What's new in Houston

Houston 1.0.15

What's new

• Security and performance updates.

Before you upgrade

• Quit Houston before installing. macOS doesn't always replace a running app. If in doubt, delete Houston from Applications and drag the new copy in.

Houston 1.0.14

What's new

• Each card on the board now shows when you last talked with that AI employee: the time if it was today, then "Yesterday", the day of the week, or the date. Hover it for the full date and time. The same time shows in the chat header and in the task list on your phone.
• On narrow boards, like with the chat open, the time on each card always shows in full. To make room, rename and delete move into a "More actions" button on the card.

Before you upgrade

• Quit Houston before installing. macOS doesn't always replace a running app. If in doubt, delete Houston from Applications and drag the new copy in.

Houston 1.0.13

What's new

• GPT-6.1 Terra is available when you connect your ChatGPT subscription.
• If you made your account with an email code, you can now also sign in with Google, Apple or Microsoft. Houston asks for a code sent to your email once, then links them.
• On yearly plans, the person who pays sees the price of a new seat before adding someone. The seat is charged right away, for the time left until renewal.
• On the web, Houston opens the space you used last, on any device.
• A task that runs for more than 4 hours now wraps up and keeps its work.

Before you upgrade

• Quit Houston before installing. macOS doesn't always replace a running app. If in doubt, delete Houston from Applications and drag the new copy in.

Houston 1.0.12

What's new

• Security and performance fixes.

Before you upgrade

• Quit Houston before installing. macOS doesn't always replace a running app. If in doubt, delete Houston from Applications and drag the new copy in.

Houston 1.0.11

What's new

• New team pricing. Each person in a paid organization costs $20 a month, or $15 a month billed yearly ($180 a year, save 25%). You choose monthly or yearly when you upgrade. Every account still has one organization on the free tier.
• Billing in every organization. Each organization has its own Billing tab. Owners upgrade from there. Members who need more see who to ask.
• Confirm or change your free-tier pick. After the switch, owners can check which organization stayed on the free tier and which one their subscription pays for, and change it for 30 days.
• Clear status for organizations that need a plan. An organization that isn't on the free tier and has no plan keeps working for 30 days, with a note that says when it locks. After that, you can still open it, read every conversation and move your AI Employees to another organization. New conversations wait until someone upgrades, and the owner can unlock it in one click.
• Back to Houston after paying. When you finish paying in the browser, Houston opens on your organization's Billing by itself.
• A heads-up for promotional prices. If your organization pays a promotional $15 a month, Houston tells you when it changes to $20 and lets you keep $15 a month by switching to yearly.
• See which routines stop before removing someone. Removing a member from a team now lists the routines that would stop running.

Fixes and polish

• Claude no longer shows "Anthropic is having a problem" on very long conversations.
• The org chart places each AI Employee under the person who created it.
• When an app has several connected accounts without names, Houston asks which one to use.
• Clearer Spanish and Portuguese wording in billing.
• Security updates and more reliability improvements under the hood.

Before you upgrade

• Quit Houston before installing. macOS doesn't always replace a running app. If in doubt, delete Houston from Applications and drag the new copy in.

Houston 1.0.10

What's new

• Teams are out of beta. From Friday, October 23, personal and team spaces work the same way, so you can invite people to any space. A short message explains the change when you open Houston.
• Pick your free organization. Each account keeps one free organization. If you own more than one, a short picker lets you choose which one stays free. If you have a subscription, you also pick which organization it pays for. You can change both later in Settings > Billing.

Fixes and polish

• Security updates and more reliability improvements under the hood.

Before you upgrade

• Quit Houston before installing. macOS doesn't always replace a running app. If in doubt, delete Houston from Applications and drag the new copy in.

Houston 1.0.9

What's new

• Post your org chart to LinkedIn. In Admin > Org chart, Share now posts the chart image straight to your LinkedIn profile. Connect LinkedIn once, edit the post if you like, and Houston opens it in your browser when it's live.
• Slack connects inside the desktop app. Connecting Slack on desktop now finishes in Houston, with no browser prompt to return to the app.
• Attach files when answering a question. When an AI Employee asks you something, you can attach files to your answer.

Fixes and polish

• If you leave Houston open for days, a downloaded update now installs when you come back.
• An update no longer downloads twice.
• A routine that was deleted no longer shows an error.
• When your Google Cloud project can't use a Vertex model, Houston now says so instead of showing a generic error.
• More reliability improvements under the hood.

Before you upgrade

• Quit Houston before installing. macOS doesn't always replace a running app. If in doubt, delete Houston from Applications and drag the new copy in.

Houston 1.0.8

What's new

• A morning briefing from Houston. Once a day, at the time you choose (8:00 by default), Houston checks on your team and writes a short briefing in its chat: what needs your attention, what got done, and what to do next. Set the time or turn it off in Settings > Morning briefing.
• Org chart. Admin > Org chart now opens on a chart of your company, with your people and AI Employees. Share turns it into an image and a ready-to-post LinkedIn message.
• Clearer Channels settings. Settings > Channels shows a connected Slack workspace as a card with its logo and a green Connected badge.

Fixes and polish

• Houston can now open files you attach in its chat.
• When your workspace's allowed models hide every AI you connected, the model picker says so instead of asking you to connect one.
• Admin > Billing no longer appears on teams where billing isn't available.
• A message sent right as a chat opens no longer disappears from the screen.
• Sending right after the app wakes up no longer waits 15 seconds.
• A local model your workspace doesn't allow no longer shows an error every time the app starts.
• Routines and chats start faster for AI Employees with lots of files or long histories.
• Spaces with many AI Employees load without errors.
• Frosted panels blur properly in dark mode on Windows.
• The Slack "already connected" message now covers an account connected to another of your spaces.
• More reliability improvements under the hood.

Before you upgrade

• Quit Houston before installing. macOS doesn't always replace a running app. If in doubt, delete Houston from Applications and drag the new copy in.

