Skip to content
| Marketplace
Sign in
Visual Studio Code>Other>Chepio AgentsNew to Visual Studio Code? Get it now.
Chepio Agents

Chepio Agents

CHEPIO.TECH

|
2 installs
| (0) | Free
One project conversation across installed coding-agent CLIs, with strict evidence-gated routing.
Installation
Launch VS Code Quick Open (Ctrl+P), paste the following command, and press enter.
Copied to clipboard
More Info

Chepio Agents for VS Code

Connect Codex and Claude Code through private CLI installations managed by the extension, or explicitly configured custom CLIs, and use your installed Antigravity CLI with Google sign-in. Each supports independent project conversations. Open the Chepio tab (CT icon) in the right sidebar, follow a full-height chat above a separate composer, tool activity, approval requests and reported model. The header opens Chat model, Agents, Attach files, Current Processes and settings. New Chat is beside the Always Allow shield in the input. The editor stays available for your code.

To display the brand icon instead of the tab label, run Show Sidebar Icons. VS Code applies this display mode to all tabs in the secondary sidebar. The native tab uses a COLR/CPAL icon font to preserve the original blue and dark background. The extension listing uses the full-color PNG. To regenerate the font after changing the brand artwork, run scripts/build-brand-font.py with build-only fonttools==4.68.0 (MIT).

Until a validated CMIE benchmark generation exists, fixed rules choose the model for each request and the chat labels that choice as provisional, not measured. An installed, validated generation replaces the rules with strict evidence-based routing. Existing manual conversations remain usable. See runtime contracts and limits.

In Settings, Task priority chooses Task fit, Quality, Speed or Economy. The leftmost Chat model header icon offers Automatic and your connected models in a compact dropdown, without recommendation text underneath. Your choice is saved for following messages in that chat, including after sending or reopening it. You can change it while a response is running; that response keeps its current model. The refresh icon reconnects the runtime and refreshes availability when needed. Automatic routing is checked again when sent without an extra model call.

If a first project command is blocked before workspace selection, the chat explains the block and shows branch choices. Select a workspace or continue in the current branch; commands then follow the usual approval controls.

0.10.15 — Chat permissions and card repair

Attach files is in the header. The aligned input controls include an always-available permission shield: red means Always Allow, green means approval for each request. The choice persists for the current chat, including its next process. One button sends while idle and stops active work; stopping preserves your draft.

When Close card checks fail, the original eligible agent can repair the source and rerun all required checks. Failed command output is retained for diagnosis, and merging still requires confirmation.

See the 0.10.15 release notes.

0.10.14 — Chat details and per-chat models

Open Chat details for separate prompt, technical, action, approval and plan histories. Prompts scroll within a fixed-height section and remain in the conversation. The leftmost Chat model icon saves the model for the current chat; changes during a response apply to the next request. The input starts on one line and grows or shrinks with its content, with motion and reduced-motion support. Pending decisions remain accessible in the curtain.

See the 0.10.14 release notes.

0.10.13 — Commands after personal approval

Connected agents can propose a command or an ordered list when they need host terminal access. Review the exact commands, working folder, shell, environment and timeout, then choose Allow and run or Deny. After approval, Chepio opens dedicated terminals, records their output and observed exit codes, and returns the result to the same chat. Changing the commands requires another approval. Stop, failure, disconnect or expiry prevents automatic replay.

See the 0.10.13 release notes.

0.10.12 — Compact chat fixes

The collapsed composer keeps an editable single-line input and expands when a draft wraps or contains newlines. Chat details collapse independently, while questions, approvals and completion controls remain accessible. The expanded controls retain per-request model selection and recommendations. Model rows are more compact.

See the 0.10.12 release notes.

0.10.11 — Task priorities and first-request recovery

Choose Quality, Speed or Economy in Settings, then use the chat's model dropdown for the current request. The engine recommends a model for the draft and preserves the saved chat preference after a one-request override. Unmeasured recommendations remain clearly labelled. Failed first project commands now explain the workspace restriction and show the branch picker, including when reopening older empty failed turns.

See the 0.10.11 release notes.

