Changelog

​

Fixed

• Keep the instructions your agent needs. Studio rejects workflow-improvement proposals that replace existing instructions with placeholder notes, omit them with an ellipsis, or remove more than half of a longer instruction set; you still review and apply eligible proposals.
• Reach the right next step. Agents can now advance to a permitted workflow step by its unique name, fixing cases where a valid handoff to an escalation step failed because the agent used its name instead of an internal reference.

​

What’s new

• Research current public information. Where public web research is enabled and granted to the agent, it can search and read public HTTPS pages, retaining source evidence when work continues or pauses for a human decision.
• Connect NetSapiens to support work. The new preset covers users, call routing, queues, voicemail, devices, and call history, with supported actions including adding or removing queue operators; configure your service address, credentials, and allowed operations before use.
• Collect missing input before a run. A manual agent task can request required information from a person, who completes the requested fields and selects Save and run to start the work.
• Turn channel work into reusable knowledge. Eligible channel evidence, including agent run reports, can produce channel-scoped knowledge proposals; run-report extraction focuses on reusable guidance, and channel facts always require review before publication. See Org Brain.

​

Improved

• Follow and review the whole case. Agent Tasks and Activity views use clearer labels, while task detail brings progress, tool calls, questions, transcripts, and outputs into Activity, with outcome review under Review.
• Edit workflows more comfortably. An open workflow step has more room for instructions, and its editor adapts to narrower panes while preserving the surrounding workflow context.
• Choose which writes need approval. Organization admins can explicitly allow selected channel endpoints to run without approval under Allowed tools; other writes retain independent exact-request approval, and allowed actions remain recorded and subject to current policy. See worker write policy.

​

Fixed

• Keep connected services available after an approval. Agents can discover every connector they are allowed to call, even in channels with large endpoint lists, reducing unnecessary requests for connection details when a task resumes.

​

Fixed

• Use the remaining organization allowance. Pooled organizations no longer receive an over-cap refusal merely because the portion selected for a member’s or worker’s debit is full while another portion of the same allowance still has room. The debit can continue against available allowance in that window. This does not remove the pool’s limit or resolve every concurrent-worker overshoot case.

​

What’s new

• A workspace for support agents. Create and manage Agents inside shared channels, configure their workflows and connections, and follow runs, human decisions, audit history, and outcomes. The web Work area brings approvals and channel boards together for the support team. See AI support agents.
• Connected action and customer response. Enabled worker connections can read a ticket, propose a service change, execute the exact approved request, verify the result, and submit a customer reply or ticket update. Each write requires its own independent authorized approval. HubSpot’s preset includes customer-visible replies and closed-stage discovery; the expanded Dialpad preset exposes its API v2 operations. See the account-change walkthrough.
• More ways to start and continue work. Eligible enabled organizations can use schedules and webhooks alongside manual runs. Supported worker chaining, durable waits, stage deadlines, and escalation keep a configured support case moving through its decisions.
• Slack questions and approvals. With Chat apps configured and identities linked, your team can answer worker questions, review actions, and receive supported outcome notifications in Slack. Decisions still pass Studio’s role and channel checks.
• Expanded Org Brain access. Eligible manual/Enterprise pooled organizations can activate Org Brain with owner consent; generated answers require separate synthesis consent. Workers can search organization-visible raw evidence, while human members can retrieve their authorized organization and channel sources. See Org Brain.

​

Improved

• Review and improve agent outcomes. Run transcripts, artifacts, request-based report titles, outcome grading, version comparisons, and refinement history make results easier to inspect. Workflow improvements still require a person to apply them. Dream settings separately control shared-fact extraction and Memory consolidation.
• Models and budgets. Added GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna choices under organization policy. Configured worker OpenAI execution and Bedrock fallback accompany clearer caps, budget-override tasks, overdue-stage indicators, and save/readiness feedback.
• Clearer notification destinations. Worker approval, failure, and budget email paths link to the web Work area. Delivery remains subject to the deployment’s email availability; the in-app task remains the place to inspect the decision.
• Managed remote-session preview. Enabled deployments gained clearer waiting and ended states, improved reconnect/cancel behavior, and more reliable frame delivery and session renewal. Windows Control and managed Copilot access are not a general release; see Managed Remote Desktop.

​

Fixed

• Workers retain earlier connector reads after an approval and show the proposed endpoint request body to the reviewer. Schedule dispatch, input-derived run names, unsaved-edit prompts, and daily usage aggregation for OpenAI workers were corrected.
• Org Brain activation can finish empty closed stream segments; empty Memory collections are explained more clearly. Deleted and recreated chats retain the correct source identity, and a Personal chat explicitly shared with the organization can supply organization-visible evidence.
• Full and catch-up syncs retrieve visible conversations more reliably. Channel pickers show joined channels, and runbook steps stop after an account or organization switch. Retired trials no longer leave stale state that blocks manual billing.
• Deleting a chat now scrubs its stored message text and other text-bearing message fields as part of the deletion cascade.