Linear — cloud-v1.0.8

(no project)

• PRODUCT-2114 — Sponsored domains: free Plus for the first 20 verified emails per company domain
• PRODUCT-2027 — Mobile apps: release pipeline and store listings
• PRODUCT-2026 — Mobile apps: over-the-air updates
• PRODUCT-2025 — Mobile apps: push notifications
• PRODUCT-2024 — Mobile apps: store-safe payments
• PRODUCT-2023 — Mobile apps: native sign-in (Google, Apple, Microsoft)
• PRODUCT-2022 — Mobile apps: Capacitor shell for iOS + Android

WhatsApp draft

🚀 Nueva versión de Houston!
✅ Sponsored domains: free Plus for the first 20 verified emails per company domain
✅ Mobile apps: release pipeline and store listings
✅ Mobile apps: over-the-air updates
✅ Mobile apps: push notifications
✅ Mobile apps: store-safe payments
✅ Mobile apps: native sign-in (Google, Apple, Microsoft)
✅ Mobile apps: Capacitor shell for iOS + Android
Gracias a todos los que reportaron 🙌

Houston 1.0.7

What's new

• Routines pause when your AI plan hits its usage limit. Instead of failing every few minutes, a routine waits until your plan resets and shows "Paused until" with the time. To keep it running now, pick another model.
• Edits on your board show up instantly. Creating, moving, renaming and deleting cards no longer wait for the server, and a change that can't be saved is undone with a short note.
• Starting an AI Employee's first day without an AI connected now shows a "Connect AI" button instead of an error.
• Routine triggers that can't fire tell you why, and what to do: pick a different event, or check the routine's settings.
• Only the person a question or approval is for can answer it in a shared chat.

Fixes and polish

• App updates continue where they left off after a dropped connection, instead of starting over.
• A message that can't start because the AI Employee holds too much saved now says so, instead of spinning forever.
• A chat no longer gets stuck when a reply was cut short; you're told it couldn't be confirmed.
• An upload cut off mid-way now says so, instead of "connection lost".
• If your GitHub Copilot billing is locked, Houston tells you, and offers to pick another AI.
• Bug reports are delivered even when our issue tracker is full.
• Saving a file to a folder that no longer exists asks you to pick another folder.
• Waking your computer no longer briefly shows sign-in errors.
• The move-to-cloud screen no longer comes back after you sign out and in again.
• Additional reliability improvements under the hood.

Before you upgrade

• Quit Houston before installing. macOS doesn't always replace a running app. If in doubt, delete Houston from Applications and drag the new copy in.

Linear — cloud-v1.0.7

(no project)

• PRODUCT-2046 — Only a conversation's owner can answer its approval cards
• PRODUCT-2033 — Split the staging gateway's GCP identity from prod

Performance

• PRODUCT-2045 — E2B template after pi 1.1 bump boots slower: create 0.41→0.6-0.9 s, first worker health 2.0→3-4 s (prod at next engine-pod release)
• PRODUCT-2044 — Prefetch times out on agents with huge stores (65k objects): dispatch 32-54 s; dispatch metrics need a kind label

WhatsApp draft

🚀 Nueva versión de Houston!
✅ Only a conversation's owner can answer its approval cards
✅ E2B template after pi 1.1 bump boots slower: create 0.41→0.6-0.9 s, first worker health 2.0→3-4 s (prod at next engine-pod release)
✅ Prefetch times out on agents with huge stores (65k objects): dispatch 32-54 s; dispatch metrics need a kind label
✅ Split the staging gateway's GCP identity from prod
Gracias a todos los que reportaron 🙌

Houston 1.0.6

What's new

• Mistral Large 4 is now the default Mistral model.
• Routine history tells you when a routine can't run as its owner. If the person a routine runs as can no longer use the AI Employee, the failed run says so and tells you the fix: pause the routine and resume it.
• Only a routine's creator can create or replace its webhook key. Everyone else still sees whether the routine has a webhook, along with who can change it.

Fixes and polish

• Pressing Stop no longer leaves an AI Employee stuck as busy.
• Skills shared with your team now load in every chat with an AI Employee running in the cloud.
• If your team's allowed models name an older Claude model, members can pick Claude again instead of being sent to OpenRouter.
• Reloading the page while Houston saves your first conversation no longer leaves your AI Manager's chat empty.
• A routine always runs as the person who last saved it.
• Additional reliability improvements under the hood.

Before you upgrade

• Quit Houston before installing. macOS doesn't always replace a running app. If in doubt, delete Houston from Applications and drag the new copy in.

Linear — cloud-v1.0.6

(no project)

• PRODUCT-2050 — Show refused routine fires in run history with a typed reason
• PRODUCT-2049 — Webhook chip: hide create/rotate for people who didn't create the routine
• PRODUCT-2048 — Control plane: bound transcript decode memory and huge-exponent numbers on read
• PRODUCT-2047 — E2B turn sync must not accept routine edits written from the shell
• PRODUCT-2036 — Trigger bindings must not move across orgs on a routine id collision

WhatsApp draft

🚀 Nueva versión de Houston!
✅ Show refused routine fires in run history with a typed reason
✅ Webhook chip: hide create/rotate for people who didn't create the routine
✅ Control plane: bound transcript decode memory and huge-exponent numbers on read
✅ E2B turn sync must not accept routine edits written from the shell
✅ Trigger bindings must not move across orgs on a routine id collision
Gracias a todos los que reportaron 🙌

Houston 1.0.5

What's new

• Your AI Employee tells you what it's about to do. Before it starts working with your apps or files, it writes one short line, so you see a reply start within seconds instead of waiting while it works.
• Claude Haiku 5.5 is available in the model picker.
• "Every N minutes" routines run exactly N minutes apart. A routine set to every 16 minutes now runs every 16 minutes all day, instead of starting over at the top of each hour.

Fixes and polish

• Messages to AI Employees with many files are accepted faster.
• Routines in the cloud keep what they remember between runs, so they no longer repeat work they already did.
• A routine deleted while it was running no longer stays "running" forever.
• The notice that your session is ending now spans the whole window.
• Additional reliability improvements under the hood.