0.10.10 — Antigravity and glass model cards

Connect the official Antigravity CLI with Google sign-in, choose models from your account catalog and keep one-time approvals for tools. Existing Gemini conversations remain available. Agent cards and model selectors use local brand icons and glass backgrounds, with solid accessibility fallbacks.

Audit & finish now recovers committed work from missing, unlocked worktrees. Locked or changed sources retain their safeguards. This release also includes the pending chat playback, composer, recovery and runtime fixes.

See the 0.10.10 release notes.

0.10.9 — One combined project result

Audit & finish now collects all branch and worktree changes into one selected destination, completes the combined work and verifies it before delivery. Choose main or one open PR; successful cleanup leaves only main in Current processes while preserving conversations and source folders. Interrupted work can resume or refresh its scope.

Gemini connects automatically through CLI-owned Google browser sign-in. New replies appear at an even typing pace with Markdown and Unicode preserved; restored history and reduced-motion preferences bypass playback.

See the 0.10.9 release notes.

0.10.7 — Gemini and chat recovery

Use your installed Gemini CLI with Google sign-in alongside Codex and Claude Code. Gemini supports model selection, streaming, tool activity, one-time approvals and supported question forms through ACP.

The composer now animates collapse/expand while retaining drafts and attachments. A Claude subscription-access refusal before work can switch to an eligible provider with compact local context and a full-history reference; no separate summary request is made. Started actions are preserved without automatic replay.

See the 0.10.7 release notes.

0.10.5 changes

Inspect retained chats through the bundled read-only debugger, including conversations from another project. Tool-enabled agents receive bounded discovery instructions; existing logs remain unchanged.

Confirmed deletion of an inactive task chat cancels unfinished tasks while retaining workflow evidence and its Git branch. Active operations, shared worktrees and unsaved work remain protected. Provider reply icons now prefer the installed extension's sidebar SVG and preserve text fallback when artwork is unavailable.

See the 0.10.5 release notes.

0.10.6 — Task completion

A finished task with all required checks passed shows green Task completed and Commit above the message input. Commit asks the current chat agent to review the combined changes, runs local checks, merges the verified result and removes the task branch/card. Chat history and files remain. Failed checks or cleanup remain visible; push, deploy and publication are separate actions. Questions, approval controls and Always allow remain accessible when the input is collapsed. Questions still need your answers.

The completion row aligns with the collapsed input control in both themes. Protocol 12 adds recoverable branch cleanup after integration. See the 0.10.6 release notes.

0.10.4 changes

Messages appear immediately, new drafts remain editable during preparation, and failed or cancelled sends retain text and attachments. Collapse the composer without losing its state; keep process status visible and replay Your prompts history. Update notices expire while details and retry remain in Settings.

Choose the current branch to start the original task with required checks. Close card verifies the combined local changes before asking to integrate the exact candidate. Failed checks, conflicts or changed files block completion; branches, working folders and chat history remain available.

Default agents use private, verified installations owned by the extension. Failed updates preserve the working release, and custom CLI paths stay manually maintained. Protocol 11 safely replaces earlier development runtimes. See the 0.10.4 release notes for verification and limits.

0.10.3 changes

Deferred Codex and Claude Code updates retry when their provider becomes available, including explicit requests with automatic updates disabled. Chat confirms a verified update with a compact bold green notice; unchanged versions say is up to date, while progress and errors remain distinct.

Codex accepts additional context during generation; Claude retains it for delivery after its current work. The branch picker and settings are compact, and live update failures preserve the draft and Stop access.

Current Processes cards enlarge on hover or keyboard focus without changing their background or moving the canvas. Moving away, Escape, scrolling or dragging restores the diagram. The model comparison uses compact rows and controls while retaining all columns, model details and selection actions. See the 0.10.3 release notes for scope and verification.

0.10.2 changes

Choosing a task branch starts implementation automatically. A branch name typed in chat selects the workspace; intermediate replies continue until the task reports completion and the project checks pass. Failed checks return to the same worker for repair. Questions, permissions, Stop and actionable errors remain explicit pauses.

Current Processes cards keep their size on hover. Open a chat through its title or branch; the redundant Open chat button is removed. See the 0.10.2 release notes for validation and limits.