​

Availability and limits

​

What’s new

• Controlled worker early access. The organization owner switch, staged worker editor, manual run form, member Work approvals, channel boards, and desktop waiting-on-you inbox reached the production testing path. The run drawer exposes decisions, transcripts, artifacts, and event-chain verification. At this stage, production workers used manual dispatch; broader triggers and connector writes followed in the next update.
• Org Brain in the desktop. Production desktop builds gained the search and cited-answer tools for activated organizations. Select a Memory citation to inspect the current authorized source, with a clear unavailable state if it was deleted or access changed. Owner activation, separate synthesis consent, source visibility, and the controlled rollout still determined access.
• Durable desktop audit. Organization administrators can inspect, verify, and export hash-linked tool-decision streams. Signed-in desktop events buffered offline upload after reconnection.
• More support connector definitions. Added HubSpot ticket/conversation operations and Dialpad service endpoints. The HubSpot preset includes a public customer reply separately from an internal note. Definitions still need a connected account, scopes, and tool grants.

​

Improved

• From completed work to a procedure. Save a conversation or completed procedure run as a reviewed draft, with derived tools and the run’s final report carried into the authoring flow.
• Easier recall and reading. Saved responses lead with conversation titles and excerpts, keep their position during navigation, and offer Undo after removal. Chat headings, code blocks, and tables are easier to read, and new-chat suggestions appear without shifting the composer or starter cards.
• Human decisions and budget visibility. Worker questions use clearer language, reviewers can give a rejection reason, and per-run/daily spending is visible with a waiting-on-budget state. Dispatch and member work views enforce channel access.
• Desktop connection protection. Local sidecar connections require session-specific authorization and validate their intended host and origin. Browser and connector requests gained stronger destination checks and corporate-proxy support. Update Studio to receive the desktop changes.

​

Fixed

• Long assistant replies are saved in full instead of as an early streaming fragment. Synchronization preserves messages written during a full pull, and conversation/message ownership checks were tightened.
• Deleting an organization also cancels its associated Stripe subscriptions. Worker cancellation and failed-dispatch recovery leave clearer task and run state.

​

Availability and limits

​

Fixed

• Approved commands execute consistently. Streaming chat and plan steps now use the same signed approval path for SSH, Telnet, serial, local shell, and file changes. This fixes commands returning “approval required” after Studio’s permission policy had already allowed them.
• Stop cancels pending approval work. A late approval after Stop no longer proceeds to cache credentials, trust an SSH host key, reconnect, or dispatch a command. Commands also retain their intended order while an earlier approval is pending.

​

What’s new

• Managed Google Workspace and Microsoft 365 connections. Connect supported accounts through Altostrat-managed OAuth, with per-member credentials and additional permissions requested as a workflow needs them.
• Key Chain inventory. Find credentials in a dedicated inventory, inspect which resources use them, sort by usage or last use, and review affected resources before deletion.
• SIP call context in chat. Associated conversations can use bounded, speaker-labelled call transcripts, including follow-up questions after hangup. A new chat can pick up the operator’s active call, and ongoing call controls remain visible across workspace navigation.
• New-chat starters. Suggested workflows help you begin a support task by preparing an editable draft.

​

Improved

• Procedure lists and editors expose missing content, input details, and side-effect risks more clearly, with readable collapsed previews and better keyboard and narrow-window navigation.
• Chat separates the answer from turn activity and displays submitted messages immediately. Sync status distinguishes unresolved changes from active syncing and lets you discard only the change you reviewed.
• Subscription setup can resume bank authentication, gives clearer feedback after successful 3D Secure payment, and lets trial-ineligible customers proceed with paid signup. Desktop billing forms also gained address autocomplete.

​

Fixed

• Saved Hosts and open SSH sessions remain stable during background refresh. New connections wait for a current secure snapshot, and automatic device-identity saves no longer overlap.
• Managed OAuth disconnect and reconnect preserve the correct member credential without creating duplicate rows. A rejected request no longer temporarily blocks unrelated authorized requests.
• Corrected incomplete channel-membership reads and restored AI entitlement for valid AWS Marketplace licenses during the transition from retail billing.

​

Rollout status

​

What’s new

• AWS Marketplace setup. A fulfillment flow links a Marketplace agreement to a Studio organization and synchronizes licensed seat limits. Marketplace usage metering remains disabled.
• Faster artifact handling. Files gained direct downloads, multi-select checkboxes, bulk actions, a flatter list, and a dedicated archive tab.

​

Improved

• Account and team menus make identity and organization switching clearer, distinguish members with duplicate display names, and keep full-page workspace views out of the drawer.

​

Fixed

• Host lists stay visible while the vault starts, and secure credentials recover after a delayed or restarted native agent.
• Key recovery handles organization-switch races and incomplete key listings more reliably, with bounded retries so a stalled recovery does not hold the vault indefinitely.

