oneship

Version log

Desktop release history for the macOS preview. The latest release is at the top.

1.0.0 June 18, 2026

First production desktop release

  • Ships the first production desktop package against the prod server and prod download feed instead of the dev channel.
  • Includes the latest 0.1.11 workspace refresh, with clearer workspace quickstart entry points, redesigned composer attachment controls, and tighter onboarding layout polish.
  • Adds production Runway deployment support for the server, browser service, and website so the desktop app can rely on the same production infrastructure used by the launch site.
  • Prepares the production website entry point with runtime-origin API routing, prod server proxying, and download redirects through the prod server.
  • Makes web Google auth origins configurable so production login can safely hand sessions back to app.oneship.space while dev keeps its existing domain.
0.1.11 June 18, 2026

Latest main workspace refresh

  • Ships from the latest main snapshot and includes the 0.1.10 desktop update payload.
  • Refreshes the workspace home quickstart so new workspaces surface clearer next steps, member activity, and entry points for projects, channels, and tasks.
  • Redesigns the composer attachment toolbar and attachment chips so file, folder, and inline attachment actions are easier to scan without crowding the message area.
  • Refines onboarding and composer layout details across desktop widths, including tighter empty-state and quickstart copy.
  • Adds production Runway environment declarations for both oneship-server and oneship-browser so the browser service can be deployed consistently outside dev.
0.1.10 June 18, 2026

Files, workspace polish, and runtime recovery

  • Improves file handling across workspace files and artifact outputs, including file deletion, safer visibility filtering, and hiding sandbox artifacts until they are actually delivered.
  • Refines workspace UI around project empty states, project creation, the workspace skills file browser, and agent detail layout.
  • Improves agent failure recovery and status handling so runtime failures, thought streams, and follow-up work are tracked more reliably.
  • Hardens multi-instance workers with singleton leases, speak-lock fixes, safer timestamp handling, and Tencent IM webhook mirroring improvements.
  • Adds GLM 5.2 to hosted cloud agent model choices and improves Haru employee profile summaries.
0.1.9 June 17, 2026

Agent creation, skill profiles, and process details

  • Simplifies agent creation by removing provider-heavy setup from the first pass and keeping the creation flow focused on the agent identity and model choice.
  • Adds a skill detail view on agent profiles and keeps profile skill lists scoped to skills actually attached to that agent.
  • Improves agent process inspection so process dialogs open on the intended message and show fuller execution details.
  • Fixes sandbox continuation Squad routing so follow-up work returns to the right squad context after artifact task handoffs.
  • Clarifies sandbox task manifest fields and updates hosted agent defaults around GLM-backed models and output language handling.
0.1.8 June 17, 2026

OpenRouter cloud agents and quieter activity

  • Routes hosted cloud agents through the OpenRouter-backed Anthropic-compatible runtime and keeps cloud model choices consistent across agent creation, model settings, and member capability surfaces.
  • Improves agent hiring and integration behavior so provider-specific OpenRouter access is handled through workspace grants instead of exposing unrelated provider setup paths.
  • Expands the activity hub and agent work modal with conversation-aware activity, making active agent work easier to trace from the workspace header.
  • Keeps unpersisted group-agent interim output out of the main chat stream until it is durable, reducing transient duplicate or stale group replies.
  • Treats provider API failures as explicit turn failures and makes Squad no_action directives structurally silent, so failed runs and deliberate silence are recorded without noisy placeholder messages.
0.1.7 June 16, 2026

Workspace activity, Squad turns, and task reliability

  • Refines workspace navigation and task activity surfaces with clearer left-nav states, updated task flow details, and a more polished agent work modal.
  • Adds richer group and avatar presentation, including generated group avatar marks, improved avatar style handling, and better syncing of the current user's avatar in the workspace shell.
  • Improves Squad turn handling so explicit mentions, handoffs, and assistant message metadata are carried through conversations more reliably.
  • Hardens browser control during owner handoff by treating browser-close operations more safely when a new owner takes over a session.
  • Improves sandbox and artifact task failure handling so failed background work is recorded, surfaced, localized, and recoverable more consistently.
0.1.6 June 13, 2026

Default agent DMs and safer DM navigation

  • Opens the default agent DM immediately after creating a workspace, preferring Haru and falling back to Navi when the server returns those default channel ids.
  • Fixes DM navigation while Tencent IM activation is still catching up by falling back to channel hydration when an ensured DM does not yet have an active provider group.
  • Applies the DM activation fallback consistently from the left nav, Direct Messages picker, Contacts page, and member inspector so message buttons navigate to the intended conversation instead of silently doing nothing.
  • Uses in-app navigation and shell refreshes for newly opened DMs to avoid unnecessary full-page reloads after choosing a teammate or agent to message.
  • Refines the workspace switcher around account membership, current-workspace unread badges, and switching state so the workspace menu stays accurate across newly created and locally registered workspaces.
0.1.5 June 13, 2026

Artifact tasks, Haru and Navi, and runtime reliability

  • Adds a unified sandbox artifact task pipeline so agents can queue longer document, media, data, browser, and general artifact work, preserve input files, attach generated outputs, and continue the conversation when a task succeeds, fails, or is cancelled.
  • Improves artifact storage, blob handling, and cleanup around generated files so background artifact work is tracked more reliably across server workers and sandbox runs.
  • Renames the default workspace agents to Haru and Navi, and updates onboarding defaults, product guidance, system skills, and workspace copy to match the new roles.
  • Fixes local daemon launch environments by expanding CLI path detection for Codex, Cursor, and Claude runtimes while keeping daemon-private tokens out of child processes.
  • Refines the desktop runtime chip so packaged sessions show local runtime startup as "Connecting..." and keep legacy/system cloud daemons out of the normal runtime health summary.
  • Hardens Tencent IM sync handling by classifying idempotent success, retryable, and permanent provider errors more precisely so conversations and memberships recover without repeatedly retrying known-bad operations.
0.1.4 June 12, 2026