0.9.3 changes

Technical reasoning is collected in one expandable section above the conversation. Images have small previews; open one to browse the turn's images with Previous/Next or arrow keys. Inspection screenshots are labelled Debugging. Missing images offer Retry image. The latest-message arrow is hidden when the end is visible. The native header paperclip follows the VS Code theme. The composer shield toggles Always Allow; its red checkmark and border indicate Always Allow, while the green shield means approval for each action. Native image output up to 8 MiB is retained; local Markdown references and raster file attachments up to 20 MiB can also be previewed. Remote images are not loaded automatically. See ADR-0041.

Add exact model IDs alongside the native catalog from Agents or the model picker. Added IDs remain labelled unverified; aliases and historical scores stay separate. Family-scoped Opus/Sonnet limits no longer exhaust Haiku.

See the 0.9.3 release notes for scope and validation.

0.9.2 changes

  • Chat normally before choosing a task branch. Implementation starts after approving the task and its required checks; saved conversations retain their history.
  • Open saved history and live replies by clicking a branch name or chat title on Current Processes. Ordinary conversations have their own Chats area; coding chats show the actual working branch.
  • Keep prompt-history controls available, preserve reading position and return to the latest reply in long conversations.
  • Recover when a selected chat disappears, retain drafts and attachments after failed sends, and cancel message preparation while provider metadata is loading.
  • Ignore stale requests from another chat and show image previews and capability refusals without leaving the composer waiting.

See the 0.9.2 release notes for validation and release status.

0.9.0 changes

  • Resume saved chats after idle workspace owners are cleaned up; selecting another model preserves the current task.
  • Refresh model availability on every send and immediately retry a recovered preferred model.
  • Start new Git chats with read-only English branch proposals, then continue in the explicitly selected worktree.
  • Preserve reading position while scrolling, selecting text or using history controls; follow new content after an idle pause.
  • Retain optional concurrency limits, queued tasks and paged chat history.

0.8.1 changes

  • Continue old chats with another available model, including missing CLI and quota recovery.
  • Restore missing managed workspaces from the retained branch or current committed project state without moving the open project. Deleted uncommitted files require a backup and are not recreated by Git.
  • Keep full history on disk with bounded model handoffs and paged replay, including journals larger than 8 MiB.
  • Show branch-name reply buttons and retain model diagnostics in optional Technical reasoning blocks.
  • Replace standalone consolidation with a shared-chat audit request, retaining task approvals and saved-chat workspaces.

0.6.3 changes

  • Current Processes fills the editor without a page header or toolbar. Branches and chats surround the project.
  • Drag or scroll to pan; pinch or Alt+scroll to zoom at the pointer; +/− zoom, arrows pan and 0 fits the diagram.

0.6.2 changes

  • When a model needs usage credits, Chepio now steps down to the next weaker model and keeps going until one accepts the request. This historical policy is extended by the 0.8.1 recovery rules above.
  • Claude Fable models count as the most capable tier, so regular tasks no longer start on them.
  • A model-access error after Claude Code's "requesting" status now triggers the switch instead of failing the chat.

0.6.1 changes

  • Parallel commands no longer stop a chat with "Invalid or excessive Codex requests". Each process keeps up to sixteen pending approvals and questions; another chat cannot use its allowance. A request beyond the limit is declined with an explanation, and the chat continues (ADR-0030).
  • Permission requests show the action, exact command, file and folder; the original parameters stay under Technical details. Waiting for your decision no longer counts toward the turn timeout.
  • Always allow approves action requests for the selected running process only with a remembered choice for that chat.
  • Prompt history at the top lists every prompt of the chat. Thinking summaries are saved and grouped per turn.
  • If a model needs usage credits or is unavailable, Chepio silently tries weaker eligible models before any work starts.

0.6.0 changes

  • New chats run a provisional model chosen by fixed rules: short questions prefer a fast model, regular tasks a balanced one and complex tasks the most capable signed-in model with remaining capacity (ADR-0027).
  • After an extension update, an older background runtime is replaced automatically before the next message; one that is still running agents asks first (ADR-0028). This fixes "Runtime disconnected" when a runtime from an earlier version was still running.
  • Delete on a Current processes card removes an inactive chat from Chepio memory (ADR-0029).
  • The canvas has no bottom bars; zoom with a trackpad pinch or Alt+scroll, or press +, − or 0 while it is focused.
  • Files from a refused request stay in the draft and no longer block later messages in that chat.

