What's new

#AI tools that understand Orca

AI tools connected to Orca now get a map of how it works (stacks, orca.yaml, workspaces, the gateway, databases, worktrees), and can read Orca's whole documentation through a new orca_docs tool. The docs are built into Orca, so they always match the version you run, and need no web access. Before an AI tool writes orca.yaml or orca-workspace.yaml, it's told to read the relevant page instead of guessing keys.

#Move projects onto shared databases

The Databases page now lists every project's own database container. Once a project requires its database on a shared server, its row offers Import into …, which copies the data over:

• Orca takes a snapshot of the new database first, so you can go back.
• A stopped container is started for the copy and stopped again.
• The copy streams from one container straight into the other.
• Between MariaDB and MySQL, the dump is translated on the way: collations, and the first line of MariaDB's dumps that MySQL rejects.
• Containers with a random admin password are read with the app's own user.

AI tools get the same with find_database_services and database_import. So you can now ask:

Consolidate our apps' databases into shared databases defined in the workspace, and move their data over.

In our test, Claude:

1. read the docs;
2. took stock of every app's database;
3. noticed a WordPress site ran MariaDB while the team wanted MySQL, and asked;
4. added the shared servers to the workspace;
5. wrote requires: for each app;
6. copied the data over;
7. compared rows on both sides, and checked that the apps still worked.

#Also

• Update on the Workspace page no longer fails when the workspace's repo can't be pulled (your own commits, or offline). It says so and applies the workspace as it is in its folder, including your local edits.

Downloads and full changelog on GitHub

#Databases

A project can now say in its orca.yaml which database it needs, and Orca provides it:

requires:
  database:
    engine: mysql
    version: ">=8.0"
    replaces: db

Orca gives the project a database and a user of its own, with a generated password, and writes the connection into its .env. The database comes from:

1. your team's shared server, when the workspace has one for that engine and version;
2. the project's own database service, when it has one and the workspace doesn't (as without Orca);
3. a server Orca runs itself: MySQL, MariaDB or PostgreSQL, one per version, started the first time a project needs it.

With replaces:, the project's own database container doesn't start, and the connection is set on the services that used it, even when their compose file hardcodes it. Once a database exists, it stays on its server, so adding a shared server later never leaves data behind.

#Shared servers for a team

A workspace shares servers with a databases: section naming services in its shared compose file. Each project gets its own database and user there, instead of every developer running a database container per project. The workspace file can choose per project with database: mysql or database: own.

#The Databases page

A new Databases page shows every server and the databases on it, which project each is for, its user and size, and what each project requires (or why it can't be provided). For each database:

• Connect: host, port, user and password, and a URL for TablePlus, DBeaver or DataGrip.
• Snapshot and Restore: compressed SQL dumps on your computer, with a label such as before-migration.
• Reset: empty it and run the project's seed command.

A project's database also shows on its stack page, on its card on the Workspace page and on the dashboard. Setting up a workspace project creates its database before the setup command runs, so migrations find it.

#For AI tools

list_databases, database_connection, database_snapshot, database_restore and database_reset. Passwords never reach the AI tool, and restoring or resetting needs your approval. Ask "snapshot onecrm's database, then run the migration", and "restore it" if it goes wrong.

See Databases for everything.

Downloads and full changelog on GitHub

#See every request

A new Requests tab on the Gateway page lists requests through the gateway as they happen: time, status, address, route, how long it took and how large the answer was. Filter it by hostname, path, project or status, or keep Only problems.

A failed request says why, in plain words: "Nothing on the gateway's networks answers to crm-app: its container isn't running", or "crm-app runs but doesn't accept connections on that port: it's still starting". Click it for the gateway's own error.

#Check this machine

Check this machine on the Workspace page checks everything a computer needs for the workspace, and says what to do about each problem:

• the gateway on ports 80 and 443;
• the domain pointing at this computer;
• the team certificate's expiry;
• Orca's certificate authority being trusted;
• the shared network;
• registry logins;
• the secret manager's CLI;
• access to every project's repository.

#No surprise certificate expiry

When a workspace's certificate is within 30 days of expiring, the Workspace page, the Gateway page and the gateway's start page say so, in yellow, then red once it has expired.

#Group a product's projects

A workspace project can name its group, such as group: oneWebshop for its backend, frontend and admin. They then show together on the start page, the Gateway page, the dashboard and the Workspace page.