Login welcome, onboarding guide, and Tencent IM reliability

  • Redesigns the login welcome screen and refreshes the download page so the desktop entry points feel clearer and more polished.
  • Adds a guided first-run onboarding experience with beta visuals, workspace setup, join-invite handling, language selection, theme preference, and avatar style choices.
  • Improves Tencent IM reliability with webhook mirror hardening, sync retry tracking, DB-backed history fallback, and richer message hydration so conversations recover more consistently after missed events or reconnects.
  • Fixes active conversation unread count rebound, chat hover timestamp layout, and broadens Simplified Chinese/static i18n coverage across the web UI.
  • Refines agent and daemon-visible output behavior by deduplicating delivered interim failures, keeping repair-only HAR finalization in the agent voice, making no-action squad leader turns silent, and ordering daemon facts scans by per-job sequence.
0.1.3 June 11, 2026

Cloud browser reliability, Studio Blue, and live chat fixes

  • Improves cloud browser reliability by launching Chromium through CDP, cleaning stale profile locks, and tightening automation readiness so browser sessions can recover more predictably.
  • Fixes browser automation signaling so agent browser actions report success and failure states more consistently.
  • Adds the Studio Blue theme and beta desktop package identity, including separate bundle schemes and update-channel metadata for beta builds.
  • Fixes agent terminal status, message grouping, realtime event cache syncing, and Tencent IM hydration so live conversations stay grouped and up to date across reconnects.
  • Preserves channel rules in hydrated groups and fixes daemon runtime file downloads for unicode filenames.
0.1.2 June 10, 2026

Packaged links, realtime terminal updates, and artifact previews

  • Fixes packaged desktop navigation for dynamic workspace routes, including agent drafts, task detail pages, schedules, and chat links, so the macOS app opens the intended in-app screen instead of falling back to the wrong shell.
  • Adds realtime terminal job completion events for agent work so completed, failed, or cancelled runs can update connected views without waiting for a manual refresh.
  • Improves artifact handling with clearer previews, retry states, and fallback uploads from local daemon runs so generated files are easier to inspect and less likely to be missed.
  • Restores local daemon artifact submission through the tool bridge so local agent jobs can reliably attach generated outputs back to oneship.
  • Polishes the desktop web UI baseline, workspace settings, and avatar settings preview for a more consistent warm-editorial interface across packaged screens.
0.1.1 June 10, 2026

Browser sessions, agent status, and identity polish

  • Reworks cloud browser sessions around explicit viewer/takeover links, scoped human ownership, browser profile routes, and noVNC viewer support so agent browser work is easier to inspect and control.
  • Improves agent run state across chat, composer, task rows, run logs, and conversation previews, including online agent status, fewer group-agent anchor flashes, and clearer terminal side effects after cloud interruptions or failed resumes.
  • Adds richer avatar and identity controls with generated avatar styles, feminine portrait variants, workspace settings consolidation, and onboarding/settings polish.
  • Hardens chat, Tencent IM hydration, unread state, artifact attachment refresh, group delete refresh, and agent status delivery so conversation state stays more consistent after updates and reconnects.
  • Adds local Cursor CLI runtime support and daemon recovery improvements for suppressed finals, interim chat delivery, and runtime Skill lease cleanup.
0.1.0 June 8, 2026

Runtime incident recovery and artifact delivery

  • Stabilizes agent artifact delivery with dedicated reply artifact slots, stronger turn artifact buffering, and Tencent IM overlay recovery so generated files are less likely to disappear, duplicate, or detach from the visible reply.
  • Improves local daemon visible output handling so substantive interim agent messages are preserved and linked to process activity instead of being replaced by a short final follow-up prompt.
  • Hardens local Claude and Codex runtime failure handling with transient network error classification, retries, and teammate-facing fallback copy that keeps raw CLI failure details in diagnostics instead of chat.
  • Adds shared language-continuity handling through account locale and runtime prompts so visible agent replies and daemon failure fallbacks better match the teammate or channel language.
0.0.99 June 7, 2026

Squad runtime reliability and chat polish

  • Improves Squad and daemon runtime reliability so active jobs claim, release, and recover work more consistently across conversation scopes, reducing stuck runs and duplicate follow-up execution.
  • Hardens local and cloud agent execution around browser bridge startup, tool progress, Codex CLI runtime output, and runtime Skill leases, making visible activity and final replies easier to recover after transient failures.
  • Updates chat rendering so markdown file path links and HAR draft links display as clickable markdown, while long chat bubbles wrap more reliably.
  • Fixes workspace navigation details for DM profile cards and agent-led group leader labels so people and team context route to the expected views.
0.0.98 June 5, 2026

Visible agent delivery and structured mentions

  • Adds shared visible-delivery finalization so agent runs are less likely to end without a user-visible answer, including repair paths for missing Squad activity, HAR draft links, and artifact submission gaps.
  • Strengthens artifact delivery with an explicit retry protocol and guarded delivery claims, reducing cases where an artifact appears delivered before upload or attachment recovery is actually safe.
  • Improves HAR and PMO runtime readiness by exposing the expected core tools by default, adding capability blocker/no-progress guidance, and making effective agent profiles easier to inspect.
  • Updates chat composer mentions to emit structured agent mentions scoped to channel members, while preserving group mentions during membership loading and removing text-only routing fallbacks.
  • Hardens Squad handoff and activity repair with deterministic fallback behavior, stricter validation, and quieter correction delivery when repair activity is internal.
0.0.97 June 5, 2026

IM navigation and HAR hiring fixes

  • Tightens Tencent IM workspace navigation so inactive ensured conversations are not opened, hydrated rows stay scoped to active groups, and the active channel unread badge clears when the selected conversation is read.
  • Improves chat readability and navigation by grouping adjacent agent transcript bubbles and applying a desktop link policy that keeps app-safe links in context while sending external web links through the system browser.
  • Refines HAR agent hiring prompts and draft schema output, and fixes proposal approval attribution so created agent employees are linked to the approving human identity.
  • Dispatches mirrored backfilled messages through the normal reconciliation path, improving recovery for messages that arrive after the initial mirror event.
0.0.96 June 5, 2026