0.5.0 changes

Native Claude/Codex questions share a stepwise bottom sheet. Answers retain Markdown structure; tool calls are grouped by turn with visible progress. New Chat preserves running work, Current processes links local branches and chats, and Add agent is in Settings. Prepared Agents cards now ship with the current read-only model catalog.

The cards were absent from the 0.4.0 source commit because their redesign remained in an uncommitted worktree; changing the version alone could not include that work. This version includes the ported implementation. File/image input and image previews use private snapshots and explicit capability checks; details and unsupported native features are listed in chat capability coverage.

Install and start

Requires desktop VS Code 1.143 or newer, on macOS/Linux or connected to a Linux Dev Container. Default agent installation also needs Node.js 22+ with npm on that workspace host. Packages are platform-specific: macOS ARM64/x64, Linux ARM64/x64 and Alpine ARM64/x64.

The Marketplace extension ID is chepiotech.chepio-agents. Install by ID for automatic Marketplace updates, or use a matching VSIX for a local build.

# Either command installs from Marketplace:
chepio install-vscode
code --install-extension chepiotech.chepio-agents

# Direct installation from an existing local package:
chepio install-vscode --vsix /path/to/chepio-agents-darwin-arm64-0.5.0.vsix
# Or use VS Code's own installer:
code --install-extension /path/to/chepio-agents-darwin-arm64-0.5.0.vsix

The Chepio terminal menu also offers Install extension for VSCode. --code /path/to/code selects a launcher; CHEPIO_VSCODE_CODE and CHEPIO_VSCODE_VSIX also work from the menu. Without an override, the installer uses code (code.cmd on Windows), with the standard macOS application path as a fallback. It never launches a shell or forces a downgrade. Windows can install a Linux remote package through VS Code; running the bundled runtime on Windows is not yet supported.

  1. Install Node.js 22+ with npm on the VS Code workspace host. Chepio installs its own Codex and Claude Code copies into its private extension storage. An optional custom CLI path uses your existing installation without updating it.
  2. Open Chepio in the right sidebar or run Open Chat. With no connected agents, a welcome page replaces the chat and composer.
  3. Choose Add Codex, Add Claude Code or Add Antigravity. Codex and Antigravity open their official browser flows and connect automatically after completion. Antigravity uses Google sign-in in a separate Chepio profile. Claude Code opens its native sign-in terminal and browser; finish there, then select Finish sign-in (or close the completed terminal). Cancel stops the pending connection. Use an existing CLI sign-in verifies the login in the selected provider profile. For Antigravity this means the separate Chepio profile, not the default terminal profile.
  4. After a verified connection, chat appears. Adding an agent creates no worktree and sends no model request. Open Agents to see your account, inspect the read-only model catalog, or check usage. There is one connection per provider on this host; signing in again changes the shared native CLI account.
  5. Send your first message from a trusted open project folder. With multiple folders, choose the project. Connected providers can contribute to the same conversation; each chat retains its own history. Reply labels identify the model that answered. Internal model selection and retry diagnostics stay collapsed under Technical reasoning; if execution cannot start, a short notice explains the failure. In a Git project, the agent first suggests 2–3 short English branch names in read-only mode. Choose a suggestion, enter a custom name, or select an existing free branch before implementation begins. Plain folders run in place. Open existing chats from Current Processes.
  6. New Chat, inside the message field immediately left of the Always Allow shield, opens an empty draft without stopping other processes. Its first task receives a name from your message. Ordinary messages receive answers immediately. Before implementation, choose a separate feature/fix or free existing branch, or explicitly continue in the current branch, including main. Separate worktrees start from committed files; the current-folder choice keeps local changes and instructions.
  7. Current Processes fills the editor with a canvas of local branches and associated chats, without an in-page header or toolbar. Cards occupy fixed groups around their project: Ready at upper left, Failed at upper right, chats in progress or paused at lower left, and branches with no active chat at lower right. Cards can be reordered within their group. Click a process title or branch to open its chat in the current VS Code window without changing the workspace. Ready cards and their status are green. Each card separately reports whether its commits are in the default branch, whether it has unique commits, and whether its working tree contains uncommitted or untracked changes. A Ready chat can still have unsaved Git work. Unknown status is shown explicitly. Dashed lines indicate the shared project, not merge ancestry. Drag empty canvas space or scroll to pan. Pinch on a trackpad or hold Alt and scroll to zoom around the pointer; with the canvas focused, + and − zoom, arrow keys pan, and 0 fits the whole diagram. The view retains its position and scale through process updates. The top-right × appears on every card. It is disabled for active chats and checked-out branches; its tooltip explains why. For an inactive branch it requests branch deletion with a unique-commit warning and a final safety recheck. For a stopped chat it removes the chat after confirmation: its messages and attachments leave Chepio memory, a clean worktree folder is removed, and its branch and commits stay. A worktree with uncommitted or untracked changes keeps the chat until you commit or discard them.
  8. Enter sends; Shift+Enter inserts a newline. Drafts are retained per conversation. The same composer button becomes Stop response while sending or working; stopping terminates the process tree and preserves history and workspace. Approval requests and the question-opening control appear under Chat details → Needs your attention. Structured questions open across the full chat content: answer each question, use Back/Next, then send all answers together. Hiding the sheet keeps the request pending; Cancel questions sends an explicit cancellation.