#More for AI tools

• gateway_requests: recent requests and why the failed ones failed.
• join_workspace: join a workspace from scratch. Your AI tool asks you first.
• check_workspace: the same check as the Workspace page.
• fetch_project_secrets: fill in a project's secrets with your own Passbolt or 1Password CLI; the values never reach the AI tool.
• identify_ports: what each port of a container is.
• gateway_routes now also says whether each route's service runs.

#Also

• The start page loads much less: icons are separate files the browser keeps, instead of being embedded in a page that refreshes every 20 seconds.

Downloads and full changelog on GitHub

#A tidier start page

• Shared services as a strip. A workspace's shared services, such as Mailpit, Grafana and MinIO, come first as a row of small chips with a status dot, instead of rows of full cards.
• Small projects share a row. A project with one or two services no longer takes a whole row: it sits next to others.
• Room for the search box. How many services run is now under the title, so the search box gets the space, and everything in the top bar lines up.

#Your own apps next to a workspace

When a workspace brings its own certificate, such as one for *.local.acme.dev, your own routes outside that domain, such as myapp.localhost, keep a certificate from Orca's own certificate authority. Before, they got one your browser didn't trust. Names two levels down, such as api.crm.local.acme.dev, which a wildcard certificate doesn't cover, get one too.

Downloads and full changelog on GitHub

#The Gateway page, like the start page

The routes table on Orca's Gateway page now works like the gateway's start page:

• Grouped by project, with shared services last. Click a group's name to fold it; Orca remembers which ones you folded.
• Whether each service runs: Running, Stopped or Unhealthy, instead of only whether the route is on. A route you switched off says Paused.
• Not set up yet: a workspace project that isn't on your computer says so, with a Set up link that takes you to it on the Workspace page.
• Search: with more than a few routes, a box above the table finds them by hostname, container, project or status. Press / to jump to it.

#Table headers stay put

Scrolling a long table (containers, images, volumes, routes) no longer shows rows above its header: the header now sits flush at the top.

Downloads and full changelog on GitHub

#A start page for lots of services

The gateway's start page has three new controls in its top bar, for when a workspace has dozens of services:

• Compact: each service on one short row instead of a card, so many more fit on a screen. Cards switches back.
• Running only: hides what's stopped or not set up yet. Searching still finds everything.
• Fold a group: click a group's name to fold it away; it shows how many services it has.

Each browser remembers your choices.

#Steadier search

Typing in the start page's search box no longer makes the page jump: the hint under the box (what Enter will do) keeps its space.

#Join a workspace you already have

The Workspace page now offers Choose its folder… for a workspace you've already cloned: Orca joins it where it is, and its projects go beside it. Pasting a path that starts with ~/ works too.

#Also

• The dashboard's app cards show the full name: the pin and hide buttons appear over the corner on hover instead of taking space.

Downloads and full changelog on GitHub

#Your team's logo on the start page

A workspace can now put its logo in the gateway start page's top bar, next to its title, and on the Workspace page. Add gateway.logo to the workspace file: an image in the workspace repo, a built-in icon or an https URL. Leaving the workspace puts back the start page's previous title and logo.

#My links

Your own cards at the top of the start page: your sprint board, a staging admin, anything you open every day. Add them on the Gateway page's Landing page tab. They're kept on your computer only.

#Projects you haven't set up yet

The start page now shows a workspace's projects that aren't on your computer yet, as dashed cards marked Not set up. Clicking one opens Orca at that project, ready to set up with one click.

#What changed in the workspace

When Orca pulls new changes to your workspace, the Workspace page lists them: Workspace updated: Add Gravity, Renew the certificate.

#Search the web from the start page

The start page's search box now searches Google when no service matches, without setting it up first. Pick another engine, or turn it off, on the Landing page tab.

#Joining puts things where your code is

When you join a workspace, Orca suggests the folder where your git checkouts already are, such as ~/projects, and shows where the workspace and its projects will go. If you already have a copy of the workspace there, Orca joins it where it is.

#Worktrees

The Projects page is now called Worktrees, so it's no longer confused with a workspace's projects. Its page is much more compact: your repositories are listed right away. Paths starting with ~/ now work when you add one.

#Less to read

Pages say what they're for in one line, with a How it works link to the docs, instead of paragraphs of explanation. This covers the Gateway, Kubernetes, Settings, container and stack pages, the worktree setup wizard and more.

#Also

• The note about switching from Docker Desktop now says correctly that containers and images aren't copied over.