Scoped workspace conversations

  • Keeps the workspace left navigation scoped to Tencent conversations that have current OneShip channel metadata, so cached SDK groups from other workspaces no longer appear as stray rows.
  • Preserves the empty and syncing states while workspace conversation metadata hydrates, avoiding accidental navigation or hydration requests for unmapped SDK conversations.
0.0.95 June 5, 2026

Message ordering and proposal approval cleanup

  • Stabilizes chat timelines by reserving display sequence numbers for conversation messages and applying the same cursor ordering on server reads and rendered channel streams.
  • Makes Tencent IM message projection more recoverable: failed pending projections can be reclaimed for retry, while overlay and mirror reconciliation avoid conflicting duplicate chat rows.
  • Simplifies HAR agent hiring approvals by removing free-form confirmation text parsing and keeping creation behind explicit proposal actions, with runtime scope preserved for agent tools.
0.0.94 June 5, 2026

Visible outputs and scoped agent collaboration

  • Adds an agent visible output delivery ledger so interim and final agent replies are linked to chat delivery, deduplicated through Tencent IM mirror events, and recoverable when send intents need retry or reconciliation.
  • Introduces collaboration scope handling for channel-based agent teams, including channel roster prompts, agent-led group routing, and delegation limits that keep follow-up work inside the active scope and assigned task participants.
  • Adds fileRef-based agent file reading so agents can safely inspect attached text, PDF, DOCX, and XLSX files through prepared runtime inputs instead of relying on raw attachment context alone.
0.0.93 June 4, 2026

Agent draft review flow

  • Adds a focused agent draft detail page so HAR-generated candidates can be reviewed with prompt, capability, auth, runtime, and Skill bundle details before creation.
  • Simplifies the HAR agent draft review flow by saving and revising drafts through chat-driven proposal actions instead of repeatedly posting fragile proposal cards.
  • Routes agent draft links correctly inside the workspace shell so draft review pages stay in the desktop workspace frame and remain covered by route-param tests.
0.0.92 June 3, 2026

Runtime tools and agent card improvements

  • Improves agent runtime tools by moving local files, browser, tasks, work activity, and visual understanding through clearer domain tool groups with stronger normalization and error handling.
  • Enhances HAR and agent proposal cards by showing skill bundle details and normalizing profile tools so proposed agents are easier to evaluate before adding them.
  • Hardens active agent run handling with runaway guards, interim reply persistence, artifact attachment progress, and better chat-stream status replay for completed runtime activity.
0.0.91 June 3, 2026

Runtime media, artifacts, and cloud reply fixes

  • Fixes failed partial cloud replies and improves cloud-worker locking so interrupted or partial agent responses can recover with less risk of losing the active run state.
  • Expands agent runtime media, auth, and artifact handling with visual understanding support, authenticated media access, safer artifact storage, and stronger remote tool checks.
  • Merges work history search into work activity and tightens chat composer status spacing so recent activity and active-agent status stay easier to scan.
0.0.90 June 2, 2026

Agent cards, recovery context, and status polish

  • Improves HAR and agent proposal name cards by preserving profile capabilities and presenting agent context more consistently across server and web surfaces.
  • Adds richer task context and truncation metadata to agent recovery snapshots while unifying runtime prompt assembly across Claude Code, Codex CLI, and Anthropic SDK runs.
  • Refines chat process and composer status behavior so active agent state stays lighter, more stable, and out of the message stream while follow-up process status updates behave correctly.
0.0.89 June 2, 2026

Parallel cloud sessions and process detail polish

  • Improves the chat process detail dialog and folder binding display so agent progress, process context, and linked folders are easier to inspect from the conversation.
  • Enables parallel cloud session handling so cloud agents can continue independent work without unnecessarily serializing unrelated follow-ups.
  • Hardens Stage2 follow-up enqueue coalescing and cloud-worker session recovery, reducing duplicate follow-up jobs and race-prone reply scheduling.
0.0.88 June 2, 2026

Runtime resilience and process status polish

  • Adds a public Agent Hiring marketing-site entry point and fixes Google OAuth callback and handoff handling so browser-based authorization returns to oneship more reliably.
  • Refines group and task creation flows, plus chat process status presentation, so task setup and agent run progress are easier to scan from the workspace.
  • Hardens agent runtime recovery around scoped writable sessions, final reply publishing, visible failure messages, stale Skill leases, and cloud runtime context restoration.
0.0.87 June 1, 2026

Agent Teams and integration auth

  • Adds Agent Teams and multi-agent task ownership so tasks can be assigned to agent groups, surfaced in task detail views, and routed with clearer team/member ownership.
  • Adds the chat-to-agent hiring page and GitHub integration authorization flow, including server-side integration storage, Composio-backed auth handoff, and runtime env injection for connected agents.
  • Hardens agent workflows around HAR proposal cards, browser approval output, and task follow-up schedule recovery so visible chat output and task status stay aligned with the underlying tool results.
0.0.86 June 1, 2026

HAR proposal cards and Squad routing fixes

  • Keeps HAR agent proposal cards as the visible review request when a card is sent, pending, or already exists, avoiding redundant follow-up text after the interactive name card appears.
  • Fixes Squad leader activity handling by stripping hidden routing directives out of visible replies, accepting compatible legacy directives, and showing a clear stop message when a leader reply is missing required activity.
  • Returns agent-authored chat text to the standard compact bubble treatment so agent replies stay visually consistent with the rest of the conversation.
0.0.85 June 1, 2026

HAR agent review flow improvements

  • Expands HAR agent name cards into a fuller review experience with detail tabs for overview, capabilities, data/auth, brain and tools, and workspace Skills, making agent drafts easier to inspect before hiring.
  • Hardens HAR agent creation proposals by creating or reusing a single active proposal for the prepared draft, returning a stable proposalId, and letting text confirmation approve that proposal without resending the full draft.
  • Improves HAR creation guidance so agents ask humans for plain text confirmation while keeping backend identifiers and workflow details out of the visible chat.
0.0.84 May 31, 2026