​

What’s new

• Save responses for later. Keep useful answers in a dedicated saved list and return to the original conversation. Saved responses sync across your Studio workspace and Studio Remote.
• Suggestions based on unfinished work. New-chat suggestions use recent activity, unresolved tasks, and blockers to propose a useful next step. New users also receive starter suggestions based on the tools available in their workspace.
• Secure connector credential requests. When an integration needs a new secret, Studio can open a dedicated credential form instead of asking you to paste it into chat. The form makes private and organization-shared storage explicit.
• Credit-based Luna trials. Eligible new trials use a fixed credit allowance and Luna model access. Trial indicators show remaining credit instead of an artificial renewal countdown.

​

Improved

• Per-conversation usage supports search and sorting, and collapsed chat-sidebar groups keep their saved state.
• Chat waits for its channel policy and required integrations to be ready before offering tools, reducing incomplete startup state and unnecessary retries.
• Work suggestions better distinguish completed work from open follow-ups and offer direct local actions where available.

​

Fixed

• Stabilized Save for later during response streaming, startup, and sync; Studio Remote now receives saved-response changes reliably.
• Improved terminal readiness, reconnection, and workspace-tab persistence so reopening or switching chats does not unnecessarily disrupt the session.
• Preserved host connection settings for older desktop clients during the backend transition, and corrected cancelled-subscription fallback and AI entitlement checks.

​

What’s new

• OpenAI models in Studio. Choose supported OpenAI models alongside Anthropic models. Luna is recommended for fast, everyday work, and usage identifies the provider and model that served each turn.
• Organization model policy. Admins can choose which models members may use, set the organization default, and control whether the model selector is visible.
• A clearer Web Admin workspace. The organization overview, team and channel pages, worker administration, and usage dashboard have clearer navigation, counts, and empty states. The admin assistant can use the organization’s configured context to answer setup questions.
• Rendered mathematics. Chat and generated response panels display supported LaTeX expressions alongside Markdown.

​

Improved

• Copy text, code, tables, plans, and tool results from responses more reliably, including content inside generated panels.
• OpenAI conversations compact long context, recover from transient failures, and handle stop requests more consistently.
• Per-message cost badges are optional and off by default, while detailed usage remains available in the usage views.

​

Fixed

• Prevented plan execution controls from appearing before the response finishes streaming.
• Corrected OpenAI usage attribution and cached-token accounting, and preserved the correct subscription tier when an existing subscriber uses an older price.

​

What’s new

• Studio Remote is available in production. Pair the desktop app with Studio on your phone, continue a conversation, follow live activity, and answer approval requests while the paired desktop remains the execution bridge. The pairing panel includes Android and iPhone download options.
• Connector activity in the status bar. Waiting connector connections appear alongside other live work, making it easier to find an integration that needs attention.
• Navigable usage summaries. Usage overview cards lead directly into the relevant AI, call, transcription, and credit details.

​

Improved

• Faster connector use in chat. Studio gets an integration into the conversation with less setup narration and clearer progress.
• Free low-cost background assistance. Eligible thread suggestions and other bounded Nova Micro side calls no longer consume the member’s paid balance.
• Longer-lived Remote sessions. Regular activity renews an existing phone session, reducing unnecessary re-pairing while inactive sessions still expire.
• More stable long tool runs. Tool configuration and MCP policy carry across continuation turns with fewer unnecessary prompt-cache changes.

​

Fixed

• Improved interrupted-run cost estimates, billing attribution, and connector preload recovery during long conversations.

​

What’s new

• Guided mobile pairing. The desktop connection flow now supports a QR code or short pairing code, reports connected devices clearly, and lets you revoke a phone that should no longer have access.
• Reusable generated artifacts. Files produced during a conversation remain available as reusable work products instead of being tied only to the turn that created them.
• Organization work-product memory. Studio can preserve useful organization-scoped outcomes across chats, with policy and source context separating them from a member’s private operational memory.
• Per-conversation AI cost. Chat details now show the models, tokens, and cost attributed to that conversation.
• Organization-wide context. Admins can write shared guidance once and have it appear in members’ Copilot conversations, plans, and delegated work alongside their channel context.
• Guided partner invitations. A dedicated referral flow takes invited customers through sign-up, organization setup, teammate invitations, and the appropriate Studio download for their operating system.

​

Improved

• Safer credit top-ups. Purchase review now shows the amount, payment method, and required card confirmation before the charge is submitted. Top-up value remains visible after plan credits are exhausted.
• Long-run guardrails. The agent loop now has tool deadlines, a wall-clock boundary, cost preflight, and mid-run spend checks. Completion, cancellation, and recoverable-stop states are preserved in the transcript.
• Clearer chat navigation. Turn navigation sits beside the conversation, completed runs have compact summaries, and the latest-response controls are easier to reach.

​

Fixed

• Concurrent approvals are queued instead of replacing one another, cancelled tools keep a truthful cancelled state, and follow-up prompts survive a conversation remount.
• Enterprise contract activation stops conflicting self-service billing. Invoice delivery and manually recorded payments keep invoice state aligned, and eligible pooled accounts receive a clearer message at their usage ceiling.