Downloads and full changelog on GitHub

#Ports that say what they are

The Gateway page's Suggested routes now tell you what each port is:

• Names: :8123 ClickHouse HTTP, :5432 PostgreSQL, :3478 STUN/TURN, worked out from the image and the port number.
• Web or not: a blue dot marks a web page the gateway can serve, a hollow dot anything else.
• Port runs: a run of ports, like a TURN server's relay range, is one label (:49160–49200 relay ports) instead of forty.

Check asks a container's ports directly. It names a web page by its title, and anything else by its protocol: PostgreSQL, Redis, SMTP, SSH, MySQL and more. It only runs when you click it.

Databases and other containers with no web pages no longer show up as routes to add. They're listed under Other services, with the address to reach them on your computer, such as localhost:5432. Add to Gateway now picks the container's web port.

#A sturdier gateway start

If something else on your computer is using port 2019, such as another Orca's gateway, Orca no longer sends your gateway's settings to it by mistake. It now says what's wrong and how to fix it. Starting the gateway also no longer fails when a just-created certificate takes a moment to appear.

#Also

• Containers with many ports no longer push their port labels out of the suggestion card.

Downloads and full changelog on GitHub

#Watch Orca in action

There's now a two-and-a-half-minute tour of Orca, recorded from the real app: a project folder that just runs, every container live, a team set up from one repo, and Claude setting up and fixing apps through Orca. It's on orca-desktop.com and YouTube.

#A start page that gets to the point

The gateway's landing page now keeps its title, search box and counts in one bar across the top that stays put as you scroll. Your services start right below it instead of halfway down the page.

#Logs in colour

Log views show the colours apps print. Ghost, Laravel, Vite and many others colour their logs, and until now that came through as [36mINFO. Copying and searching use the plain text.

#Fewer false alarms

"Container exited" alerts are now only for real crashes and out-of-memory kills. Stopping a container, and Orca restarting its own gateway, no longer raise an error.

#Clearer app cards

On the dashboard's Your apps:

• an address (marked ↗) opens the app in your browser;
• the rest of the card opens it in Orca, with its containers, logs and settings. Open → appears when you hover over it.

#Workspaces

• Addresses from a project's own orca.yaml now show on the Workspace page, in setup results and to AI tools, not only the routes in the workspace file.
• A project's icon is used for the routes its own orca.yaml defines.
• gateway.title in the workspace file sets the landing page's title, such as "Acme dev".
• Updates without clicking: the Workspace page refreshes every few seconds, so projects an AI tool sets up show up there by themselves.
• Shorter paths: progress steps show ~/code/… instead of the full path.

#Also

• Addresses no longer end in :443, and app cards keep a route's path.
• The Open a project folder dialog and a stack's page show ~/… paths.
• Ghost joins the built-in icons.

Downloads and full changelog on GitHub

#Every app gets its icon

Apps now show their logos on the gateway's landing page, in the dashboard's Your apps, on the Workspace and Gateway pages and in the ⌘K palette.

• Most need no setting at all. The container's image picks the icon: postgres:17 gets PostgreSQL's logo, grafana/grafana Grafana's, axllent/mailpit Mailpit's. Orca knows about 80 common images.
• Or choose one with icon: in orca.yaml, a label, the workspace file or a route's Edit dialog:
  • a built-in icon, such as icon: grafana, icon: redis or icon: keycloak. There are 71: brand logos, plus generic ones such as mail, database, api and worker;
  • an image file in your repo, such as icon: assets/logo.svg. Orca reads it in, so it works offline;
  • or an image URL.
• Workspaces get icons too. Projects and shared services take icon, and routes take a description. A team lead sets each project's logo once, for everyone.

The brand icons come from Simple Icons (CC0) and the Mailpit logo from Mailpit (MIT).

Docs: Icons

#Ask your AI tool whether Orca is up to date

Two new MCP tools, 74 in all:

• check_for_update tells your AI tool which version you're running, what the latest one is, and what each newer release is about. Ask "is my Orca up to date?". When you ask about a feature your version doesn't have yet, the tool can tell you it needs an update.
• offer_update brings Orca to the front with its Update prompt. Nothing installs until you click Update. On a server running only the daemon, it gives the upgrade command instead.

Docs: Orca itself

Downloads and full changelog on GitHub

#Workspaces: your team's local setup, in one repo

A workspace is a git repo with an orca-workspace.yaml. It describes everything a developer needs to work on your team's projects locally: the shared services everyone uses, the Docker network they meet on, your own domain and certificate, and every project with its setup steps and secrets. A new developer installs Orca, joins the workspace, and has any project running in minutes. Or they ask Claude to "spin up billing".

