Multica Release Notes

Per-run cost details, shorter onboarding, revoked access for former members, and steadier runs

NEW FEATURES

The execution log shows a cost bar for each run, along with its tokens, time, and model.

Run details compare runs by cost, tokens, or agent time, with a cumulative view one click away.

Expand any run to see its token and cost breakdown.

IMPROVEMENTS

Skipping runtime setup during onboarding opens the install guide Issue right away.

BUG FIXES

Members who leave or are removed from a workspace stop getting its notifications and live updates.

The agent builder fills in the configuration even when its draft comes back slightly malformed.

The agent builder's retry warning stays gone once you close it or edit the form yourself.

Tasks picked up but never delivered to your computer go back to the queue instead of getting stuck.

Resuming in a local folder no longer replays committed changes as drafts, and keeps your other drafts.

Cache cleanup no longer deletes tracked folders like .next, or Git checkouts inside them.

Claude Sonnet 5.5 and Haiku 5.5 runs now show an estimated cost.

Arabic, Hebrew, and other right-to-left text reads in the right order, even with English mixed in.

Text and link list properties, workspace switching in Cmd+K, one-click starters, and steadier runs

NEW FEATURES

Custom properties can hold a list of text or links, and be filtered by exact value.

Press Cmd/Ctrl+K to switch to another workspace.

On desktop, switching workspace brings back the tab you last had open there.

Clicking a conversation starter in an empty chat sends it right away, and never overwrites a draft.

Turn on start-at-login for your local daemon from the command line; it stays off until you do.

IMPROVEMENTS

The Issues list loads its first page faster.

Desktop navigation and the tab bar now line up.

BUG FIXES

Long Codex threads continue across local folders, and rate limits no longer stall a run.

The Codex CLI inside ChatGPT.app is detected.

Pi tasks get the MCP servers you configured.

Failed Pi and Kimi runs show as failed instead of finishing quietly.

Antigravity keeps its answer even when the connection drops right at the end.

A CodeBuddy startup error no longer stops runs already in progress.

Oh My Pi starts a fresh session when asked, instead of resuming the old one.

Tasks on Windows keep access to the system temp folder.

Tasks are no longer cut off as unresponsive after your computer wakes from sleep.

The local daemon recovers even when its health-check port is taken.

Hermes picks up its earlier session across local folders.

Busy Autopilot schedules no longer go silent, and runs keep their Autopilot attribution.

PR wakeups no longer miss a change, and explain problems they can recover from.

Lark clears its Typing reaction once the task ends.

WeCom Issue notifications come from the bot of the agent doing the work.

Issue prefixes that contain digits show up in search.

Editing a project at the same time as someone else no longer wipes their fields.

Follow-ups sent while a chat is still loading are no longer lost.

Plain attachment download links keep working, and text steps in run details are labeled correctly.

Server errors no longer expose internal error details.

Cumulative run cost curves, a faster Issues list, and steadier workspace switching

NEW FEATURES

The run chart follows your pointer, and shows the quiet stretches between runs.

An Issue's sidebar shows the cumulative cost of its runs as a curve.

Mark a duplicate Issue from the command line and keep the link to the original.

IMPROVEMENTS

The Issues list stays steady as runs come and go, and the working-agents filter returns faster.

BUG FIXES

Codex offers GPT-6.1 Sol, and cost for the GPT-6 models reads at current prices.

Hitting a provider's usage limit reads as a quota, not as a credential problem.

Closing the floating chat leaves your keyboard shortcuts working, and keeps your draft.

On Windows, a repository checks out even after your workspaces folder was moved.

Switching workspace from an open chat lands you there, with your session and project still selected.

Conditional wakeups, instant search, deliverable previews, and a new Settings page

NEW FEATURES

An Issue can wake its agent when a status changes, a sub-issue finishes, or a pull request moves.

Give a wakeup an expiry, and see every check it made.

Issue and project search return results instantly on web and desktop.

Reply to a running agent and choose to add to this run, queue a new one, or restart it.

See a receipt telling you the running agent got your reply.

Find an Issue's deliverables in one sidebar, with versions and details.

Attachments sit in a grid, and HTML, Markdown, CSV, JSON, and YAML preview in place.

Mermaid diagrams show whole, zoom in close, and images copy in one click.