HAR system Skill startup fix

  • Fixes default HAR system Skill startup validation by removing copyable fake supporting-file paths from Skill-writing guidance, so HAR can load its built-in agent builder Skill without being blocked by missing example files.
0.0.83 May 31, 2026

HAR name cards and steadier agent setup

  • Adds interactive HAR agent name cards in chat, giving agent drafts a visible profile with Hire, Modify, and Reject actions plus readable skills, brain, tools, env, and workspace skill previews.
  • Hardens HAR agent creation around shared human confirmation checks, proposal state handling, workspace auth key reuse, and partial-provisioning recovery so employee creation is less likely to leave ambiguous or unsafe state.
  • Keeps chat attachments steadier during refreshes by preserving local pending attachment ids, files, and media preview URLs when server messages replace local optimistic messages.
  • Fixes packaged daemon MCP bridge startup by switching the stdio bridge to runtime-resolvable MCP SDK ESM subpath imports.
  • Repairs the public homepage CTAs so changelog, docs, and macOS download links point at real destinations, and adds a static docs landing page.
0.0.82 May 30, 2026

Download page refresh and runtime import hardening

  • Reworks the public download page into a fuller oneship landing page with an interactive hero, capability sections, install section, and clearer product positioning while removing the stale hardcoded desktop version.
  • Hardens the contracts package ESM import surface so daemon and desktop code can load public contract subpaths directly in raw Node ESM, reducing runtime import failures from extensionless internal imports.
  • Improves Claude runtime guidance around Oneship MCP startup: agents are told not to treat early ToolSearch misses as proof that tools are unavailable, and to call canonical mcp__oneship__... tools directly when the runtime spec lists them.
  • Clarifies DM operating policy so tool-backed workspace administration can still happen from a direct message, while visible teammate participation still requires an explicit routing mechanism.
0.0.81 May 29, 2026

Daemon queue and agent reply fixes

  • Prevents a running daemon job from blocking unrelated queued jobs: the local daemon now reports active conversation scopes while claiming work, so same-thread work stays serialized without starving other conversations.
  • Serializes per-agent runtime home preparation and strips internal Squad activity directives from visible agent process output, reducing runtime setup races and noisy hidden coordination text in chat progress.
  • Keeps Tencent IM agent job start anchors and final replies as distinct stable messages, so the "working" marker and final agent reply no longer collide when they share the same daemon job owner.
0.0.80 May 29, 2026

Realtime agent replay and steadier job recovery

  • Restores agent process history more reliably after reloads or reconnects by replaying missing agent events, caching per-job event streams locally, and using start/end anchors so progress rows stay attached to the right job.
  • Moves the desktop realtime connection to the user-scoped realtime feed, refreshes scopes when workspace or channel membership changes, and retries transient authenticated GET failures so the app recovers more cleanly from brief network, TLS, or DNS interruptions.
  • Hardens daemon job completion and cancellation paths: stopped, stale, failed, and reply-less runs now post terminal agent messages or anchors and release runtime leases without leaving jobs appearing active.
  • Makes artifact, reply, and Tencent IM side effects durable and idempotent, recording retryable failures and fencing sends so transient provider errors are retried without duplicate or ambiguous agent messages.
0.0.79 May 28, 2026

Artifact bundles and a clearer Agent Work modal

  • Replaces the compact agent work inspector with a job-first Agent Work modal that groups active, attention-needed, and recent jobs, and lets teammates expand a job for transcript and tool progress.
  • Improves artifact delivery for agent-created files: shell scripts are accepted, richer code/data file types are classified correctly, and multi-file deliverables written as folders are uploaded as zip bundles.
  • Hardens artifact upload failures so unsupported or oversized individual artifacts are reported as skipped warnings instead of failing the whole agent job.
  • Clarifies Oneship cloud browser approval and tool copy so browser requests are described as cloud browser access, not local desktop browser control.
0.0.78 May 28, 2026

Clearer task history and steadier agent activity

  • Redesigns the to-do detail page around a clearer thread and work history view, with richer agent reply summaries, hidden process counts, warning counts, and expandable run history.
  • Stabilizes agent work activity summaries with paginated activity reads, safer payload previews, clearer activity counts, and more reliable status/history responses for PMO and HAR workflows.
  • Shows the failing file path when local skill scan or import requests error, making broken local skill definitions easier to find and fix.
0.0.77 May 28, 2026

Agent work status and cleaner PMO/HAR tools

  • Adds a live agent work status entry in the workspace header, with an inspector panel for active agents, queued or running to-dos, review counts, and the selected agent's activity log.
  • Normalizes Oneship tool and system skill names for PMO/HAR workflows, including work history/status reporting, self capability changes, and HAR-managed agent capability administration.
  • Keeps Squad leader no_action turns visible with a concise reply and keeps process-only no-action traces anchored to the original trigger message instead of merging into the next human message.
0.0.76 May 27, 2026

Document-style agent replies and stronger history recovery

  • Improves local Claude history recovery by scanning the correct Claude projects directory, preserving runtime session metadata with stable project roots, and guiding agents to call Oneship history tools directly when they are available.
  • Renders agent-authored text as document-style content instead of compact chat bubbles, making longer drafts, reports, and setup notes easier to read while keeping human messages in the existing bubble treatment.
0.0.75 May 27, 2026

More resilient local runtime handoffs

  • Starts a fresh local Claude Code session when available runtime tools change, while attaching recent visible conversation context so agents can recover prior work without stale tool state.
  • Keeps in-progress agent process output attached while switching snapshots, avoiding a detached or missing process trace when the anchor message changes.
  • Keeps the workspace shell visible if a refresh request fails and cached or current workspace data is still available.
0.0.74 May 27, 2026

Cleaner HAR replies and steadier run status

  • Keeps HAR replies focused on product language by hiding internal tool names, action names, question IDs, and backend workflow details.
  • Fixes the packaged desktop local MCP bridge so local runtime tools inherit the Electron runtime environment they need.
  • Refreshes the run log status glyph and keeps run status current when agent activity finishes or messages arrive.
0.0.73 May 27, 2026