• Join in one step. Open Workspace in the sidebar and paste the repo's URL. You can also click an orca://workspace?repo=… link from your wiki, or run orca workspace join <url>. Orca then:
  • clones the repo and creates the network;
  • serves *.<your domain> with your team's certificate;
  • starts the shared services and adds every route.
• Projects are a menu. Each one shows Running, Cloned or Not cloned. Set up runs these steps:
  1. Clones the project, or uses the checkout you already have.
  2. Creates its .env from the template.
  3. Fills in secrets.
  4. Runs its setup command once.
  5. Starts the shared services it uses.
  6. Starts it and shows its URLs.
• Secrets stay in your secret manager. The workspace file holds references only: passbolt:<id> or op://vault/item/field.
  • Without the manager's CLI, each missing value comes with a link to open its exact entry.
  • With go-passbolt-cli or the 1Password CLI set up, Fill in automatically fetches them all.
  • Values only ever go into the project's .env, which only your user can read.
• AI tools set projects up for you. With Orca connected over MCP, "set up the CRM" finds the project (workspace_status) and sets it up (setup_project), then gives you its URLs. AI tools are told never to ask for secrets in the chat.
• Always up to date. Orca pulls the workspace when it starts and applies whatever changed: a new project, a route, a shared service or a renewed certificate. Update does it right away. Leave puts your gateway's domain and certificate back as they were.
• Coming from a hand-run reverse proxy? orca workspace import-haproxy haproxy.cfg turns an HAProxy config into workspace routes, and the gateway now handles what such setups need (below).

More on the site: orca-desktop.com/workspaces, and eight new docs pages starting at Workspaces.

#Your apps, grouped

The dashboard's Your apps now groups your stacks by where they come from: the workspace's projects first, with its shared services as one compact row of links, then each project's worktree instances by branch, then everything else. Click a group's name to fold it. Hover a card to pin it to the top or hide it.

#Gateway

• Hostname patterns. A route for *-shop matches north-shop, south-shop and every other tenant. A route for * alone catches every hostname under the domain. The most specific route wins.
• WebSocket-only routes. websocket: true sends only WebSocket connections to a container, so an app's Reverb or Vite socket can live apart from its pages.
• Path routes that work. A route with a path no longer matches every request to its hostname, and a longer path wins over a shorter one (/api/admin/* before /api/*).
• A wildcard certificate per pattern with Orca's certificate authority, and your own certificate is in place before the gateway's first start.

#Also

• orca workspace join|status|setup|update|leave|import-haproxy on the command line.
• After-start hooks in a stack's orca.yaml also run for stacks whose compose file is in a subfolder.
• A failing setup command shows the end of its output, wherever the tool printed its error.

Downloads and full changelog on GitHub

#Safer and more solid

We put the last few releases through a round of adversarial reviews: security, backend and app. This release fixes what they found.

#Security

• Your AI tools' config files stay private. Connecting Orca to Claude, Cursor, Codex and the others keeps each config file's permissions, keeps symlinked dotfiles as symlinks, and never loses a change the tool itself made at the same moment.
• Codex's config.toml keeps its comments, and Orca writes it as valid TOML.
• More secrets stay on your machine. Before the built-in assistant sends tool results to a cloud model, Orca now also blanks out:
  • Kubernetes Secret manifests
  • .env-style command output
  • every password in a URL
  • keys such as APP_KEY and *_PAT
• The built-in assistant can't delete things. It runs tools without asking, so tools that remove containers, images, volumes, routes or projects, prune, or overwrite orca.yaml aren't available to it. Do those in Orca, or from an AI tool connected over MCP, which asks you first.
• MCP clients ask before anything that changes state. Stopping a container, switching profiles, running a command or applying Kubernetes YAML is now marked destructive; only tools that add or start something aren't.
• The compose file editor only edits YAML files inside the stack's own folder.
• On Windows, Open in… refuses folder names that the command prompt would run as commands.

