 works again. If your Mac kept showing "Update ready" for 1.0.2 after restarting, this version installs normally.
• If an update ever can't be installed, the app now says so and links to the download, instead of offering the restart again and again.

• The memory search model now downloads from hesperan.com, so nothing goes to another company when the app starts.
• When a bot writes to other people for you, it says that it's an AI assistant acting for you.
• Settings → About lists the licences of everything inside the app, and links to the privacy policy and terms.
• Routines that watch a folder no longer miss a file that arrives right after the app starts.

• The bot computer now downloads from hesperan.com with your Hesperan account, so every Mac can give its bots a computer of their own.
• Settings → About links to the licences and source code of the programs that run the bots' computers.

The first release of Hesperan Bots for the Mac.

• A team of AI bots on your Mac: chat with them, give them routines, and let them hand work to each other.
• They think with your ChatGPT or Claude plan, an API key or a model on your Mac.
• Each bot has its own browser and, if you want, its own computer; sending, paying and deleting wait for your approval.
• Sign in once with your Hesperan account. Updates come from hesperan.com, and your bots, chats and memory stay on your Mac.