​

What’s new

• Sandboxed Studio apps. Generate a focused interactive app inside a dashboard. Apps run in a restricted frame and request named host capabilities rather than inheriting unrestricted workspace access.
• Immutable app revisions. Preview a revision, grant capabilities to its exact content hash, and explicitly promote the version that should be active. A changed revision must earn its own grants.
• Dashboard version controls. Batch snapshots, inline previews, and explicit promotion make it easier to inspect and recover dashboard changes.

​

Improved

• Generated apps compile in packaged desktop builds, start more reliably, recover saved state, and present requested layouts and controls consistently.
• Dashboard panels resize more smoothly and show truthful loading, readiness, failure, and retry states.
• Credit top-ups are limited to the billing owner.

​

Fixed

• Closed revision activation races, stale capability refreshes, app-startup transport failures, parent-visibility gaps, and dashboard polling hydration races.

​

What’s new

• Organization control plane. Administrators can manage member context, integration definitions, per-member or shared credential mode, tool policy, and account-readiness tasks from a coordinated organization surface.
• Status-based chat board. Organize conversations by status and move them through the board without losing their channel or transcript.
• Per-chat model selection. Choose a model for the current conversation instead of changing every chat globally.
• Live browser activity. Active browser sessions appear with the rest of a chat’s running activity.

​

Improved

• Computer Use now prefers direct, OS-aware tools before falling back to broad visual interaction.
• Studio Remote refreshes conversation snapshots, publishes approval changes reactively, and can wake an eligible paired desktop.
• Chat streaming, typing, scrolling, code-workspace routing, and general workbench interaction are more responsive.

​

Fixed

• Tightened organization switching, integration-editor isolation, member-profile updates, shared-credential resolution, browser session reuse, and cloud credential scoping.

​

What’s new

• Integration policy and account checklist. Admins can allow an integration, choose personal or managed shared credentials, set defaults, and see whether members have completed required personal account setup without reading their secrets.
• Layered tool policy. Organization and channel controls now constrain connector, MCP, and first-party tools before they are offered to a conversation.
• Detailed usage views. Usage now breaks AI activity down by model and feature, shows remaining allocation as a percentage, and distinguishes plan value from top-ups.

​

Improved

• Member context and approved tool preloads are synchronized so a conversation starts with the effective organization policy.
• Integration editors preserve supported configuration fields across the desktop and admin surfaces.
• Billing state, usage summaries, and shared-credit reads are more consistent across organization changes.

​

Fixed

• Closed races around integration hydration, OAuth registration, member profile updates, partial policy, catalog visibility, and cross-organization editor state.

​

Removed

• Conversational onboarding retired. Removed the first-run Q&A, local integration scanners, role and app pickers, and guided first-task handoff. Existing onboarding conversations remain readable.

​

What’s new

• Local OCR for Computer Use. Studio can read visible application text through the local Computer Use path and use stronger window and coordinate context before targeting an action.
• Digital-worker foundation behind the owner gate. Early-access worker definitions gained staged execution, trigger dispatch, checkpoints, resume, and channel-bound control tools. Workers remain disabled unless explicitly enabled for the organization.
• Clearer plan lifecycle. Planning cards now expose the active plan state and execution progress more consistently.
• Managed-customer handover. Partners can retry customer invitations and complete a managed organization’s handover with clearer lifecycle actions and cleanup. Earnings and payout views show clearer status, exact amounts, and confirmation before a payout request.

​

Improved

• Long chats compact and recover with more stable cache boundaries, model attribution, and exact-usage estimates.
• Computer Use prefers native inspection and checks the current window before applying coordinate-based interaction.
• Conversation failures remain visible after reload, truncated internal tool payloads stay out of the transcript, and procedure editors handle narrow layouts better.

​

Fixed

• Corrected over-cap top-up accounting races, stale credit warnings after recovery, model-specific cache attribution, and several long-conversation retry loops.

​

What’s new

• Model-aware usage display. Studio replaced raw internal credits with percentage remaining for plan allowance and currency for purchased top-ups, while attributing cost to the model that ran each turn.
• Postman-style connector defaults. Connector definitions can set reusable headers and parameters, reference Key Chain entries, and sign supported AWS requests with SigV4.
• Studio Remote preview foundation. The first pairing, relay, conversation snapshot, suggestion, and approval paths entered gated desktop preview ahead of the production launch.
• Investigate dashboard changes. Panels highlight changes since your last view, with a Mark as seen control. Right-click an anomaly marker for an explanation grounded in the dashboard’s current data, with citations to the relevant panel.

​

Improved

• Adaptive thinking and the effort control now apply consistently to supported models; retired Sonnet 4.5 and Opus 4.6 choices were removed.
• Billing and Key Chain now open as first-class workspace destinations, and the macOS menu bar can show active Studio work.
• Hidden live dashboards pause work, Markdown and Mermaid rendering is steadier, and completed tool results remain visible.

​

Fixed

• Improved connector parameter handling, default synchronization, billing-plan refresh, top-up state, mobile pairing-code expiry, and remote message snapshots.