Before you upgrade

• Quit Houston before installing. macOS doesn't always replace a running app. If in doubt, delete Houston from Applications and drag the new copy in.

Linear — cloud-v1.0.5

Performance

• PRODUCT-2010 — AI Manager history read on a missing conversation wakes its pod (~5.7 s 404)

WhatsApp draft

🚀 Nueva versión de Houston!
✅ AI Manager history read on a missing conversation wakes its pod (~5.7 s 404)
Gracias a todos los que reportaron 🙌

Houston 1.0.4

Fixes and polish

• The first reply in a new chat finishes sooner. For AI Employees running in the cloud, Houston now names the conversation while the reply is written, not after it. That saves about 3 seconds.
• Replies start faster in chats with many shared images. Your AI Employee no longer waits for every earlier image to load before it starts answering.
• The Houston icon shows in your browser tab on every Houston site, in white on dark tabs.
• Additional reliability improvements under the hood.

Before you upgrade

• Quit Houston before installing. macOS doesn't always replace a running app. If in doubt, delete Houston from Applications and drag the new copy in.

Houston 1.0.3

What's new

• Use your AI Employees from other apps. Settings has API keys again, and each AI Employee has an API access page with its ID. Houston can also create a key for you right in the chat.
• A heads-up before Claude signs you out. Claude's sign-in ends about every four weeks. Houston now warns you a few days before and shows where to reconnect, so your routines keep running.
• New AI Employees are ready in seconds. Their first day shows up the moment you create them, and you can start it right away.
• Empty folders show in the sidebar, and it's clear where each folder ends.

Fixes and polish

• Replies stream reliably, even when the connection catches up late.
• A message you send while your AI Employee is still answering is never lost. It waits and goes out when the reply ends.
• A reply that stalls recovers on its own, and a reply stuck repeating itself now stops.
• The effort level you pick is used on every message.
• A routine run that couldn't start says why, in your language.
• Additional reliability improvements under the hood.

Before you upgrade

• Quit Houston before installing. macOS doesn't always replace a running app. If in doubt, delete Houston from Applications and drag the new copy in.

Houston 1.0.1

A small update after 1.0.

What's new

• Your plan is always in view. On Free, the bottom of the sidebar shows how much of your week you have used, with an Upgrade button that opens Billing. On your phone it sits in More.

Fixes and polish

• An edit you make to a routine is never undone by your AI Employee saving its own change at the same moment.
• Houston stays connected while it sits in the background, instead of reconnecting every few minutes.
• AI Employees you aren't using stay asleep instead of being woken by the app in the background.
• Additional reliability improvements under the hood.

Before you upgrade

• Quit Houston before installing. macOS doesn't always replace a running app. If in doubt, delete Houston from Applications and drag the new copy in.

Houston 1.0.0

Houston 1.0 is here, and Houston is out of beta. Thank you for building it with us.

Plans, and a gift for our beta testers

• Plans start on October 1, 2026. Your plan is personal and covers you in every space.
• Free is for personal use. It has a weekly usage limit, shown as a percentage of your week. Houston lets you know as you get close, and if you reach the limit it tells you when your usage resets.
• On Free, one routine runs at a time, at most every 15 minutes. If you have several, you choose the one to keep running. The others stay saved.
• On Free, routines pause after 7 days away. When you come back, Houston asks if you want to resume them.
• Plus is USD 15 a month with our launch discount, down from USD 20. You keep going without waiting, all your routines run, and they keep running while you are away.
• Teams is coming soon, so your whole team can work on Houston together.
• A thank-you gift for our beta testers. If your account was created before October 1, 2026 and you have never had Plus, your first month of Plus is USD 10. Then it is USD 15 a month. There is no deadline.
• A new Billing screen in Settings shows your plan, your weekly usage, your routines and your invoices. Upgrade or manage your payment from there.

Meet Houston, your AI Manager

• Your AI Manager is named Houston. It builds your team, hands out the work and keeps track of it.
• New to Houston? It welcomes you in its own chat. It helps you connect your AI, asks a few questions about your work and builds a starter team with you. Then it takes your first goal and you watch it pick who does what and get started.
• AI Employees give each other missions. A Teamwork setting on each one decides who it gives missions to and who it takes them from. Every mission shows who started it, including "Started by Houston".
• Houston can do more for you: connect your apps and your AI through secure cards, rename missions and find your past chats.
• Talk to Houston from Slack. Connect it in Settings > Channels and continue the same conversation in the app.

A new home for your team

• The sidebar is your team. It lists Houston and your AI Employees, with "Connect your apps" and "Connect your AI" at the bottom. Your account menu holds your spaces, Profile, Admin, the Academy and Settings.
• Groups replace teams. Arrange your AI Employees into your own groups with drag and drop. Your arrangement is yours and never changes anyone else's.
• Admin opens on your org chart, with the hours your AI Employees worked this month, who ranks first, and who manages or uses each one.
• Pick your colors. Settings > Appearance offers Light, Dark or System, plus a palette for each mode that you see change live.
• On Mac, the title bar blends into the sidebar for a cleaner window.

AI Employees that learn your work

• Hiring starts from an industry and a role, picked from a catalog you can search, then a name and a color.
• A new AI Employee starts with a first day. It introduces itself, offers three concrete jobs, does one with you on a real case and saves it as a Skill.
• Skills read as workflows: numbered steps you edit in each AI Employee's settings.
• GPT-6 Luna is the default ChatGPT model.

Work and routines you can count on

• If Houston restarts in the middle of a task, the task picks up once on its own, and your AI Employee tells you what carried over.
• A routine that keeps failing for the same reason pauses itself after 10 tries and tells you the one fix, like reconnecting an account.
• A routine that has run for a long time keeps running, instead of failing once its chat grows too long.
• Routine history keeps every run, even when several run at the same time.
• The routine screen shows the model the routine runs.
• A routine your AI Employee sets up runs on the same AI as the chat it came from.
• Replies start sooner.