See a run's timeline on the Issue, along with failed and cancelled states.

Let an Issue move itself the way you choose once its pull requests are merged.

Preview an Issue beside the list without leaving it.

WeCom notices arrive in each person's own language.

Pick Claude Opus 5.5 and GPT-6 Sol/Luna from the model list.

See a repository's project description as you pick it.

IMPROVEMENTS

Settings are grouped into personal, workspace, and this device, and you can search them.

Agent task history pages through and adds up the time spent.

Back, forward, and the sidebar toggle sit together on the left in the desktop app.

See how many agents each MCP server is assigned to.

BUG FIXES

Model settings you saved stay saved.

A Hermes task runs with the settings you picked.

MCP tool inputs stay fully visible.

An attachment you quote in Telegram reaches the agent with your message.

WeCom no longer answers the same message twice.

Tools with the same name across plugins no longer replace each other.

A link opened by Issue identifier expands the comment it points to.

Opening a project on mobile shows its Issues right away.

Select all, and every selected Issue is deleted.

Run status reads accurately.

A saved view keeps the name you typed.

The wakeup menu shows its options correctly.

A pull request row shows the whole diff, and spins while checks run.

NEW FEATURES

Send photos, videos, audio, and files to an agent on Telegram.

Attachments an agent produces come back to you in Telegram.

Pick Simplified Chinese, English, or your system language in the mobile app.

See an Issue's linked pull requests, and let it complete itself once they are all merged.

Watch an Antigravity agent's tool steps as they happen.

Add new instructions to a Grok Build task while it is still running.

Browse every attachment on an Issue full-window and page through them.

Read the whole documentation site in French.

Find licensing answers, the privacy policy, and the team behind Multica on the site.

Self-hosted admins can make automatic titles and quick actions respond faster.

IMPROVEMENTS

Comment actions are grouped by what you came to do, with Edit and Resolve first.

BUG FIXES

Usage for a resumed Claude session counts this run only.

Cache hit rates on Issues and runtimes no longer read higher than they are.

Agent replies in a thread appear in the order they were sent.

A task on Windows starts even when a file is briefly in use.

An Issue link in DingTalk shows up without stray characters.

A self-hosted server explains an untrusted certificate, and can trust your own CA.

Steering running tasks, duplicate Issue marks, and steadier task runs

NEW FEATURES

• Add new instructions to a Claude Code or Codex task while it is still running.
• Mark an Issue as a duplicate from the status picker, jump back to the original, and see that link on lists.
• Set an Issue's custom properties as you create it from the command line.
• Mention an agent in a Telegram group and it already knows the recent conversation.
• Install the command line tool on Windows straight from the download page.

IMPROVEMENTS

• Each OpenClaw agent works in the folder you configured for it.
• Attachments you upload while creating an Issue show up in its description.
• A Lark bot that stays silent now points you to where delivery is stuck.
• Scheduled Issue wakeups are shown in your own time zone.
• Getting to a task's GitHub pull request is faster.
• The running indicator is smoother and lighter on your machine.

BUG FIXES

• New Codex models show up in the picker as soon as they are out.
• A command line sign-in that cannot reach the server says so, instead of waiting forever.
• An invited member can finish signing up on a self-hosted server that restricts signups.
• A task whose start goes unconfirmed is picked up again instead of stalling.
• Cancelling a task answers right away, and a reply in a thread reaches the agent that owns it.
• The mobile app reconnects on its own after the connection drops.
• Desktop toolbar buttons are spaced correctly again.
• The Windows installer runs on PowerShell 5.1.
• Confirmation dialogs in French no longer scroll sideways.
• An Autopilot keeps a record of the Issues it creates.
• A guest squad leader wakes up and picks the work up.
• You can tell where a WeCom reply was lost on its way back.

NEW FEATURES

Set an Issue to wake an agent up again when a comment arrives, or on a schedule you choose.

Manage those wakeup rules from the Issue sidebar or from an Autopilot.

Choose the branch or commit a project's repository work starts from.

Copy a direct link to any comment or reply, and open it with that comment highlighted.

A WeCom answer comes back inside the message you asked from.

Point a self-hosted Multica at Gitea or a compatible mirror for updates.

IMPROVEMENTS

A long WeCom answer arrives in full instead of being dropped.