Antigravity CLI

Install the official Antigravity CLI on the workspace host and set Antigravity Path (chepioAgents.antigravityPath) to agy or its absolute executable path. The adapter requires stable CLI 1.5.2 or newer and its native ACP server. Chepio does not install, update or bundle this external CLI. Open Agents → Add Antigravity, complete Google's browser sign-in, and wait for Connected. Cancel terminates the pending sign-in process. If sign-in expires, use Sign in again.

Chepio maintains a separate ~/.chepio-antigravity/antigravity-cli profile with strict action approvals, workspace access restrictions, terminal sandboxing and additional AI credit spending disabled. It does not copy credentials or alter the normal Antigravity/Gemini profile. Sign-in and credential storage belong to the CLI. Models come from the account's actual ACP catalog, including model/effort variants; missing usage figures remain Unavailable. Select a model in Current Processes. Streaming, tool approval, supported questions, image attachments, Stop and native-session resume use the shared conversation. Running-task updates follow the current response. Before choosing a task branch, Chepio denies every tool permission request, including reads requested by strict mode. The native plan mode alone is not a read-only guarantee.

Existing Gemini connections and histories are preserved; new-agent onboarding offers Antigravity. Selecting an Antigravity model in an older conversation retains its journal and starts a new native provider session. It never sends a saved Gemini session identifier to Antigravity.

Verified boundary: official agy 1.5.2 on macOS arm64, Google authentication and one isolated edit request denied without writing the file. Other interaction flows have deterministic fake-provider/native editor coverage; that bounded check is not an end-to-end production project acceptance test.

Agents cards and model-selection windows use glass surfaces in light/dark themes, with solid high-contrast and reduced-transparency fallbacks. Their brand SVGs are bundled from Lobe Icons under MIT: attribution and pinned source hashes are in media/model-brands/LICENSE.txt and sources.json. No icon package or runtime network request is required. Model-family icons are independent of the connecting agent.

Run npm --prefix vscode-extension run test:antigravity for isolated mock coverage. For a persistent source review, use the manual-review.py --development --scenario antigravity workflow below; its guide describes reset/replay.

Gemini CLI account eligibility

Google ended Gemini CLI consumer access on June 18, 2026, including Google AI Pro, Ultra and free individual accounts. Those users should use Antigravity CLI. Connect it using Add Antigravity above; setting Gemini Path to its executable does not make its protocol compatible. Reinstalling Gemini or repeating Google sign-in does not restore retired consumer access. Chepio reports this refusal without exposing the upstream response's account data. Existing Code Assist Standard/Enterprise access is a separate eligibility path.

Keep the official Gemini CLI available on the workspace host only for existing connections with an eligible Code Assist Standard or Enterprise account. The saved Gemini card retains Sign in again, browser authentication and Refresh usage. New-agent onboarding offers Antigravity. If Gemini is not on the editor's PATH, set Chepio Agents: Gemini Path to its executable. Chepio uses this installation without installing or updating it. Its CLI source is Apache-2.0; Chepio bundles no Gemini SDK or model weights.