Faster local runtimes and clearer agent traces

  • Lets owned local runtimes run multiple agent jobs at once, with inspector controls for a 1-10 job concurrency limit and live active-job counts.
  • Improves agent run transcripts with compact branded status glyphs, grouped tool activity, and expandable process history for easier scanning.
  • Makes HAR-created agent setup text-first and binds human confirmation replies to the recorded draft before creating the agent.
  • Tightens desktop release operations so public version notes are prepared from repository diffs before the update feed is activated.
0.0.72 May 27, 2026

Durable workspace file and artifact storage

  • Stores workspace artifacts and chat file uploads in durable object storage on Runway-backed deployments.
  • Keeps existing artifact records from being marked removed when the server container restarts and local disk is replaced.
  • Lets artifact previews, downloads, and agent read_artifact use the same storage path as newly submitted files.
0.0.71 May 27, 2026

Development-channel update checks and version log links

  • Re-enables auto-update checks for packaged development-channel desktop builds while keeping local source runs disabled.
  • Points desktop Version log links at the public release notes page instead of the dev update feed URL.
  • Documents the recent development-channel desktop releases in the static website timeline.
0.0.70 May 27, 2026

Runtime status visibility without optimistic response text

  • Removes optimistic "agent is responding" header text and keeps live work visibility in the runtime status panel.
  • Shows runtime system Skills in agent settings so PMO/HAR built-in capabilities are visible without turning them into editable workspace Skills.
  • Tightens the chat header status behavior across direct messages, normal groups, and squad groups.
0.0.69 May 27, 2026

Separate development desktop channel

  • Adds a separately signed development desktop channel with its own app identity, user data directory, deep-link scheme, and dev server update feed.
  • Moves desktop artifact publishing to the server-managed release pipeline so the updater follows the configured current desktop release tag.
  • Keeps release and development package versions in separate version files for safer preview builds.
0.0.67 May 27, 2026

Local runtime network access and leaner agent tools

  • Allows trusted local Codex and Claude runtimes to run with full local access by default, so workspace-provided API keys can reach external services such as OpenRouter from desktop agent jobs.
  • Adds progressive tool slimming for agents, selecting tool groups from workspace context and task intent instead of loading the full tool surface into every run.
  • Improves workspace auth schema startup handling and refreshes Runway deployment configuration for the website and browser service.
0.0.66 May 26, 2026

macOS update restart hotfix

  • Fixes the macOS update restart handoff so clicking Restart no longer quits the app before Squirrel.Mac has finished taking over the downloaded update.
  • Shuts down local daemon and realtime processes before handing control to the updater, then lets the native updater own the final quit and relaunch.
  • Keeps the updater from falling back to a plain app quit while it is still waiting for the native macOS install handoff.
0.0.65 May 26, 2026

Workspace Auth, admins, and realtime runtime updates

  • Adds a workspace Auth tab for shared encrypted environment variables, so humans and agents in the same workspace can use configured keys such as Gemini API credentials without copying them into each task.
  • Adds workspace admin management in Settings. Owners and admins can add, edit, reveal, and delete Auth secrets; regular members see secret names and masked values only.
  • Lets workspace admins add or remove other admins while preserving owner protections and keeping agent members out of the human admin role.
  • Injects workspace Auth environment variables into both local daemon jobs and cloud worker jobs, with failure handling that marks jobs clearly when secrets cannot be loaded.
  • Moves workspace realtime traffic onto a WebSocket-based fanout layer so channels, task activity, and desktop shell updates stay synchronized without relying on the older SSE stream path.
  • Improves agent runtime behavior with background task support, better workspace-aware read/search tools, and local runtime handling for Anthropic SDK jobs that need workspace-provided credentials.
  • Adds the new server-backed desktop release upload path, so the public download feed can serve stable macOS URLs from oneship storage while keeping auto-update metadata on the generic updater endpoint.
0.0.64 May 26, 2026

Cloud browser and local runtime bridge

  • Adds the first cloud browser tool path with persistent browser profiles, approval cards, viewer takeover controls, and server-side audit redaction for browser actions.
  • Lets local daemon jobs receive job-scoped browser bridge configuration so local Claude Code and Codex CLI agents can use server-backed browser tools instead of a local-only browser.
  • Adds browser access controls in the desktop UI: approve or deny browser access and risky browser actions from chat, revoke agent browser grants, and clear your cloud browser profile from Me.
  • Runs the cloud browser as headed Chromium on a managed virtual display, with readiness checks and cleanup for failed launches.
  • Resolves Claude Code and Codex CLI binaries from the desktop app install before falling back to system paths, avoiding stale or guessed local runtime targets.
  • Aligns preset agent tool catalogs so runtime switching keeps the expected server-routed tool set.
0.0.63 May 26, 2026

Local agent runtimes and HAR creation

  • Lets owners move existing agents between cloud and local computers from the Contacts Runtime tab, including Claude Code and Codex CLI local runtime targets.
  • Keeps agent instructions, server-routed tools, and Skills server-authored so PMO, HAR, and user-created agents keep the same behavior when their runtime driver changes.
  • Wires Claude Code local runtime jobs through the Oneship MCP bridge so server tools can be used from local execution without losing the server as the source of truth.
  • Adds HAR agent creation proposal cards earlier in the conversation and preserves HAR-created agents' human owner in the Contacts roster.
  • Improves cloud-agent tool error handling so missing or failed tool results do not trigger repeated error loops.
  • Adds richer daemon claim logs with job id, agent, runtime, adapter runtime, model, and computer id to make cloud/local routing easy to verify.
0.0.62 May 26, 2026

Update prompts and agent run status

  • Reopens the desktop update prompt when you manually check for updates and a newer version is available, even if that same version prompt was previously dismissed.
  • Keeps quiet long-running agent jobs labeled as running in the run log instead of marking normal non-streaming work as possibly stalled after only a short gap in events.
  • Improves PMO channel inventory visibility and channel overview output so agents can reason from the workspace channels they are allowed to read.
0.0.61 May 26, 2026