Pages open faster, and runtime usage figures fit on a phone screen.

BUG FIXES

Two tools with the same name running at once no longer mix up their results.

OpenCode 2.x runs again, and custom Oh-My-Pi runtimes are recognized and discovered as before.

Telegram replies once per message, even after a restart or a retry.

Telegram and DingTalk on a self-hosted server accept the secrets you set.

Cancelling sub-tasks tells you which stage was affected and how many.

Comments keep their order, and an Issue link still resolves after you reopen a view.

A local folder resource no longer offers a rename that cannot work.

An image pasted into the editor keeps the format it already had.

Inbox wording about agent activity matches what actually happened.

Tasks on Windows deliver their results without extra steps.

NEW FEATURES

Set the interface language to French, on the web and in the desktop app.

Manage skill labels from the command line, and filter the skills page by label.

Pick a thinking level for your Oh-My-Pi agents.

Edit a comment you already posted from the command line, without overwriting someone else's edit.

Narrow an Issue list down by the status of the project it belongs to.

Edit or pause any one of an Autopilot's schedules, instead of deleting it and starting over.

IMPROVEMENTS

An agent picking a task back up gets straight to work, instead of re-reading the whole Issue and every comment.

Staying active keeps you signed in, instead of logging you out every 30 days.

The Inbox archive scrolls all the way back, and filters and links reach every notification.

Repetitive on-screen explanations are gone, and the Chat list opens at the same width as the Inbox.

BUG FIXES

WeCom no longer drops messages when several go out at once.

Switching an agent's WeCom bot leaves nothing from the old one behind.

A message quoted in a WeCom or DingTalk group reaches the agent with your request.

A DingTalk reply names the agent answering from its first message.

A chat integration that was revoked now shows as disconnected.

Runs on Grok, Pi, Copilot, and Codex no longer fail quietly or leave part of the reply out.

A Cursor session survives a connection timeout, so you can carry on with it.

An outdated OpenCode can no longer fill up your disk.

A Hermes task no longer hangs while wrapping up.

The desktop app finds the command line tools you installed, and CodeBuddy replies show in full.

Runs on Windows follow the tool paths you set.

A private runtime no longer refuses to start over a mismatched owner.

Task cost and usage are recorded in full again.

A commit made in a task uses that task's own Git identity.

A task's final result still reaches you after a reconnect.

Cancelling a child task moves the parent's stage along correctly.

An invitation completed elsewhere no longer stays pending.

The mention picker opens mid-word, and keeps working when nothing matches.

Sidebar help about PR linking and @all no longer misleads you.

Quick Create keeps exactly what you typed.

DingTalk quoted replies, reworked Issue status settings, more accurate run usage, and safer comment deletion

NEW FEATURES

• A DingTalk reply in a group now quotes the message you sent.
• DingTalk marks your request with a reaction while it runs, and again when it is done.
• A DingTalk command confirmation keeps what you typed, and links the Issue it created.
• Issue statuses in Settings are grouped as Unstarted, Started, Done, and Closed.
• Drag a status, built-in ones included, to set its place in boards, lists, and status menus.
• Give a custom status the icon shape you want.
• DeepSeek Harness Desktop is picked up as a runtime on its own, with no manual setup.
• Self-hosted servers send one anonymous daily snapshot of deployment size and run volume; set DO_NOT_TRACK=1 to turn it off.

IMPROVEMENTS

• Archiving a status tells you how many Issues still use it, and takes you to them.
• A DeepSeek Harness runtime that cannot start now says what is missing.
• The command line shows how each Autopilot's last run went.
• Agent steps in an Issue preview what each one was working on.
• A refreshing Issue list says so in the page title, and quick refreshes no longer flicker.
• An agent posting a long final comment finishes sooner.

BUG FIXES

• Deleting a comment keeps the replies people left under it.
• A deleted reply leaves nothing behind, and a link to one lands on the message beside it.
• Sorting an Issue list by status again follows the order you set.
• You are told when an Issue your agent filed is waiting on you.
• Two comments handing the same agent work at once join one run instead of failing.
• A delegated task that failed is picked up again instead of being left behind.
• Usage and cost no longer overstate what Claude, CodeBuddy, Codex, OpenCode, and Qwen runs used.
• A cancelled task no longer counts against an agent's success rate.
• Codex runs no longer mix in a reply meant for another conversation.
• A skill archive packed on Windows imports as a skill you can use.
• Hermes picks up your local skills from the home directory its tasks actually run in.
• WeCom no longer sends a blank message ahead of a file, and says when the file did not go through.
• The Lark binding QR code lasts its full hour, and no longer dies on the first check.
• A desktop app on an older version keeps your custom statuses after the server updates.
• Help in the sidebar stays aligned once you dismiss the Discord link.
• The self-hosted web image picks up the latest OpenSSL security fix.