​

What’s new

• Dashboard data history. Persistent versioned history records dashboard changes and supports earlier-state inspection and recovery.
• Cloud dashboard refresh foundation. Source hashes, shared refresh jobs, cached results, metering, and IP-pinned outbound fetches support coordinated panel data.
• TV mode controls. Detached live dashboards gained clearer wall-display controls and source-display behavior.

​

Improved

• Dashboard type filters, edit controls, and detached-window presentation are easier to use.
• Auto routing, context estimation, prompt caching, and per-turn attribution reduce avoidable long-conversation cost.
• Billing usage reflects cached-input economics and the model and region that actually served each turn.

​

Fixed

• Improved dashboard refresh coordination, source validation, files filtering, detached-route packaging, and update-status handling.

​

What’s new

• Auto becomes the default model. New and migrated conversations can use a visible Auto choice that selects Sonnet for normal work and escalates to Opus when deeper reasoning is warranted, with a one-click revert notice.
• Synced channels. Channels became organization resources with an administrative surface and channel-aware desktop sign-in and context.
• Dashboard variables. Define static or query-backed variables and reference them in panel parameters and request bodies.
• Partner program workspace. Organization admins can enroll in the partner program, share referral links, inspect linked accounts and earnings, request payouts, and manage eligible customer organizations from the Partner area.

​

Improved

• Long agent sessions retain a bounded output tail and preserve salient terminal evidence during compaction.
• Procedure shortcut triggers are visible from the procedure list.
• Browser-anchored desktop sign-in uses a PKCE flow and respects the active release environment.

​

Fixed

• Improved CLI table parsing, tab overflow, conversation-local suggestions, dictation draft isolation, dashboard variable resolution, and build reliability for the local embedding assets.

​

What’s new

• Computer Use. On macOS, Studio can inspect and interact with the desktop after Screen Recording and Accessibility permissions are granted. A secondary cursor path supports controlled background interaction.
• Audio Use. Capture microphone or system audio into a conversation, with clear start and stop state.
• Claude Opus 4.8 with extended context. Opus 4.8 is selectable with a 1M context option and adaptive thinking on the supported Bedrock path.
• Procedure event triggers. Start a procedure on a schedule, application focus, supported host lifecycle event, or configurable shortcut. App context can be attached to the triggered run.
• Dashboard schema v3. A resilient widget registry, explicit data-source adapters, error boundaries, autosaved layout, and optimistic editing form the new dashboard foundation.
• More precise remote desktop work. Save a resolution for each RDP host, target visible controls by their text, and use application-opening and form-filling tools. Click results report whether the remote screen changed so Copilot can detect an ineffective action and reassess.

​

Improved

• New-chat shortcuts are configurable and can attach foreground application context.
• Trial conversion can carry unused value into top-up credit, plan downgrades wait for the billing-cycle boundary, and top-ups can cover use after a rolling plan window is full.

​

Fixed

• Hardened Computer Use permission startup and moved-window safety, RDP workspace behavior, local code-tool approvals, MCP tool resolution, and approval hashing for shell redirection.

​

What’s new

• Channel-scoped conversations. Choose a channel for a new chat so its tools and context follow the team’s operational scope.
• Browser takeover. Temporarily take manual control of the live browser session, then release it back to Studio without creating a second session.
• Native web research. Web search and page fetch tools can gather public source material without driving an interactive browser.
• PDF export for reports. Export a Markdown report as PDF from its artifact controls.
• Setup adapts to your role. Guided onboarding uses your answers to understand your role and tools, scans local integrations before suggesting connections, and helps you complete a first task. Support-focused roles can move directly into application and tool discovery.

​

Improved

• Procedure runs get their own linked conversations, scheduled runs prompt for inputs, and run artifacts appear with chat activity and files.
• Mentions for hosts and procedures persist as structured references in the composer and transcript.
• The billing surface consistently expresses plan and purchased value as credits and handles card confirmation when required.

​

Fixed

• Improved RDP idle connection, browser approval noise, host credential linking, procedure-run approvals, replay availability, and payment or invoice links opened from the desktop app.

​

What’s new

• Computer Use foundation. Studio gained native screen inspection and controlled desktop interaction, with explicit separation between observation and state-changing actions.
• Firmware staging over TFTP. Stage a local firmware file from a selected private-network interface and copy the generated URL into a device workflow.
• Local shell. Studio gained a workstation shell with common file, text, network, pipeline, redirection, history, and completion commands, plus direct routing from Copilot.

​

Improved

• RDP uses incremental frame updates for a more responsive remote desktop and resolves attached credential references at connection time.
• Runtime tabs survive conversation switches, and terminal sharing sends viewer interaction over the live peer channel instead of routing every keystroke through cloud state.

​

Fixed

• Improved remote desktop reconnection, local permission reporting, terminal selection context, and workspace state restoration.