Fixes and polish

• A folder Houston cannot write to says so, instead of failing in silence.
• After Houston summarizes a long chat, you see the reply, not the summary.
• Connecting a model that runs on your own computer saves correctly.
• Signing in on a computer with no default browser tells you what to do.
• An update download retries a short hiccup on its own.
• On Windows, a missing WebView2 is named, with the fix.
• A folder name with a space in it no longer blocks your AI Employee's commands.
• Additional reliability improvements under the hood.

Before you upgrade

• Quit Houston before installing. macOS doesn't always replace a running app. If in doubt, delete Houston from Applications and drag the new copy in.
• On Free with more than one routine, Houston asks you to choose the one to keep running. The others stay saved.
• Your teams become groups with the same AI Employees. Team pages and team notes are retired.
• An AI Employee already named "Houston" keeps its name. New ones can't take it.
• Some things moved or are retired: the Agent Store and the skills marketplace leave the app, the help menu is gone (Report bug stays in Settings), and the tour is two lessons in the Academy.

Known limitations

• Teams is not available yet.
• Slack appears in Settings only where your Houston is set up for it.

Houston 0.6.24

The newest Claude models

• Sonnet 5.5 and Opus 5.5 are here. Sonnet 5.5 is the new default for everyday work; Opus 5.5 takes on deeper, longer tasks.
• Claude gets its full 1M-token memory. Long conversations keep far more context before Houston has to summarize them.
• A simpler Claude menu: Sonnet 5.5, Opus 5.5 and Fable 5.1. Agents on an older Claude model move to the newest one of the same kind.

Fixes and polish

• Additional reliability improvements under the hood.

Linear — cloud-v0.6.24

(no project)

• PRODUCT-1929 — Creating an AI Employee with a taken name overwrites (and can delete) the existing one
• PRODUCT-1928 — Tag missions started by Houston (the AI Manager)
• PRODUCT-1927 — Reserve "Houston" as an AI Employee name
• PRODUCT-1925 — Onboarding inside the AI Manager chat
• PRODUCT-1913 — Teams become personal sidebar views of AI Employees
• PRODUCT-1914 — Back up team memberships and team notes before dropping the team tables
• PRODUCT-1909 — Themes: System mode, a palette picker, and ten Omarchy palettes
• PRODUCT-1907 — Design tokens: burn down the UNTOKENIZED inventory and fix the dark-mode hierarchy
• PRODUCT-1904 — Setup mission redesign: permanent hello, no user bubble, Set up tag, ideas card, do the job first
• PRODUCT-1905 — pi 0.87.1 upgrade and GPT-6 Luna as the Codex default
• PRODUCT-1902 — Text typed into an interaction card's free-text option box is lost when switching missions 🐛
• PRODUCT-1901 — Dialog leftovers: update overlay, Academy lesson panel and AI Models detail on the shared dialog frame
• PRODUCT-1900 — Drop the native iOS app: phones use the PWA of the shared web app
• PRODUCT-1899 — es/pt wording: skill stays skill, industry is industria/indústria everywhere
• PRODUCT-1794 — Dev loop: the manager (assistant pod) can never be seeded under pnpm dev
• PRODUCT-1795 — Manager silently disappears when discovery keeps failing at launch
• PRODUCT-1793 — First-party product analytics: product_events table, gateway ingest, Slack signup ping, product dashboards
• PRODUCT-1102 — ui/engine-client store-catalog specs stale vs agentstore-client skills normalization (2 failing, not in CI)

WhatsApp draft

🚀 Nueva versión de Houston!
✅ Creating an AI Employee with a taken name overwrites (and can delete) the existing one
✅ Tag missions started by Houston (the AI Manager)
✅ Reserve "Houston" as an AI Employee name
✅ Onboarding inside the AI Manager chat
✅ Teams become personal sidebar views of AI Employees
✅ Back up team memberships and team notes before dropping the team tables
✅ Themes: System mode, a palette picker, and ten Omarchy palettes
✅ Design tokens: burn down the UNTOKENIZED inventory and fix the dark-mode hierarchy
✅ Setup mission redesign: permanent hello, no user bubble, Set up tag, ideas card, do the job first
✅ pi 0.87.1 upgrade and GPT-6 Luna as the Codex default
✅ Text typed into an interaction card's free-text option box is lost when switching missions
✅ Dialog leftovers: update overlay, Academy lesson panel and AI Models detail on the shared dialog frame
✅ Drop the native iOS app: phones use the PWA of the shared web app
✅ es/pt wording: skill stays skill, industry is industria/indústria everywhere
✅ Dev loop: the manager (assistant pod) can never be seeded under `pnpm dev`
✅ Manager silently disappears when discovery keeps failing at launch
✅ First-party product analytics: product_events table, gateway ingest, Slack signup ping, product dashboards
✅ ui/engine-client store-catalog specs stale vs agentstore-client skills normalization (2 failing, not in CI)
Gracias a todos los que reportaron 🙌

Notify reporters (then clear the Notify pending label):

• PRODUCT-1902 → TBD (Julian adds)

Houston 0.6.23

Your work is not interrupted

• An answer in progress finishes before Houston restarts, so an update never cuts a reply in half.
• A stuck turn tells you it is stuck, instead of quietly repeating itself.

Routines that run as they should

• A routine woken by an event runs as the person who set it up, with the right access.
• The routine screen stops spinning forever when a trigger is slow to answer.

Fixes and polish

• Choosing a model stays within what your agent is allowed to use.
• A model running on your own computer stays connected, even when the desktop sits idle.
• Saving or opening a downloaded file explains itself instead of showing a raw error.
• Signing in to Claude without internet says you are offline, plainly.
• Long answer options stay fully readable in a narrow chat panel, with their numbers in view.
• What you pin to a message is attached when you send it.
• Additional reliability improvements under the hood.

Linear — cloud-v0.6.23

AI Models

• PRODUCT-1453 — Investigate claude error 🐛

WhatsApp draft

🚀 Nueva versión de Houston!
✅ Investigate claude error
Gracias a todos los que reportaron 🙌

Notify reporters (then clear the Notify pending label):

• PRODUCT-1453 → ⚠️ no phone on issue

Houston 0.6.22

Updates that finish what they start

• A large update keeps going after your connection drops, resuming where it stopped instead of starting over.
• Houston checks for a new version once when it opens, so startup is quicker and quieter.