Comment and description annotations, desktop history menus, new Issue board defaults, and performance improvements

NEW FEATURES

Select text in a comment, add a note, and your reply carries that quote.

Select text in an Issue description to start a new thread from it.

Back and Forward on the desktop app list the pages you visited, so you can jump straight to one.

A new Issue view opens on your latest work, and keeps the sorting you saved.

The transcript now says when a tool's output was not kept in full.

Run history shows who cancelled a task.

Set or clear an agent's conversation starters from the command line.

GPT-6 Astra is available, with its published rates.

Self-hosted servers can run on a Redis cluster or a managed serverless instance.

IMPROVEMENTS

The Inbox opens faster, and long comments no longer slow the list down.

Typing a comment while a task streams stays smooth, even in a long transcript.

Rounded corners now match across the app.

The Inbox keeps what you opened readable in a narrow window.

BUG FIXES

A brief server hiccup no longer ends a task that is still running.

A retry, or checking out the same repository again, keeps what the earlier attempt left behind.

Reopening the Inbox no longer misses notifications that arrived while you were away.

A reply in a Lark group keeps the original message and mentions who asked.

A worker's reply still reaches the Squad leader after the task is claimed.

Older Autopilot triggers run again instead of being skipped.

Usage says when token counts are partial, instead of showing them as complete.

An Antigravity run reports what it used and cost.

Every tool step shows how long it actually took.

A GitHub PR links to an Issue only where it says it should.

Agent runs sit in the Issue at the time they actually happened.

A long dialog keeps its buttons reachable.

Codex stops instead of reusing a setup that failed to prepare.

An Issue no longer shows an update when nothing changed.

The web app picks up the latest security fix.

Agent runs inside comment threads, an Issue thread outline, more reliable channel replies, and steadier search in large workspaces

NEW FEATURES

An agent's queue position, run status, steps, and logs now show up inside the comment thread.

Add more instructions in the same thread and they join the run already going.

Separate comment threads queue on their own, so instructions never land in another thread's run.

The outline on the right lists every thread in an Issue, who is in it, and what is done.

IMPROVEMENTS

Search holds up in large workspaces instead of timing out partway.

A long thread title stays on one line in the outline.

BUG FIXES

Cancelling a Codex run keeps what that run already used and cost.

A runtime whose first session failed no longer stays stuck.

Cursor's background commands are no longer mistaken for idle, and they are cleaned up when the task ends.

A Telegram reply no longer arrives twice while the placeholder is still sending.

DingTalk keeps the quote you picked, and /new or /clear no longer swallow what you were typing.

Lark tells you in the group when it cannot send the binding card to your DM.

An Issue created from a channel now links straight to it.

Replying to an agent's thread stays with that agent, with no spare task left over.

Pressing Esc after opening logs with the mouse leaves nothing behind.

A quick-created Issue no longer shows activity that has nothing to do with it.

Tool output previews no longer cut characters in half.

Hitting the request limit says so, instead of asking you to sign in again.

The command line reports which page of Issues you actually got.

Reordering Issues no longer drops one in the wrong place.

The Inbox no longer shows two sidebar toggles side by side.

A reorganized sidebar, a regrouped Settings page, faster search in large workspaces, and Autopilot manual runs

NEW FEATURES

The sidebar now groups your work and your AI team, with Analytics and Settings pinned at the bottom.

Pinned items past the first five stay tucked away until you expand them.

Settings is grouped by purpose: Personal, Workspace, Issue configuration, Connections & extensions, and Desktop app.

Every integration sits on one page, with its connection status shown up front.

Issue and chat preferences now live inside Preferences.

Custom properties come back with their names from the command line, not just ids.

See how much each agent run used and cost from the command line.

Pick the Issue fields you want from the command line and get a shorter response.