​

What’s new

• SSH host-key verification. Studio now checks an SSH host’s key fingerprint on every connection. It asks you to confirm a device the first time you reach it, and warns you clearly if a known host’s key changes later — a built-in guard against man-in-the-middle attacks.
• Invoices for credit top-ups. Every credit purchase now produces a downloadable invoice, ready for expenses and accounting.

​

Improved

• A chat-centric workspace. The workspace was reorganized around the conversation — Copilot stays front and centre, with a single focused slot for whatever you’re working on, whether that’s a diagram, a data table, an editor, or a report.
• A broad security-hardening release. Device credentials are kept out of Copilot’s tool calls, logs, and telemetry; stored connection settings are encrypted at rest; and isolation between organizations was tightened across sync and sharing.
• Fewer interruptions from approvals. Routine, low-risk actions prompt you less often, while anything that changes a device still asks first.
• Per-conversation drafts. Unsent text in the chat composer is saved per conversation, and each chat keeps its own artifact space — switching threads no longer mixes your work.

​

Fixed

• Call startup. Fixed a set of issues that could delay or drop audio when a call first connected.

​

What’s new

• One-click connector presets. Add a ready-made connector in a single click. Altostrat SDX and Zendesk are the first presets, with their endpoints already mapped — no manual setup.
• Mention Studio objects in chat. Type @ in the chat composer to drop a host, file, diagram, connector, or other object straight into your prompt, so Copilot works from the exact reference instead of a description.

​

Improved

• Your workspace persists and syncs. Open tabs, splits, and layout are saved as you work, restored when you reopen Studio, and kept consistent across your Studio windows.
• Smarter, safer Copilot. Copilot now judges whether a command is destructive from the full conversation context rather than simple pattern-matching, gathers available evidence before acting, and asks instead of guessing when something is ambiguous.
• Faster long conversations. Copilot compacts conversation history in a way that preserves its working cache, cutting latency and cost on long threads.

​

Fixed

• MikroTik terminal. Cleared stray colour codes in MikroTik output and a multi-second hang when opening some SSH sessions. SSH sessions also wait for their credentials to be ready before connecting.
• Procedure edits. Copilot editing a procedure no longer leaves a duplicate copy behind.
• Sync and sharing. Closed a class of bugs affecting shared-resource visibility and data isolation between organizations.

​

What’s new

• Just-in-time runbooks. Copilot can now propose a multi-step terminal sequence as a single approval, then run each step on your okay — quick fixes get the same audit trail as formal procedures.
• Copilot drives dashboards and data tables. Ask Copilot to update a panel, inspect a widget, or edit a data table inline, and it changes what you see without leaving the conversation.
• Calls open in their own tab. When a call connects, Studio now opens a dedicated workspace tab with live stats, dual audio meters, and auto-starting transcription. A new Calls settings panel controls the auto-open and auto-transcribe toggles.
• Public-client OAuth for connectors. Wire up APIs that use OIDC discovery and PKCE — no client secret required — when defining a connector.

​

Improved

• Resource sharing, rebuilt. The visibility model and Share dialog were redesigned for clearer private/shared distinctions across hosts, files, procedures, diagrams, and chats.
• Smarter voice transcription. Voice input is now anchored with NetOps/MSP vocabulary so jargon and vendor names come through correctly, and streaming partials show in italics for quick visual confirmation.
• Terminal replays as private files. Saved sessions are now archived as deduped private files, browseable from the Files sidebar with an improved playback UI — sidebar entries open the replay directly.
• Better default image generation. Image generation now defaults to Stable Diffusion 3.5 Large with a richer fallback chain, and you can request a specific model by alias.

​

Fixed

• Single-click HTTP hosts and browser auto-login. Hosts using HTTP/HTTPS now open straight into the in-app web view, and Copilot’s browser tool correctly threads stored credentials into text fields when filling forms.

​

What’s new

• Copilot can schedule itself. Ask Copilot to follow up later — “check this in an hour”, “remind me Monday morning” — and it will wake up and continue the conversation on time.
• Live sub-agents in chat. When Copilot delegates work to a sub-agent, you now see the sub-agent’s progress as a live, expandable card in the conversation and in the status bar.
• Open a Zendesk ticket from chat. When Copilot can’t help, it can escalate to Altostrat support without leaving the conversation.
• Faster, peer-to-peer shared terminals. Sharing a terminal session now uses a direct peer connection where possible, with smoother typing and fewer drops. Past sessions land in a new Session Replays tab.
• Browse and install MCP servers from inside Studio. A new MCP Directory lets you search a curated catalogue, see what each server does, and install it in one click — including support for self-signed certs.
• Vulnerability scanning. Copilot can now look up CVEs that match the devices you’re connected to and explain the risk in plain English.
• Native, redesigned billing. Top up credits, add a card, and manage your plan with native Stripe Elements styled to match Studio. Country, tax-ID, and payment-method pickers are all searchable.
• High contrast and personal themes. A new high contrast theme for accessibility, plus dynamic themes that pick accents from your email domain.