Delegated process placement

  • Fixes Squad delegated process placement so the delegated agent's running PROCESS appears under that agent instead of inside the Squad leader's previous message.
  • Keeps chat timeline attribution aligned with the run log when a Squad leader delegates work to a coder, reviewer, or other member.
0.0.60 May 26, 2026

Squad routing and delegation recovery

  • Requires Squad leader replies to include explicit activity routing, separating real no-action decisions from missing delegation metadata.
  • Adds one-shot correction recovery for leader replies that omit Squad activity, so delegated follow-up work can still wake the intended member without posting duplicate chat noise.
  • Applies the same Squad activity gate to to-do-driven Squad assignments, keeping group and task workflows aligned.
  • Improves member prompts so agents sign off cleanly when a mention is informational instead of claiming unrelated work.
0.0.59 May 25, 2026

HAR proposal cards and run log visibility

  • Fixes HAR agent employee proposal cards in the main conversation stream so editable card fields and the Create button render instead of fallback text.
  • Preserves proposal metadata through Tencent message mirroring so newly generated HAR proposal messages hydrate as cards after sync.
  • Adds a group run log panel for inspecting recent agent jobs, including queued, running, completed, failed, and possibly stuck runs.
  • Refreshes the run log while it is open so long-running agent activity is easier to diagnose from the channel header.
0.0.58 May 25, 2026

Editable HAR agent creation and local runtimes

  • Adds editable HAR agent employee proposal cards so agent creation starts from a reviewed schema instead of chat-only confirmation.
  • Lets HAR revise a pending proposal when humans continue with new requirements, with version and draft-hash checks that reject stale cards.
  • Adds local runtime targeting for HAR-created agents when an online local computer advertises Codex CLI, Claude Code, or Claude Agent SDK.
  • Keeps cloud execution as the default when no usable local runtime is detected.
  • Shows detected local runtimes in HAR computer listings and in the creation card runtime selector.
  • Supports creating and attaching folder-style workspace Skill bundles during agent creation, including supporting files and scripts.
0.0.57 May 24, 2026

To-do List threads and project-aware tasks

  • Renames Tasks to To-do List across the workspace navigation, tool panes, agent detail surfaces, and loading states.
  • Adds task detail threads so completed agent replies appear as comments and humans can reply in-context to continue the same task.
  • Lets task comments mention humans and agents, passing visible thread context, prior final replies, and mentioned participants into follow-up runs.
  • Adds optional Project binding when creating tasks and schedules, so task runs can resolve the intended Project repository context without relying only on channel context.
  • Shows task metadata for assigner, assignee, bound Project, execution history, and token usage on the task detail page.
  • Improves Project creation with icon/avatar controls, project notes, social account fields, and a fuller Projects workspace surface.
0.0.56 May 24, 2026

Workspace Projects and project-scoped agent work

  • Adds Workspace Projects for organizing repository context, notes, owners, and related work away from the old workspace-goal surface.
  • Adds project-bound channels so group/public conversations can carry project context into agent prompts and resource resolution.
  • Adds project-aware repository defaults, keeping unassigned legacy repositories separate while hiding repositories from archived projects.
  • Adds list_projects and read_project tools so agents can inspect project names, descriptions, owners, repositories, social accounts, and notes.
  • Adds Projects UI in the workspace rail, project inspector details, create/archive flows, and channel project selection.
  • Improves Contacts with Squad directory/detail views, a Create Squad action, member instruction/Skill summaries, and direct links to edit agents.
  • Refines agent creation and Skill management with a two-column Create Agent modal, all-Skills selector, searchable Skill import dialog, stable import height, and faster selected Skill imports.
  • Restores the sidebar loading animation, removes duplicate mention-style notifications where the conversation list already carries unread state, and refreshes app/icon handling.
0.0.55 May 23, 2026

Squad briefing, delegation, and chat sync fixes

  • Gives Squad leaders and delegates the current channel roster and Squad briefing directly in local daemon jobs, so they can answer who is in the group without probing local files, sessions, tokens, app storage, or databases.
  • Restores reliable Squad delegation from leader replies by carrying structured mention ids through Tencent IM messages, so @coder and other Squad teammates receive delegated jobs.
  • Keeps Squad leaders focused on coordination by separating the full group membership roster from the agent teammate roster used for delegation.
  • Adds stop controls for running agent work so local agent runs can be canceled from Oneship.
  • Stabilizes chat conversation sync and scrolling so active conversations keep their message state while Tencent IM history and live events arrive.
  • Includes the earlier Squad roster context fixes from the post-0.0.54 line, covering both cloud worker and local daemon execution paths.
0.0.54 May 22, 2026

Local Skill import and agent reply grouping fixes

  • Fixes local Skill import so a completed local scan stays importable long enough to select Skills instead of expiring with local_skill_scan_not_ready.
  • Fixes the Import Skill dialog Close action in the desktop app by keeping the modal controls outside the Electron titlebar drag region.
  • Keeps separate agent replies and process cards tied to their original jobs, so later agent responses no longer collapse older job output into the newest reply bubble.
  • Preserves agent reply metadata from Tencent IM history overlays and includes the original trigger message id on new agent reply metadata for stable process placement.
0.0.53 May 22, 2026

Squad responses and broader local Skill imports

  • Shows squad leader response state immediately after sending to a Squad group, and refreshes active process cards even when the live job event arrives before the current chat view catches it.
  • Keeps Claude Code and Codex CLI agents on the user's default login state while still loading only the Skills assigned to that Oneship agent home.
  • Expands local Skill import scanning to include user agent Skills, Codex superpowers, Codex plugin Skills, and Promptfolio Skills.
  • Fixes the local Skill import dialog close action and adds Select all / Deselect all controls for large scan result sets.
0.0.52 May 22, 2026

IM routing and local Skill import polish

  • Preserves squad leader reply metadata through Tencent IM mirroring, so squad conversations keep their leader and delegate routing context after provider callbacks.
  • Uses the full conversation row width for message previews, with timestamps inline instead of reserving a narrow trailing column.
  • Filters Oneship bootstrap custom messages from live Tencent timelines so new groups do not show [unsupported message type].
  • Keeps the local Skill import dialog usable when many Skills are found by making the result list scroll inside the modal.
0.0.51 May 22, 2026