IMPROVEMENTS

Agent, skill, and runtime detail pages stay aligned on wide screens.

Issue search is faster in large workspaces, and multi-word searches no longer stall.

The Inbox list is narrower by default, leaving more room for what you opened.

Self-hosted servers can set how much memory search is allowed to use.

BUG FIXES

An agent can start an Autopilot run for you when you could start it yourself.

Every Autopilot change is judged by the permissions of the person behind it.

Autopilot webhook credentials no longer show up in live updates.

A Pi task whose working folder is gone starts a fresh session instead of failing over and over.

A new chat from a channel takes its title from what you just asked for.

Archiving an agent clears its chats, so Slack no longer shows them as still working.

Members can create tasks with agents on runtimes they can't see.

Tab and Enter in a nested list only affect the level you are on.

Codex explains how to fix the setting behind a failing conversation instead of retrying silently.

MCP server renames and connection swaps, steadier task starts during upgrades, cleaner sign-out on expiry, and accurate shared-agent status

NEW FEATURES

Rename an MCP server without losing its connection details or the agents using it.

Swapping in a new connection is now a separate, deliberate step.

Saving an MCP server points you straight at the field that needs fixing.

IMPROVEMENTS

Each agent execution is called a run everywhere, from the apps to the docs.

Search results come back faster.

BUG FIXES

Tasks still start while your local app and the server are on different versions.

An expired session takes you back to sign-in, with the previous account's data cleared.

Being offline no longer signs you out.

Agents shared from someone else's runtime show as available instead of offline.

Checking out a repository you have used before no longer fails.

Tab-switching shortcuts, server performance, and Autopilot identity authorization

NEW FEATURES

Refresh the model list yourself and get what your runtime supports right now.

WeCom brings a failed run back to the chat you started it from.

Switch desktop tabs with number shortcuts.

Filter and sort Issues by custom properties from the command line.

Find Issues where a custom property has no value yet.

IMPROVEMENTS

Assigning an Issue takes fewer steps, and you can still hold off starting the run.

Self-hosted servers do less background work to keep runtimes connected.

Deleting a runtime takes effect right away, with no stale connection left behind.

BUG FIXES

Replying to a squad leader stays with that leader, keeping the briefing, folder, and session.

Long Issue identifiers no longer cover the title in lists and tables.

Sign-in, error, and not-found pages follow the language you picked.

Signing in still works after you switch languages part way through.

Hermes keeps the model provider you picked instead of switching back on its own.

OpenClaw's managed MCP servers start, and your own MCP settings stay untouched.

Scheduled and webhook Autopilot runs use the permissions of whoever created the trigger.

An older trigger whose creator can't be confirmed stops instead of running.

Members who left the workspace keep their name in activity history.

Claude Code model discovery, property filter operators, session continuity, and steadier providers

NEW FEATURES

The model picker shows only the Claude Code models your machine can actually run.

Models you can't use are called out up front, not after a run fails.

Fable 5.1 is available, with its pricing already in place.

Filter Issues by what a custom text, number, date, or URL property contains.

Number and date properties can also be filtered by range.

See which agent runs are still active from the command line, across an Issue and its sub-Issues.

Custom status colors show up on Issue links, hover cards, and editor suggestions in chat.

IMPROVEMENTS

Recovering a handed-off task that failed no longer slows down everything else running.

Japanese and Korean read consistently across the product and the docs.

Shortened text keeps emoji intact instead of cutting them in half.

BUG FIXES

Editing an Issue description saves cleanly, with no false conflict prompts.

A task that fails while preparing says so, instead of blaming the model or the agent.

Task logs show the full task ID, so parallel runs are easy to tell apart.

A replacement task waits for the previous one to release the working folder.

Every task in the same session keeps working on the same branch.

Repositories you already checked out follow your Co-authored-by setting right away.

Kimi reports provider errors instead of finishing with an empty reply.

Pi sessions run on Windows.

Codex can be set to Standard speed on purpose.

Background cleanup only removes a working folder that belongs to the task.

Nothing is quietly dropped from saved data when unusual characters are cleaned up.

Long Markdown lists on iOS no longer overlap while you scroll.

Images in task details load without flickering.

Dates and amounts follow the language you picked.

The daemon controls in the desktop app are fully translated.