Open Current Processes → Model and select the Pro model reported by your CLI to use it for that chat. Model IDs and entitlement come from Gemini; the app does not infer your plan from a model name. Usage limits and account email are shown as unavailable when ACP does not report them. The adapter uses Google sign-in rather than API-key billing. If that sign-in expires, use Sign in again on the saved Gemini card or the native terminal; background refreshes do not open a browser. An expired account also offers Sign in again in Agents. Cancel stops a pending browser sign-in; an incomplete sign-in times out after five minutes without adding an account. Automatic memory extraction is disabled in Chepio's Gemini subprocesses.

Gemini supports streamed replies, native session continuation, file/image inputs, tool activity and one-time action approvals. Context sent while it is working is delivered after the current response. Forms appear only when the CLI actually exposes a structured ACP question; otherwise an ordinary question remains text. Close a form to answer later; closing it never selects a default. Stop terminates Chepio's owned process group.

This integration requires Gemini CLI with --acp and ACP v1. Chepio disables hooks, extensions and MCP startup in its Gemini subprocesses, and restricts read-only discussion to reading/search tools. Native slash commands belong in the Gemini terminal. An existing system settings policy is refused rather than replaced. Other installed CLI versions and your account's real eligibility need a separately approved live check; deterministic tests use gemini-acp-fixture-1 without model calls. Run npm --prefix vscode-extension run test:gemini for the isolated VS Code mock scenario (set VSCODE_EXECUTABLE_PATH to an installed editor to avoid downloading one).

Models and usage

On a chat card in Current Processes, Model: Automatic opens the model comparison. Choose Use model to save that model for the chat's next request, or Automatic to restore automatic routing. Saving a choice leaves the current reply running, including in legacy chats. On the next message, idle processes sharing the chat's folder are stopped automatically before the selected model starts with the saved history. Active work and pending questions or approvals remain protected; a blocking message identifies the chat to finish or stop. If the chosen model becomes unavailable, the existing recovery policy may use another eligible model while retaining your preference and recording the switch.

The popup lists all models supplied by connected agents and the saved historical profiles, with a scrollable table. The preliminary 28 October 2026 benchmark shows five task-category check counts, a total, tied ranks and whole-response time. Timeout/limit failures and untested models have no quality score. Historical profiles that are absent from the current catalog remain visible but unavailable. These screening scores do not drive automatic routing and do not claim project-level quality. Refresh models refreshes native availability without running a benchmark.

If your CLI menu omits a newer model, use Add model ID (or Add Model ID from the Command Palette), choose the connected provider and enter its exact ID, for example claude-haiku-5-5. The additional row states that access is unverified; it becomes eligible for both manual and Automatic selection. Adding an ID sends no request and does not update your CLI. Account access is checked when you send a message; a provider refusal follows normal recovery. Up to 32 IDs per provider are saved locally. Remove ID removes the addition and retains chat history/preferences. Native aliases such as haiku remain separate and show the description/version reported by your installed CLI.

Agents displays compact cards with the CLI-reported account email and plan when available, plus its model catalog and usage limits. The page ends after the cards; conversation management is available through Current Processes. A catalog entry is not a guarantee of access; the CLI checks that when a model turn starts. No hardcoded model list or paid availability probe is used. Without an active CMIE generation, provisional routing permits unknown capacity but rejects known exhaustion; an active generation retains its stricter evidence requirements.

Codex reports usage through its official account methods. Each window shows the actual provider period (for example, 5 hours or a week), percentage remaining and used, reset time and observation time. These are not calendar-day allowances. Refresh usage fetches account metadata again without starting a model turn.

Claude Code supplies Session (5hr) and Weekly (7 day) through its native usage metadata, so both bars can appear before you start a conversation. Additional supported windows appear when reported. Opening Agents or using Refresh usage fetches the account snapshot, including activity in other clients, without a model turn. This CLI method is experimental; if unavailable, update Claude Code or check usage on the provider's website. Conversation observations remain a fallback. The newest observation wins; missing or expired values never become an invented 100% allowance. Old conversation observations are excluded after extension restart or a detected sign-in change. Usage failures do not prevent recording a chat request. Family-only Opus/Sonnet limits do not exhaust unrelated models such as Haiku; shared account limits still apply to all models. Unknown aliases keep conservative limit checks.