Correct desktop bundle for IM interactions

  • Rebuilds the desktop package from the current workspace code so the Tencent IM interaction updates from 0.0.50 are actually bundled.
  • Contact rows now open the contact detail view first; starting a direct message is an explicit action from the detail pane.
  • Keeps the 0.0.50 chat improvements for conversation hydration, unread state, pinned and muted conversations, media previews, attachment previews, paste-to-attach, and failed-send retry.
  • Keeps the unified local Skill scan flow with automatic polling.
0.0.50 May 22, 2026

IM chat reliability and unified local Skill scanning

  • Refactors the Tencent IM engine integration for more reliable conversation hydration, unread state, pinned and muted conversations, and desktop badge updates.
  • Improves chat media handling with image/video previews, local attachment previews, paste-to-attach, and retry support for failed sends.
  • Scans supported local agent Skill directories together instead of asking you to choose Claude Code or Codex first.
  • Adds automatic scan polling and a scanning state so completed local scans no longer look stuck on running.
  • Adds daemon scan logging for the number of local Skills found.
0.0.49 May 21, 2026

Local Skill scan fix

  • Fixes local Skill scans so a Claude Code scan ignores Codex Skill directories, and a Codex scan ignores Claude Skill directories.
  • Prevents mixed-runtime local scan results from failing with local_skill_scan_result_invalid.
0.0.48 May 21, 2026

Workspace Skills for desktop agents

  • Adds workspace-level Skill management and agent Skill assignments to the desktop app.
  • Ships the updated bundled daemon protocol required for managed Claude and Codex Skill runtime homes.
  • Includes the production schema repairs needed for workspace Skill imports, snapshots, and runtime leases.
0.0.47 May 21, 2026

Tencent IM readiness and workspace migration polish

  • Improves Tencent Cloud IM startup by falling back when the SDK does not emit the expected ready event.
  • Includes the latest workspace rename compatibility fixes for saved registries, notifications, artifacts, and server schema repair.
  • Keeps local daemon workspace protocol compatibility aligned with the production server.
0.0.46 May 21, 2026

Workspace migration and local agent compatibility

  • Renames the top-level product container to Workspace across the packaged desktop app.
  • Migrates saved desktop workspace registry entries and local daemon registrations so existing connections survive the rename.
  • Updates the bundled local daemon to the new workspace job protocol used by the production server.
  • Includes the production schema migration guard that prevents the server from booting against an unmigrated database.
0.0.45 May 20, 2026

IM routing, task delivery, and desktop polish

  • Ships the Tencent Cloud IM-primary conversation changes in the desktop app.
  • Redesigns task creation into a focused two-column composer with explicit Task or Schedule trigger modes.
  • Adds delivery controls for sending task and schedule results to your own DM, selected contacts, or selected groups.
  • Keeps the task board full-width by removing the low-value secondary task sidebar.
  • Fixes task detail back navigation, restores dragging on the signed desktop window, and opens Google sign-in in the system browser.
0.0.44 May 20, 2026

Tool rail, contacts, and agent console

  • Moved workspace tools into a compact left rail so conversations stay separate from tasks, contacts, files, and devices.
  • Renamed Employees to Contacts, added human/agent filtering, and added invite person plus create agent entry points.
  • Reworked agent details into a full-width editable console with Prompt, Skills, Runtime, Memory, Files, Activity, Environment, and Parameters tabs.
  • Added a Files workspace for shared agent outputs and simplified Tasks back to the production-style dashboard layout.
0.0.43 May 19, 2026

Tencent Cloud IM and chat files

  • Added Tencent Cloud IM-backed chat transport and session setup for new accounts.
  • Added chat file attachment support for IM-backed messages.
  • Includes the 0.0.42 desktop release build fix that keeps the packaged app connected to the production API.
0.0.42 May 19, 2026

Production API build fix

  • Fixed the 0.0.41 desktop package entering mock mode because the bundled web app was built without the production API URL.
  • Restored real account, workspace, member, and runtime data in the desktop app.
  • Kept the Squad coordination fixes from 0.0.41.
0.0.41 May 19, 2026

Squad coordination and channel routing

  • Improved Squad coordination so Squad leaders keep ownership of routing, delegation, and completion checks.
  • New Squads now always use their bound group, and Squad tasks run in that group instead of silently stopping when no channel was selected.
  • Removed the obsolete "create Squad group" option from the Squad creation flow.
  • Cleared web build lint warnings without changing chat runtime behavior.
0.0.40 May 19, 2026

Create flow usability

  • Create Agent now uses a plain model dropdown with the runtime default preselected, so model choices are visible without clearing a search field.
  • Create Agent and Create Squad dialogs now keep their header and footer fixed while the form body scrolls inside the viewport.
  • The dialog close button now has a larger, reliable click target.
0.0.39 May 19, 2026

Restart handoff fix

  • Fixed the macOS updater restart path that could still leave the app on "Restarting oneship" after ShipIt received the install request.
  • The updater now explicitly quits the app if the native Squirrel handoff does not start Electron shutdown, allowing daemon cleanup and installation to proceed.
0.0.38 May 19, 2026

Agent and squad visibility

  • Added Agent and Squad visibility controls for public/private membership workflows.
  • Tightened server-side visibility checks across channels, tasks, inbox rows, artifacts, agent activity, and agent tools.
  • Added production schema support for visibility ownership, privacy constraints, and related indexes/backfills.
0.0.37 May 19, 2026

Reliable update restart

  • Fixed a macOS auto-update handoff where clicking Restart could leave the app stuck on "Restarting oneship" while the native updater waited for the app to quit.
  • The desktop updater now retries the native Squirrel restart handoff if app quit has not started shortly after installation begins.
  • Added focused updater coverage for the stuck restart path.
0.0.36 May 19, 2026

Clearer agent runtime setup

  • Create Agent now shows available computers as explicit local/cloud selection cards instead of hiding them in a dropdown.
  • Switching runtimes in Create Agent now resets the model to that runtime's default, so a model from Claude Code is not accidentally sent to Codex CLI or another runtime.
  • The Create Agent modal now has focused tests for assignable computer visibility and runtime-specific model synchronization.
0.0.35 May 19, 2026