​

Improved

• Procedures get scheduled runs and a richer timeline. Schedule a procedure for later, browse runs in a timeline, and find their artifacts in the Files sidebar. The Markdown editor now supports an AI edit tool and a slash-command picker.
• Real-time billing across windows. Your usage, credits, and trial status update in real time across every open Studio window — no refresh needed.
• Live dashboards on a responsive grid. The dashboard canvas was rebuilt as a flexible grid that fills the viewport, with live summaries from your connectors.
• KMS-backed credential vault. Supported credential fields are envelope-encrypted under per-organization data keys wrapped by AWS KMS. KMS operations use AWS’s FIPS endpoint, and Studio schedules a fresh data-encryption key every 30 days. The old master-password flow is gone.

​

Fixed

• The “Trialing” badge now reflects remaining credits, not remaining days, so it disappears the moment you actually run out.
• Hosts no longer save before their keychain writes complete, eliminating a class of credential drift on shared folders.

​

What’s new

• RDP support. Connect to Windows hosts over RDP from the same workspace as your SSH sessions. Includes live frame streaming, fullscreen, a session indicator in the status bar, and AI-driven actions through Copilot.
• Onboarding for new users. A guided first-run flow that imports your existing hosts from PuTTY, SecureCRT, and Termius, scans for installed network tools, and walks you through your first Copilot task.
• Procedures, rebuilt around Markdown runbooks. Procedures are now plain-Markdown runbooks that Copilot can read, edit, and execute step-by-step. Sub-agents can run them as tools, and runs land in a unified timeline.
• Share files and procedures with your team. A new Share dialog lets you mark resources as private or shared with your organization, with a visibility badge on every list.
• Improved sign-in with team management. Sign-in, organization switching, billing, and team profile now share a single, themed dialog.
• VS Code-style command palette and keybindings. A unified command surface for the workspace with keybinding overrides — search commands, navigate, and run actions without touching the mouse.
• Activity indicator in the status bar. A single live indicator that shows every running task — chat agentic loops, sub-agents, sessions, calls, sync — with a popover to drill in.
• Ask mode in chat. Toggle Copilot into a read-only “Ask” mode for questions where you don’t want any actions taken.
• SNMP from chat. Copilot can now run SNMP GET, walk, and bulk-GET against your devices, with vendor MIB name resolution.
• New network probing tools. PMTUD probing, packet capture, and flow collection are now available as Copilot tools.

​

Improved

• The renderer was overhauled to use a VS Code-style workbench (now called the Workspace) with split panes, tab groups, and persistent state.
• Empty sidebar states are unified across hosts, files, diagrams, procedures, and chats.
• Onboarding events fire only on actual milestones, so the rating prompt shows up at the right moment.

​

What’s new

• Voice calling. Place and receive SIP calls from Studio with live transcription, DTMF, call history, codec pinning, and call-quality scoring. Calls survive NAT with built-in STUN.
• PTZ camera control. ONVIF detection lets you discover IP cameras on your network and control pan/tilt/zoom directly from a stream viewer.
• OAuth for MCP servers. Studio now supports the MCP standard OAuth flow (PKCE, dynamic registration, auto-discovery). Localhost callbacks are handled by Studio itself.
• MCP servers as tabs. The old MCP server dialog is gone — adding and editing MCP servers now opens as a tab, with reconnect feedback inline.
• Online/offline indicator. Studio now shows when it’s offline and reconnects gracefully.

​

Improved

• Local-first storage with a cloud twin. Your conversations, hosts, diagrams, and procedures sync continuously between local storage and the cloud, with conflict detection and force-resync controls.
• Search and command palette merged. One dropdown handles both — search local entities, app settings, and quick actions in the same list, with recent search terms.
• Dashboards rebuilt as a builder. The dashboard canvas became a rigid grid builder with viewport-aware generation and live polling panels.
• Browser tools. A picture-in-picture browser view, faster auto-approve toggle, and self-signed cert acceptance for MCP servers and connectors.
• Team presence overhaul. Manual status, idle detection, and deduplicated presence across devices.

​

Fixed

• Calling reliability. Six bugs that were preventing reliable Chime calls and screen sharing have been fixed.

​

What’s new

• Live dashboards. Build polling dashboards from your connectors in a widget grid, with summaries and per-panel refresh.
• API connectors. A new connector system lets you wire any HTTP API into Studio with OAuth2 or custom auth, then call it from chat or a dashboard. Custom MCP servers join the same sidebar.
• Account management inside Studio. A full self-service account surface — profile, organizations, billing entry, team members.
• Network status popover. See your public IP, ASN, and ISP in one click from the status bar.
• Network calculation tools. Subnet math, IP parsing, and YAML/XML data conversion are now available as Copilot tools.
• JIT components in chat. Copilot can now render rich, just-in-time UI inside its responses — tables, charts, alerts, code blocks, diff views, forms — instead of plain text.