#Fixes

• Kubernetes port-forwarding works again. Every connection was refused.
• Opening a folder:
  • It uses the compose files listed in the folder's orca.yaml.
  • It follows Compose's own rules for project names.
  • It warns you when another folder already runs under the same name, instead of replacing it.
  • Errors stay on screen, and Open tries again after you fix the problem.
• A first compose.yaml is no longer refused because an .env file it mentions doesn't exist yet.
• orca mcp answers batched requests and survives unreadable input.
• On Windows, Orca's own orca command updates even while an AI tool is running it.
• Gateway settings: if the gateway won't start with new settings, Orca puts the previous ones back instead of leaving it stopped.
• In the app:
  • Shell works on the container you're already looking at.
  • Deploying a stack from the catalog opens its page.
  • The assistant settings no longer mix up one provider's models with another's.
  • The AI window says when no supported AI tool is installed.
  • A custom search URL you typed wrong stays in the field so you can fix it.
  • ⌘K keeps your selection in view.
  • A stack page's URLs refresh by themselves.
  • The Ollama container restarts with your computer as intended.

Downloads and full changelog on GitHub

#Your start page: search on the landing page

The gateway's landing page at https://localhost now opens with the cursor in a search box. Set it as your browser's home page, and every new start puts your services one keystroke away.

• Find any service as you type. Search by name, hostname, description, project, container or status: mail, admin and even unhealthy all work. The matching text is highlighted, and Enter opens the best match.

• Search the web from the same box. Choose Google, DuckDuckGo, Bing, Brave Search, Kagi, Ecosia or Startpage on the Gateway's Landing page tab. When no service matches, Enter searches there; Shift+Enter always does.

• Or your own search. Custom URL… takes any search address with {query} in it: your company wiki, Jira, or GitHub code search (https://github.com/search?type=code&q={query}).

• Go straight to a site. Type something like github.com/acme/shop and Enter opens it.

• All from the keyboard.

  • ↑ ↓ choose between matches.
  • Ctrl/⌘+Enter opens in a new tab.
  • / or Ctrl/⌘+K jumps to the box from anywhere on the page.
  • Esc clears it.

  A hint under the box tells you what Enter will do before you press it.

The search box is on by default and only finds services until you pick a web search. Switch it off on the Landing page tab if you don't want it. The tab also shows the address to use as your browser's home page.

Thanks to Yannick for the idea!

Docs: Search, and using it as your start page

Downloads and full changelog on GitHub

#Open a folder. It's running.

Starting a project you already have is now one step. Click Open folder on the dashboard or the Containers page, press ⌘K, or just drop the folder onto Orca's window.

Before anything starts, Orca shows you what's inside: the compose file, its services and profiles, whether there's an orca.yaml, and whether it's a git repository. Start runs it as a stack and opens its page.

• Git repositories can be added to Projects in the same step, so every worktree can run side by side.
• A folder without a compose file gets a first compose.yaml, which you can edit before it starts.
• Your files are safe. Orca never overwrites an existing compose file. The old Compose dialog did: Validate replaced the project's docker-compose.yml with a sample. That dialog is gone.

#Jump to your code

A stack's page now shows its HTTPS URLs as buttons, plus Open in VS Code (or Cursor, Zed, Windsurf, IntelliJ, PhpStorm) and Open in Ghostty (or iTerm, Warp, Terminal), whichever you have, and Finder.

Every service has Shell, which opens a terminal inside its container. You'll also find Open shell in each container's "…" menu.

#⌘K finds everything

The command palette now searches as you type:

• stacks
• gateway URLs (Enter opens them in your browser)
• projects
• containers, including a shell in each one
• images

Its actions do the real thing: Run a container…, Build an image… and Add a gateway route… open their dialogs, and Open a project folder… starts right there. Press ? for keyboard shortcuts.

#A dashboard about your apps

The dashboard leads with Your apps: every stack, how many services run, and its URLs, one click away. Stopped stacks get a Start button.

On a fresh install it shows three ways to begin instead: open a project folder, add a git repository, or start from the catalog. The CPU and memory charts now show what your containers use, scaled so the lines actually move.

#Easier to find your way around

• Sidebar: grouped into Develop, Resources, Platform and System. Kubernetes is marked "off" when it isn't enabled, and Fleet appears once you add a remote host.
• Stacks: stack names on the Containers page are now links, with an Open stack button.
• Container pages: labelled buttons (Start, Stop, Restart, Expose via gateway, Save as image, Remove) instead of a row of icons.
• Gateway:
  • the two overlapping settings tabs are now one Settings tab;
  • routes that come from orca.yaml link to Edit in orca.yaml;
  • stopping the gateway asks first.
• Kubernetes: row actions are tidied into a "…" menu, and stopping Kubernetes asks first.
• Builds: now has a Build an image button.
• Volumes: the empty page explains what volumes are for and offers Create volume.
• Labels: sentence case throughout, and delete icons stay quiet until you point at them.