Skills and connected apps behave

• Finding and installing a skill is more dependable, with searches that end on their own and skills that no longer exist kept out of the list.
• Disconnecting an app while it is still connecting no longer looks like a failure, the row simply goes back to available.

Fixes and polish

• Houston opens reliably on Windows.
• Your saved routines keep loading, even when an odd character slips into them.
• A model key that gets rejected explains what to fix, in plain words.
• Dictation tells you what went wrong when it stops unexpectedly.
• Your usage limit clears as soon as it resets.
• A mission you just started appears on your board while the agent wakes up.
• Additional reliability improvements under the hood.

Houston 0.6.21

New integration: ManyChat

• Connect ManyChat from the curated integrations list, so your agents can work with your chat marketing right away.

Custom integrations you set up yourself

• Sign in to a custom integration with your own account when the service supports it, instead of hunting for an API key.
• Give a custom integration your own name and icon, so it looks right alongside the rest.

More room to read

• Mission chats open in a wider view, so long answers are easier to follow.

Fixes and polish

• Connecting a model on your own computer tells you the truth, with honest wording instead of an endless wait.
• The restart to update button is easier to spot.
• Very long conversations keep working, and a heavy task no longer takes the agent down with it.
• A dropped connection no longer reads as a failed answer, and an expired session recovers on its own.
• Additional reliability improvements under the hood.

Houston 0.6.20

New integration: HighLevel

• Connect HighLevel from the curated integrations list, so your agents can work with your CRM right away.

Smoother answers and a steadier board

• Replies and their follow-up suggestions arrive together, as one complete answer.
• Your task board stays accurate, so a slow refresh no longer undoes a change you just made.

Fixes and polish

• Houston reopens properly after an update on Windows.
• Moving a workspace between machines is more reliable, with large transfers no longer running out of time.
• Additional reliability improvements under the hood.

Linear — cloud-v0.6.20

(no project)

• PRODUCT-1708 — Sentry HOUSTON-APP-5D6: list_routines fails - routines.json has a bad control character in a string literal (500)
• PRODUCT-1707 — Sentry HOUSTON-APP-5D5: provider sign-out in a space with no agents 404s on managed cloud (gateway setup-runtime allowlist)

AI Models

• PRODUCT-1709 — Sentry HOUSTON-APP-5AP: view-docs boot warm 404s for 12 minutes after a rename-triggered pod wake; doc shadow stays bound to the old name

Backend

• PRODUCT-1649 — Sentry HOUSTON-APP-5AD: sandbox credential heal fails (fetch failed)

Integrations

• PRODUCT-1633 — Setup GHL integration

WhatsApp draft

🚀 Nueva versión de Houston!
✅ Sentry HOUSTON-APP-5AP: view-docs boot warm 404s for 12 minutes after a rename-triggered pod wake; doc shadow stays bound to the old name
✅ Sentry HOUSTON-APP-5D6: list_routines fails - routines.json has a bad control character in a string literal (500)
✅ Sentry HOUSTON-APP-5D5: provider sign-out in a space with no agents 404s on managed cloud (gateway setup-runtime allowlist)
✅ Sentry HOUSTON-APP-5AD: sandbox credential heal fails (fetch failed)
✅ Setup GHL integration
Gracias a todos los que reportaron 🙌

Houston 0.6.18

A new model

• Use GPT-6 Astra for your work, now available in Houston's model picker.

Start from an agent you already have

• Create a new agent by copying an existing one, right from the create dialog.

Fixes and polish

• Conversations with Gemini recover on their own, so a hiccup mid-answer no longer stops your turn with a retry card.
• Connecting your own local model explains what went wrong, with the reason shown next to the connection instead of a silent failure.
• Signing in again clears the old error, so a provider that failed before stops looking disconnected once you reconnect.
• Additional reliability improvements under the hood.

Linear — cloud-v0.6.18

(no project)

• PRODUCT-1663 — Non-admin member connecting the org AI subscription gets a 403 reported as an unexpected error instead of an expected-state toast

AI Models

• PRODUCT-1682 — Add GPT-6 Astra (gpt-6-astra) to the OpenAI Codex and Azure OpenAI catalogs

Global Interfaces & Components

• PRODUCT-1681 — Phone agents home as a chat list: latest-task rows, team selector, desktop-parity Teams tree
• PRODUCT-1678 — Add new option when creating an agent 🐛

WhatsApp draft

🚀 Nueva versión de Houston!
✅ Add GPT-6 Astra (gpt-6-astra) to the OpenAI Codex and Azure OpenAI catalogs
✅ Phone agents home as a chat list: latest-task rows, team selector, desktop-parity Teams tree
✅ Add new option when creating an agent
✅ Non-admin member connecting the org AI subscription gets a 403 reported as an unexpected error instead of an expected-state toast
Gracias a todos los que reportaron 🙌

Notify reporters (then clear the Notify pending label):

• PRODUCT-1678 → ⚠️ no phone on issue

Houston 0.6.17

Updates that wait for you

• Houston no longer restarts itself to update. A new version downloads quietly in the background, and a small "Restart to update" button appears in the corner of the window. Restart whenever you like; nothing you are doing gets interrupted.
• Your agent finishes what it is doing before it restarts, so a task in progress is never cut off by an update.
• A message sent while your agent is restarting is delivered once it is back, instead of failing.

Calmer, more honest messages

• Everyday moments no longer look like failures, so a waking agent, an expired sign-in, or a permission you do not have shows a clear notice instead of a red error.

Sign-in fixes

• Signing in with Claude on Windows now works on more setups.
• A Claude subscription stays connected on self-hosted setups.
• When your ChatGPT sign-in expires, Houston asks you to sign in again instead of showing an error.