​

Improved

• Faster, smarter Copilot. A new agentic loop runs tools concurrently, streams progress as it works, and routes to Claude Opus 4.6 or Sonnet 4.6 based on the task. Long conversations are now compacted in stages instead of failing.
• Streaming status bar. A persistent bar shows Copilot’s current phase, turn progress, and per-tool durations.
• Usage tracking. Per-feature usage is now metered server-side and surfaced in a usage page, with a status-bar badge for app resource use.
• Files sidebar. Streamlined with cleaner empty states and faster host file workflows.

​

Fixed

• Sign-in now exchanges tokens through the main process so production CORS no longer blocks Bedrock calls.

​

What’s new

• Local credential vault. Store device credentials encrypted on your machine with AES-256-GCM. Credentials redact themselves before anything is sent to Copilot.
• Vendor-aware terminal. Studio detects Cisco, Juniper, and MikroTik prompts, parses their CLI output into interactive tables, and shows a two-tier AI explanation for any error inline in the terminal.
• Quake-mode terminal. A global hotkey drops a terminal pane down from the top of the screen, ChatGPT-launcher style.
• Up to 8 split panes. Run more sessions side by side, with per-pane close controls.
• Saved commands and templates. Save commands for reuse, parameterize them with {{variables}}, and configure post-connect scripts that run automatically when you open a session.
• SFTP file management. Browse, upload, and download files on remote hosts from the Files sidebar.
• Structured logs panel. A bottom-panel Logs tab with structured agent log timelines, plus video support (HLS and RTSP) for CCTV troubleshooting.
• Markdown editing. Markdown documents now have a toolbar with formatting actions and a built-in Mermaid diagram creator.

​

Improved

• Memory budget per session. Long terminal sessions automatically prune scrollback to keep performance steady.
• Sidebar consistency. Hosts, diagrams, chats, and procedures now share the same context-menu actions, folder behavior, and item layout.

​

What’s new

• Shared terminal sessions. Invite a teammate into a live terminal session — they see your output, you see them typing, and the session can be replayed afterwards.
• Memory system. A dedicated sidebar and detail view for Copilot’s long-term memories, with date grouping, search, and full CRUD.
• Workflows. Multi-step procedures Copilot can run end-to-end, with grouped nodes, a context menu for editing, debug copy, and a workflow reflector for after-action review.
• Browser auto-approve. Toggle to let Copilot drive the headless browser without an approval prompt for each step.
• Auto-updater. Studio now checks for updates and notifies you in the status bar — with a manual check option.
• Custom title bar on Windows and Linux. Native minimize/maximize/close in a dark frame that matches the rest of the app.

​

Improved

• Diagram editor. A new layers panel and selection spotlight in the inspector, AWS draw.io fidelity import (with gradient fills and theme-aware contrast), section shapes, multi-select property edits, and smarter responsive toolbars.
• VS Code-style workspace menus. Unified tab chrome, workspace-level commands, and a faster team switcher.
• Chat composer. New prompt-kit based chat UI, deep-linked AI settings, and grouped tool cards for repeated actions.
• YOLO mode. Optional auto-approval for chat tools when you want Copilot to keep moving.
• Theming. A new theming architecture with smooth, native-feeling transitions between light and dark.

​

What’s new

• Studio for Windows. Studio now ships as a signed installer for Windows alongside macOS, with auto-incrementing versions and an integrated update flow.
• Voice in chat. Talk to Copilot — recordings transcribe in real time and an audio visualizer shows you the level. Microphone and camera permissions are now properly entitled on macOS.
• Extended thinking. Copilot can now think harder when needed, with cryptographically-signed reasoning sessions you can audit.
• Team presence. See who else from your organization is in the workspace and what they’re doing.
• AI usage tracking. A usage page shows token consumption per feature, with a token-usage indicator wired into the status bar.

​

Improved

• Diagram editor v2. A new MaxGraph-based editor with cleaner shape rendering, draw.io import and export, viewport-preserving edits, and improved label and edge styling.
• Local search. Built-in full-text search now uses an on-device embedding model for relevance, downloaded the first time you launch Studio.
• System tray. A macOS tray icon and About dialog, plus dynamic window icons.

​

Fixed

• Team switching is faster and more reliable, with better error logging when an org load fails.

​

What’s in the box

• AI Copilot. A conversational sidebar that can read your screen, run commands in your terminals, navigate web admin UIs, and edit code or markdown — with a clear approval flow for anything that touches a host.
• Multi-protocol terminal. SSH, Telnet, and serial console sessions in tabbed editor groups, with private-key and jump-host support.
• Hosts and credentials. A hosts sidebar with folder organization plus a local keychain that stores credentials encrypted on your machine.
• Diagrams. A diagram canvas with shape libraries, draw.io import, and Mermaid support — opened side-by-side with terminals and chat.
• Browser automation. A headless browser tab Copilot can drive on your behalf for web-based device admin and service dashboards.
• Search and command palette. Full-text search across hosts, diagrams, conversations, and settings, plus a quick-action command palette.
• Image attachments and Monaco code blocks. Drop images into chat for Copilot to look at, and view structured output in a real code editor with syntax highlighting.