#Fixes

• Kubernetes: the pod Restart action deleted the pod. It now restarts it, and asks first.
• Confirmation buttons: confirmations for actions like "Stop all containers" no longer show a red Delete button; the button names the action.
• Compose wizard: creating a stack no longer overwrites an existing stack with the same name.
• Docker unreachable: when Orca can't reach Docker, Containers and Images now say so, with a button to System Health, instead of loading forever or quietly showing stale data.
• Daemon errors: the "daemon not running" screen shows the actual error, and only points you to a reinstall when the daemon is really missing.
• Charts: chart axes no longer read "now now now".

Docs: Quickstart · Stacks: open a project folder · Find your way around

Downloads and full changelog on GitHub

#Run Orca from Claude: one click, no tokens

Orca is an MCP server, and connecting your AI tool to it now takes one click. Open Settings → AI & Agents and press Connect next to Claude Code, Claude Desktop, Cursor, VS Code, Windsurf or Codex. Orca shows which of them are on your computer and adds itself to their config, leaving everything else in the file alone.

Then just ask: "Start the project in ~/code/shop and give me its URL", "Turn on the search profile", "Why does the api container keep restarting?"

• No tokens to copy. The AI tool runs orca mcp, which reads Orca's token itself. Nothing secret ends up in an AI tool's config or a committed .mcp.json.
• Uses the subscription you already have. No API key or per-token billing: Claude Pro/Max, ChatGPT, Cursor and Copilot plans all work.
• Survives updates. Orca keeps its own copy of the CLI and refreshes it on start, so updating or moving the app doesn't break anything.
• From a terminal too: orca mcp install connects every AI tool it finds, and orca mcp status shows where things stand.
• Remote hosts: orca mcp --host build-server lets an AI tool manage a host from Remote Hosts.
• If one of your AI tools is installed but not connected, the dashboard invites you to connect it.

#Your AI can now do (almost) everything you can

21 new tools, 69 in total, so an AI tool can manage projects from start to finish:

• Projects: deploy_project starts a folder with a compose file or orca.yaml; add_project / list_projects / remove_project handle worktree projects.
• Stacks: restart, pull, logs, switching compose profiles, and reading and writing orca.yaml (validated before it's saved).
• Gateway: status, start, stop, and listing, adding and removing routes.
• Builds: list targets, start a build, follow its status and log.
• Volumes: remove one.

They run through Orca's own API, so they validate exactly as the app does. Tools that delete or overwrite something are marked so your AI tool asks you first.

#The built-in assistant goes local

The chat window behind ✦ is now optional and built for models on your own computer: free, private, and offline.

• Finds your model servers. Orca looks for Ollama, LM Studio and llama.cpp's llama-server, and lists their models. Pick one and click Use. No key needed.
• Made for small models. A local model gets a lean set of 19 tools and up to five minutes per answer.
• Ollama in a container on Linux and Windows now pulls qwen3:8b. On a Mac, Orca suggests the Ollama or LM Studio app instead, since containers there can't use the GPU.
• Cloud API keys (Anthropic, OpenAI, Gemini, any OpenAI-compatible endpoint) are still there under Cloud API key, with current Claude models.
• When no model is set up, the assistant window offers to connect Claude instead, and Ask AI on a container or build gives you the question to paste.

#Fixes

Writing the new docs at orca-desktop.com turned up a list of things that didn't work as described. They do now:

• Schedules run in your local time, not UTC.
• Stacks with several compose files start with all of them, and the stack page lets you pick which files are used.
• Kubernetes: enabling Orca's cluster no longer changes your kubeconfig's current context. Port-forwarding services works, jobs show their logs, deployments and services show their labels and selectors, and start, stop and reset work on macOS (Lima).
• Auto-deploy webhooks: Docker Hub webhooks can authenticate with ?token=, tag rules are enforced, and pulls use your saved registry credentials.
• Cleanup only ever deletes Orca's own Lima VM.
• orca deploy accepts a project folder or a compose file inside one.
• Saving gateway settings restarts the gateway so the change takes effect, and the landing page merges sections with the same name.
• dev.start: on-create starts new worktrees as it should.
• The assistant blanks out secrets (passwords, tokens, DATABASE_URL and the like) in tool results before they reach the model.
• On macOS, Start Docker restarts the VM when the engine is down.
• Settings only offers GPU options when there's a GPU, and shows where an AI key comes from (setting or environment variable).

Full docs: AI agents and MCP · Built-in assistant

Downloads and full changelog on GitHub