Fixes and polish

• The integrations catalog now shows every available app, instead of stopping partway through the list.
• Houston only offers models your connected provider can actually run.
• Additional reliability improvements under the hood.

Linear — cloud-v0.6.17

(no project)

• PRODUCT-1670 — Forced desktop update interrupts users mid-task: download in the background, "Restart to update" pill instead of the countdown dialog
• PRODUCT-1668 — Codex sign-in code submitted after the login session ended surfaces "no active login for openai-codex" as an unexpected error
• PRODUCT-1667 — Windows: Claude sign-in fails when CLAUDE_CODE_GIT_BASH_PATH inherits the WindowsApps bash.exe stub
• PRODUCT-1665 — Quiet-class Sentry warnings ignore their fixed fingerprint and regress old resolved issues (load_chat_history "Load failed", integration_status "Failed to fetch")
• PRODUCT-1657 — Allowed-models ceiling forces an OpenRouter model id onto Claude: "Reconnect to OpenRouter" loop / "model is not available"

AI Models

• PRODUCT-1621 — Sentry HOUSTON-APP-5AP: view-docs boot warm for providers gives up (503)

Backend

• PRODUCT-1479 — Review stateless arch in staging

Global Interfaces & Components

• PRODUCT-1659 — Improve the web-app UI in mobile
• PRODUCT-1555 — Responsiveness overhaul

WhatsApp draft

🚀 Nueva versión de Houston!
✅ Forced desktop update interrupts users mid-task: download in the background, "Restart to update" pill instead of the countdown dialog
✅ Codex sign-in code submitted after the login session ended surfaces "no active login for openai-codex" as an unexpected error
✅ Windows: Claude sign-in fails when CLAUDE_CODE_GIT_BASH_PATH inherits the WindowsApps bash.exe stub
✅ Quiet-class Sentry warnings ignore their fixed fingerprint and regress old resolved issues (load_chat_history "Load failed", integration_status "Failed to fetch")
✅ Improve the web-app UI in mobile
✅ Allowed-models ceiling forces an OpenRouter model id onto Claude: "Reconnect to OpenRouter" loop / "model is not available"
✅ Sentry HOUSTON-APP-5AP: view-docs boot warm for providers gives up (503)
✅ Responsiveness overhaul
✅ Review stateless arch in staging
Gracias a todos los que reportaron 🙌

Houston 0.6.16

Clearer waiting, fewer alarming errors

• Send a message to a sleeping agent and see it right away, with a calm waking up notice instead of a red error.

Connections that just work

• Connecting an app now opens its sign-in reliably, even when your browser blocks pop-ups.
• Signing in with a Claude subscription now works on Mac-hosted setups.

Fixes and polish

• Amazon Bedrock now offers only the models your key can actually run.
• Running a routine now uses the connections of the person who pressed Run.
• Additional reliability improvements under the hood.

Linear — cloud-v0.6.16

(no project)

• PRODUCT-1643 — Sending to an asleep agent freezes the composer ~5s before the message shows

AI Models

• PRODUCT-1641 — Bedrock picker offers model ids Bedrock cannot invoke (bare foundation ids, missing regional profiles) 🐛
• PRODUCT-1636 — Sentry HOUSTON-APP-57B: nvidia gpt-oss-120b fails with Unknown role: final 🐛
• PRODUCT-1632 — Amazon bedrock is not working 🐛

Backend

• PRODUCT-1634 — Sentry HOUSTON-APP-PE: TLS certificate verification fails (UnknownIssuer) 🐛

Chat Experience

• PRODUCT-1640 — Waking and connectivity failures never reach Sentry; capture them low-noise and escalate a stuck wake 🐛
• PRODUCT-1639 — Users see raw EngineError bodies in toasts instead of plain product copy 🐛
• PRODUCT-1638 — Waking 502 at turn start pushes "Couldn't update the board status" into the chat feed 🐛
• PRODUCT-1624 — Error resuming conversation 🐛

Global Interfaces & Components

• PRODUCT-1631 — Verify onboarding on mobile
• PRODUCT-1630 — Improve UI/UX in chat in the web-app mobile 🐛
• PRODUCT-1628 — packages/web/README.md still says the store UI was cut
• PRODUCT-1627 — Web self-host path relies on an app module-load side effect to set window.HOUSTON_CP (no test pins it)
• PRODUCT-1626 — Claude subscription OAuth fails from a browser against a macOS self-hosted engine (CLI writes to Keychain) 🐛
• PRODUCT-1625 — Web: popup blockers can block the integration-connect tab on Safari/Firefox 🐛

Scheduled

• PRODUCT-1637 — Can't change routine provider 🐛

WhatsApp draft

🚀 Nueva versión de Houston!
✅ Sending to an asleep agent freezes the composer ~5s before the message shows
✅ Bedrock picker offers model ids Bedrock cannot invoke (bare foundation ids, missing regional profiles)
✅ Waking and connectivity failures never reach Sentry; capture them low-noise and escalate a stuck wake
✅ Users see raw EngineError bodies in toasts instead of plain product copy
✅ Waking 502 at turn start pushes "Couldn't update the board status" into the chat feed
✅ Can't change routine provider
✅ Sentry HOUSTON-APP-57B: nvidia gpt-oss-120b fails with Unknown role: final
✅ Sentry HOUSTON-APP-PE: TLS certificate verification fails (UnknownIssuer)
✅ Amazon bedrock is not working
✅ Verify onboarding on mobile
✅ Improve UI/UX in chat in the web-app mobile
✅ packages/web/README.md still says the store UI was cut
✅ Web self-host path relies on an app module-load side effect to set window.__HOUSTON_CP__ (no test pins it)
✅ Claude subscription OAuth fails from a browser against a macOS self-hosted engine (CLI writes to Keychain)
✅ Web: popup blockers can block the integration-connect tab on Safari/Firefox
✅ Error resuming conversation
Gracias a todos los que reportaron 🙌

Notify reporters (then clear the Notify pending label):

• PRODUCT-1641 → ⚠️ no phone on issue
• PRODUCT-1640 → ⚠️ no phone on issue
• PRODUCT-1639 → ⚠️ no phone on issue
• PRODUCT-1638 → ⚠️ no phone on issue
• PRODUCT-1637 → ⚠️ no phone on issue
• PRODUCT-1636 → ⚠️ no phone on issue
• PRODUCT-1634 → ⚠️ no phone on issue
• PRODUCT-1632 → ⚠️ no phone on issue
• PRODUCT-1630 → ⚠️ no phone on issue
• PRODUCT-1626 → ⚠️ no phone on issue
• PRODUCT-1625 → ⚠️ no phone on issue
• PRODUCT-1624 → ⚠️ no phone on issue

Houston 0.6.15

Routines are easier to use

• Routines now feel more polished, with cleaner layouts, clearer navigation, and a simpler sign-in.