Local runtime ownership and CLI discovery

  • Local computers are now scoped to the signed-in owner, so one teammate's machine and runtime choices no longer appear as another teammate's "My Mac".
  • Local runtime detection now checks PATH, login shell paths, package-manager shims, fnm/nvm/mise/asdf-style installs, which -a, whereis, and known local state folders for Codex, Claude Code, Cursor, Gemini, Kimi, and related CLIs.
  • The Create Agent flow now uses a structured model catalog with searchable model IDs, provider defaults, and runtime-specific discovered models.
  • Agent creation and computer APIs now reject attempts to use another human's local computer while still allowing hosted cloud runtimes.
0.0.34 May 18, 2026

Faster, cleaner app updates

  • Fixed the "Restarting oneship" step hanging for a long time during an app update. The app now shuts the local daemon and its agent CLI processes down cleanly before restarting, instead of waiting on leftover background processes.
0.0.33 May 18, 2026

Stable local daemon and live agent progress

  • Fixed a local daemon reconnect loop where a single transient authorization error could trigger an endless cycle of token rotations and restarts, leaving local agents unable to claim jobs and the agent runtime list stuck on stale data.
  • The local daemon now tolerates brief authorization blips, only refreshes its credentials after repeated failures, and stops with a clear "sign in again" message instead of looping when it truly cannot authenticate.
  • Claude Code agents now stream their progress — tool use and intermediate steps appear live while the agent works, instead of the chat sitting silent until the final reply.
  • Claude Code and Codex CLI agents now share the same execution time limit, so longer local tasks are no longer cut short inconsistently.
0.0.32 May 18, 2026

Agent status and unread consistency

  • Agent employee status now follows the bound runtime computer and active jobs, so stale local daemons no longer leave agents looking online or working.
  • Agent detail pages now show a human owner for each agent when one can be resolved from creation or confirmation history.
  • Sidebar unread dots and the macOS Dock badge now use the same visible unread rules, preventing stale Dock badges when the conversation list is already clear.
  • Local daemon token recovery no longer rotates tokens or restarts the daemon for job-claim authorization errors, reducing disconnect loops.
0.0.31 May 18, 2026

Faster reopen and unread badge

  • The macOS Dock icon now shows the current workspace unread message count and clears it when messages are read or the app exits.
  • Closing the desktop window now hides it instead of destroying it, so reopening from the Dock is immediate instead of rebuilding the whole workspace UI.
  • Workspace screens now render a cached shell snapshot first and refresh in the background, reducing full-screen loading after reopening the app.
  • The app header now uses the OneShip wordmark and removes the extra green status dot before the brand.
  • Fixed a desktop channel read crash that could occur when opening certain channels.
0.0.30 May 18, 2026

Local daemon recovery and CLI discovery

  • Local desktop daemons now recover automatically when their saved machine token becomes stale, rotating the token and restarting after heartbeat or job polling returns 401.
  • The Create Agent runtime picker should refresh with the daemon's latest reported runtimes after recovery instead of falling back to only Anthropic SDK.
  • Local runtime detection now asks the user's login shell for CLI paths, matching how macOS users install Claude Code, Codex CLI, Cursor, Gemini, and other agent CLIs through shell startup files.
  • Codex model detection uses the resolved Codex CLI path, so GUI-launched desktop builds no longer depend on Electron's inherited PATH.
0.0.29 May 18, 2026

Tasks workflow, transcripts, and local runtimes

  • Tasks now have clearer board status colors, timezone-aware scheduling controls, task templates, filtering, sorting, and status facets.
  • Schedule detail pages now support editing, archiving, pause/resume, and toast feedback for task workflow actions.
  • Task detail pages now show run history, transcripts, and live execution events, with safer transcript authorization and task event error handling.
  • The task module now has i18n coverage and skeleton loading states instead of bare loading text.
  • Agents now receive configured workspace repository context when reading workspace overviews.
  • Claude Code and Codex CLI installed in ~/.local/bin, /opt/homebrew/bin, or /usr/local/bin are detected from the desktop app.
  • Runtime child processes still strip Oneship daemon/user tokens from their environment.
0.0.28 May 17, 2026

Agent DM tools and completed process state

  • PMO, HAR, and agent employees can use the proactive messaging tool to DM teammates or post into accessible channels.
  • The runtime permission gate now accepts both SDK MCP tool names and raw MCP tool names, so enabled tools are not denied by name mismatch.
  • Completed or failed agent jobs synthesize a terminal process step when an idle event was missed, preventing stale "正在思考" process labels after a reply appears.
  • The agent tools UI now matches the server catalog for send_message_to.
0.0.27 May 17, 2026

Agent messaging and SDK normalization

  • Agents can now proactively message teammates or accessible channels with send_message_to.
  • Agent-to-member DMs are created automatically when an agent sends a direct message.
  • Claude Agent SDK events now pass through a server-side normalizer before driving process state and chat output.
  • Agent process blocks now receive explicit completion events, reducing stale "thinking" states after a reply appears.
  • The web build now runs with a local Next ESLint configuration, so lint errors are caught during release builds.
0.0.26 May 17, 2026

Process display and version log links

  • Process updates keep each agent job in one process block, avoiding duplicated running or completed states.
  • Completed process blocks stop the breathing highlight but remain expandable, including one-step process history.
  • Older process steps stay folded by default while the latest live step is highlighted.
  • Inline confirmation buttons update immediately inside the agent card after selection.
  • The app and download site link to version logs, and the public download page includes the oneship favicon.
0.0.25 May 17, 2026

Agent tools and inline confirmations

  • Cloud agents and HAR-created agents keep the full tool capabilities they are assigned, including shell tools.
  • Process updates collapse older steps by default and highlight only the latest active step.
  • Agent question confirmations stay inline with the agent reply instead of creating a separate user echo message.
  • Update prompts can show release notes from the desktop update feed.
Back to download