The runtime continues when the chat or VS Code closes. Reopening Chepio reconnects to it. Stop Runtime and All Agents explicitly shuts it down. It is not registered as a system service and does not restart after reboot. After a runtime crash, interrupted turns are marked for review; they are never automatically repeated.

Each reply shows its provider's installed extension icon and the model reported at that point in the conversation. When the icon is unavailable, the label reads Codex (model) or Claude Code (model); without model evidence it says model not reported. Replaying history preserves earlier model labels. Reply icons are read locally from installed official extensions, never downloaded or bundled by Chepio. The Codex account card uses a transparent vector mark that follows the current theme. Only provider-emitted reasoning summaries are displayed; hidden or encrypted reasoning is never inferred. The preview retains legacy manual sessions and gates shared-chat execution on CMIE evidence; it does not yet decompose a task, assign subtasks or automatically merge their changes.

The chat opens its saved conversation and follows incoming replies. Coding chats show the current Git branch of their actual working folder under Chat details → Chat context; ordinary chats omit it. Account status remains on Agents. Opening or selecting a chat focuses its message field and preserves the draft. Background replies and account refreshes leave your current keyboard focus alone. The history scrollbar ends above the composer, including when drafts or approval requests grow. Long conversations mount only the visible rows and a small surrounding buffer (at most 80 rows); other rows are created on demand during scrolling. Collapsed tool payloads are created only when expanded. The native event transport still uses its existing bounded pages and cache.

New assistant messages buffer until the complete message arrives, then reveal their formatted text at a steady pace. Emoji and combined characters appear intact; links, code and image controls retain their behavior. Replies play in order, while saved history opens immediately. Reduced motion shows text immediately, and hiding or switching the chat ends its current playback. Failed or stopped turns retain any partial reply. A message stays marked busy for assistive technology until its text has finished appearing.

Run npm --prefix vscode-extension run test:typing from the repository root to replay the synthetic typing scenarios in an isolated VS Code development host. The temporary project's MOCK-CHAT.md describes the scenarios and reset command. Set VSCODE_EXECUTABLE_PATH to an installed editor to run without downloading VS Code.

On Current Processes, click a branch name or chat title to restore that chat's messages and current generation. The project root's branch label opens its most recent active or saved coding chat when one exists. A branch without a saved chat has no history to open. Ordinary chats appear in a separate Chats group at the upper right, without Git status or connecting lines to the project. Selecting a workspace for an ordinary chat moves the same conversation into its branch group and preserves its history.

Rich messages

Assistant responses retain headings, bold text, lists, code, bordered tables and explicit links. Project links such as README.md:1 appear as file chips with a filename and line number; opening one stays within the conversation’s project. JSON error details use an error.json code panel with indentation, syntax colors and Copy. Formatting preserves exact string/number tokens and duplicate keys. Inline code and commit hashes use orange emphasis. Commands are grouped into one expandable activity block per turn, with counts for running, completed, failed or interrupted actions. Nested output stays collapsed until requested. The work indicator distinguishes thinking, command activity, permission waiting and unanswered questions. Plans, diffs, native tool results and returned images have dedicated interpretations. See the capability matrix for the supported native messages and explicit limitations.

File attachments

Use the paperclip in the chat header to select files, or paste/drop them into the composer. Attach up to four files of 20 MiB each, inspect their names and sizes, and remove any before sending. Enter sends; Shift+Enter inserts a new line. Failed or refused requests keep their draft and attachments.

Provisional model selection sends images, documents and other files; it refuses audio, video and media generation. With a validated CMIE generation, the selected model, client and tools need verified support for the input format and requested result; reading an image does not imply image generation. The chat explains every refusal. Native raster input is supported by the transport; other formats use private file references and require verified tools. See attachment behavior and limits.