A new Claude model

• Choose Claude Fable 5.1 for your work, now available in Houston's model picker.

Fixes and polish

• Downloads from the web now save as files, instead of opening an extra browser tab.
• Voice typing now clearly tells you when your computer cannot support it, without failed retries.
• Additional reliability improvements under the hood.

Linear — cloud-v0.6.15

(no project)

• PRODUCT-1616 — Add Claude Fable 5.1 (claude-fable-5-1) across all model surfaces

AI Models

• PRODUCT-1618 — Investigate chatGPT usage in Houston 🐛

Global Interfaces & Components

• PRODUCT-1619 — Web file downloads open a blob tab instead of saving the file 🐛
• PRODUCT-1561 — feat(app): routines mobile usability + tier-2 sweep + full mobile CI gate

WhatsApp draft

🚀 Nueva versión de Houston!
✅ Web file downloads open a blob tab instead of saving the file
✅ Investigate chatGPT usage in Houston
✅ Add Claude Fable 5.1 (claude-fable-5-1) across all model surfaces
✅ feat(app): routines mobile usability + tier-2 sweep + full mobile CI gate
Gracias a todos los que reportaron 🙌

Notify reporters (then clear the Notify pending label):

• PRODUCT-1619 → ⚠️ no phone on issue
• PRODUCT-1618 → ⚠️ no phone on issue

Houston 0.6.14

Publish agents straight from Settings

• Copy an agent and publish it to the Agent Store, right from Settings — now the default tab when you open an agent, so sharing what you built takes one click.

New integration: Croma

• Croma joins the curated integrations list, discoverable by your agents for Colombian public data (traffic cameras, procurement records, company filings).

Fixes and polish

• Clearer status while things wake up. Renaming an agent, logging into a provider, or reconnecting after a pause now shows a calm "waking up" notice instead of looking broken.
• Local AI connections report their real state. A disconnected tunnel to your own computer's AI model no longer shows as still connected.
• Usage and provider error messages are easier to read and stay visible instead of scrolling out of view.
• Additional reliability improvements under the hood.

Linear — cloud-v0.6.14

(no project)

• PRODUCT-1612 — Sentry regression: launch_provider_login 503 — setup pod unreachable 🐛
• PRODUCT-1607 — Sentry regression: rename_agent — engine PATCH to pod failed 🐛
• PRODUCT-1606 — Sentry regression: reconnect_local_bridge — no saved local-bridge to reconnect 🐛
• PRODUCT-1605 — Sentry regression: [claude-login] Claude sign-in failed (exit signal) 🐛
• PRODUCT-1604 — Sentry regression: create_activity — "agent not found" 🐛
• PRODUCT-1602 — Sentry regression: [serve] credential openrouter 500 "fetch failed" 🐛
• PRODUCT-1601 — Sentry regression: [provider_error] google gemini-3.5-flash-lite MALFORMED_RESPONSE 🐛

AI Models

• PRODUCT-1600 — Azure models are not working 🐛
• PRODUCT-1594 — Sentry HOUSTON-APP-4YT: github-copilot OAuth fails with 403 no_copilot_access 🐛
• PRODUCT-1586 — Update Pi

Backend

• PRODUCT-1599 — Sentry HOUSTON-APP-595: TypeError fetch failed (undici) 🐛

Chat Experience

• PRODUCT-1578 — Show out of usage card at the bottom 🐛

Global Interfaces & Components

• PRODUCT-1589 — Sentry HOUSTON-APP-4ZA: Invalid Unicode escape in action suggestions 🐛

WhatsApp draft

🚀 Nueva versión de Houston!
✅ Sentry regression: launch_provider_login 503 — setup pod unreachable
✅ Sentry regression: rename_agent — engine PATCH to pod failed
✅ Sentry regression: reconnect_local_bridge — no saved local-bridge to reconnect
✅ Sentry regression: [claude-login] Claude sign-in failed (exit signal)
✅ Sentry regression: create_activity — "agent not found"
✅ Sentry regression: [serve] credential openrouter 500 "fetch failed"
✅ Sentry regression: [provider_error] google gemini-3.5-flash-lite MALFORMED_RESPONSE
✅ Azure models are not working
✅ Sentry HOUSTON-APP-595: TypeError fetch failed (undici)
✅ Sentry HOUSTON-APP-4YT: github-copilot OAuth fails with 403 no_copilot_access
✅ Sentry HOUSTON-APP-4ZA: Invalid Unicode escape in action suggestions
✅ Update Pi
✅ Show out of usage card at the bottom
Gracias a todos los que reportaron 🙌

Notify reporters (then clear the Notify pending label):

• PRODUCT-1612 → ⚠️ no phone on issue
• PRODUCT-1607 → ⚠️ no phone on issue
• PRODUCT-1606 → ⚠️ no phone on issue
• PRODUCT-1605 → ⚠️ no phone on issue
• PRODUCT-1604 → ⚠️ no phone on issue
• PRODUCT-1602 → ⚠️ no phone on issue
• PRODUCT-1601 → ⚠️ no phone on issue
• PRODUCT-1600 → ⚠️ no phone on issue
• PRODUCT-1599 → ⚠️ no phone on issue
• PRODUCT-1594 → ⚠️ no phone on issue
• PRODUCT-1589 → ⚠️ no phone on issue
• PRODUCT-1578 → ⚠️ no phone on issue

Houston v0.6.13

• Houston on mobile now has Agents, Tasks, and Settings tabs, plus a dedicated home for your agents.
• Mobile task boards now let you swipe between stages and open or create tasks in a focused full-screen chat.
• Files created by agents now open reliably from chat links across cloud and desktop.
• Apps that require no account connection now appear ready to use instead of showing connection errors.
• Stability and polish improvements under the hood.

Before you upgrade

• Quit Houston before installing. macOS does not always replace an app that is running.

Linear — cloud-v0.6.13

Integrations

• PRODUCT-1571 — Enable Xero integration in cloud

WhatsApp draft

🚀 Nueva versión de Houston!
✅ Enable Xero integration in cloud
Gracias a todos los que reportaron 🙌

Houston v0.6.12

• Scheduled routines now run reliably at every due time, including the top of the hour.
• Your language choice now stays in place after you restart Houston.
• Browser and mobile back and forward controls now move through Houston without leaving the app.
• Cloud agents can now use connected apps, custom tools, skills, memory, and saved files during on-demand work.
• Stability and polish improvements under the hood.

Before you upgrade

• Quit Houston before installing. macOS does not always replace an app that is running.

Linear — cloud-v0.6.12

Global Interfaces & Components

• PRODUCT-1564 — Persist user preferred language 🐛

Integrations

• PRODUCT-1542 — Cannot connect reddit integration in production 🐛

WhatsApp draft

🚀 Nueva versión de Houston!
✅ Persist user preferred language
✅ Cannot connect reddit integration in production
Gracias a todos los que reportaron 🙌

Notify reporters (then clear the Notify pending label):

• PRODUCT-1564 → ⚠️ no phone on issue
• PRODUCT-1542 → ⚠️ no phone on issue

Houston v0.6.11

• Cloud work now recovers more reliably when agents wake, laptops reconnect, or storage briefly stalls.
• OpenAI-compatible setup issues now show clear actions to reconnect or switch to an available model.
• Cloud agents can now run local code in isolated workers and keep the files they create.
• Azure OpenAI connections now keep the correct endpoint across agents and restarts after you reconnect once.
• Stability and polish improvements under the hood.

Before you upgrade

• Quit Houston before installing. macOS does not always replace an app that is running.

Linear — cloud-v0.6.11

(no project)

• PRODUCT-1535 — Pool needs autoscaling: single-use spends one Kata VM per turn, a fixed 3-worker pool drains in one conversation

AI Models

• PRODUCT-1532 — Azure models not working 🐛

Activity

• PRODUCT-1528 — Sentry HOUSTON-APP-58Q: Missions remain unread after re-sweeps 🐛

Backend

• PRODUCT-1529 — Sentry HOUSTON-APP-50B: Object store PUT fails (500) on GCS 503 🐛
• PRODUCT-1454 — Investigate about code execution and terminal usage for the agents [stateless cloud arch]

Chat Experience

• PRODUCT-1531 — Sentry HOUSTON-APP-4W4: list_conversations invalid or expired token (401) 🐛

Local Models

• PRODUCT-1530 — Sentry HOUSTON-APP-579: provider_error openai-compatible gemma4 returns 404 HTML 🐛

Skills

• PRODUCT-1526 — Sentry HOUSTON-APP-5AP: Skills refresh 404 after SkillsChanged 🐛

WhatsApp draft

🚀 Nueva versión de Houston!
✅ Pool needs autoscaling: single-use spends one Kata VM per turn, a fixed 3-worker pool drains in one conversation
✅ Azure models not working
✅ Sentry HOUSTON-APP-4W4: list_conversations invalid or expired token (401)
✅ Sentry HOUSTON-APP-579: provider_error openai-compatible gemma4 returns 404 HTML
✅ Sentry HOUSTON-APP-50B: Object store PUT fails (500) on GCS 503
✅ Sentry HOUSTON-APP-58Q: Missions remain unread after re-sweeps
✅ Sentry HOUSTON-APP-5AP: Skills refresh 404 after SkillsChanged
✅ Investigate about code execution and terminal usage for the agents [stateless cloud arch]
Gracias a todos los que reportaron 🙌

Notify reporters (then clear the Notify pending label):

• PRODUCT-1532 → ⚠️ no phone on issue
• PRODUCT-1531 → ⚠️ no phone on issue
• PRODUCT-1530 → ⚠️ no phone on issue
• PRODUCT-1529 → ⚠️ no phone on issue
• PRODUCT-1528 → ⚠️ no phone on issue
• PRODUCT-1526 → ⚠️ no phone on issue

Houston v0.6.10

• Installing a skill now keeps your search and results in place, so you can add several without retyping.
• Xiaomi MiMo now gives clearer model guidance and an easy switch when a model is unavailable.
• Linux users can now download the AppImage directly from the Houston website.
• Routines keep their selected provider, and provider reconnects no longer loop on stale status.

Before you upgrade

• Quit Houston before installing. macOS does not always replace an app that is running.

Linear — cloud-v0.6.10

AI Models

• PRODUCT-1517 — MiMo models not working 🐛

Backend

• PRODUCT-1515 — Chat asks to reconnect Anthropic while the conversation's selected provider is OpenRouter (wrong-provider auth card) 🐛

Global Interfaces & Components

• PRODUCT-1519 — Add linux download option in landing page
• PRODUCT-1512 — Skill input search gets reset after installing a skill 🐛

WhatsApp draft

🚀 Nueva versión de Houston!
✅ Add linux download option in landing page
✅ MiMo models not working
✅ Chat asks to reconnect Anthropic while the conversation's selected provider is OpenRouter (wrong-provider auth card)
✅ Skill input search gets reset after installing a skill
Gracias a todos los que reportaron 🙌

Notify reporters (then clear the Notify pending label):

• PRODUCT-1517 → ⚠️ no phone on issue
• PRODUCT-1515 → ⚠️ no phone on issue
• PRODUCT-1512 → ⚠️ no phone on issue