Dev Containers

Include the extension ID in .devcontainer/devcontainer.json:

{
  "image": "mcr.microsoft.com/devcontainers/base:ubuntu",
  "customizations": {
    "vscode": {
      "extensions": ["chepiotech.chepio-agents"]
    }
  }
}

VS Code selects the Linux/Alpine package for the container's architecture and runs Chepio in its workspace extension host. Install Git and Node.js 22+ with npm inside the container; Chepio installs its default agent copies there. Authenticate on that host through the native provider flow. Chepio never copies host credentials. Use persistent storage for the runtime home and VS Code extension storage if conversations and agent installations should survive container recreation; set the machine-scoped chepioAgents.home in remote user settings to a directory outside the Git repository. Rebuilding or stopping the container stops its processes; closing only the editor leaves them running while the container remains alive.

For a local build, use Extensions → … → Install from VSIX in the connected container window with the matching Linux/Alpine artifact. A local macOS VSIX cannot run in a Linux container. VS Code Dev Container customization.

Agent CLI updates

When the extension activates, it installs or updates private Codex and Claude Code copies before loading their models. Empty executable settings, including the legacy bare names codex and claude, select these managed copies. Explicit custom executable paths remain manually maintained. Terminal installations are preserved. Chat settings and Agents show progress and the verified version. Update agents, or Update Agent CLIs, retries and reloads connected catalogs. Authentication remains native to the provider; existing sign-ins can be reused.

Chat confirms a verified version change with a compact bold green Codex was updated or Claude Code was updated strip. If the reported version is unchanged, it says is up to date. Waiting and failed updates retain their own status; they never appear as a successful update.

New releases are staged separately and become active only after their version command succeeds. A failed download or unusable binary leaves the previous verified release available and shows a retry notice. If no working copy exists, install Node.js/npm if needed and retry before signing in. Versions reported on stderr are supported. Claude's npm native binary is prepared by its adapter without enabling arbitrary lifecycle scripts.

An independent active process postpones updates for its provider; automatic tasks postpone both because recovery can switch providers. Explicit update requests remain pending while busy, even with automatic updates disabled. Set chepioAgents.autoUpdateAgents to false to retain installed releases; missing defaults still install on first use. Claude's DISABLE_UPDATES=1 policy also blocks its managed installation/updates. Updates can still leave account or model availability unchanged; Add model ID remains available. Private releases are retained for active processes. Details: ADR-0055.

Appearance and updates

The header gear opens settings with an Agents → Add agent section, the running extension version, Appearance and Check for updates. Chepio Light and Chepio Dark use the canvas colors, mesh gradients and grid from chepio.tech; Follow VS Code selects the matching scheme. High-contrast editor themes retain native colors and omit decoration. Advent Pro is bundled locally under SIL OFL 1.1, including Cyrillic; tool output and filesystem paths remain monospace. The font's license is included under media/advent-pro/OFL.txt and in the third-party notices. The original logo blue #0d67a5 remains the composer focus glow and send button color. User message text is logo blue in light mode and light blue in dark mode; agent replies are dark green in light mode and lime in dark mode. Native high contrast overrides both for readability.

Opening the gear checks Marketplace for a newer stable release compatible with the current editor version and host platform; successful checks are cached for five minutes. Offline and unpublished states are shown explicitly. Update to … delegates the exact checked version to VS Code's built-in installer; after success, Reload VS Code activates it. Reloading detaches the UI and keeps the runtime running. VS Code's own automatic extension update settings continue to apply. There is no custom binary download or executable replacement in the extension.

Settings

If a model reaches its session or subscription usage limit during a task, Chepio stops that process and looks for an available model that meets the task's requirements. The replacement receives the saved history and partial work to continue; routing and provider diagnostics stay in collapsed Technical reasoning blocks. This also applies to a manually selected model; its saved preference stays unchanged. Each candidate is tried once. If none can run, Chepio says the work is saved and no model is currently available. Other failures have a short explanation with expandable technical details.

All settings use the chepioAgents prefix. Runtime settings have machine scope and cannot be overridden by a project to substitute an executable. Appearance has application scope and does not change the editor theme.

Setting Purpose
appearance system (default), light or dark; saved by the header settings menu.
runtimePath Absolute path to chepio-agentd; empty uses the bundled binary.
home External data directory; empty uses CHEPIO_HOME, then ~/chepio. Never put this inside a Git repository.
codexPath, claudePath Installed executable name or absolute path. GUI VS Code may have a different PATH from your terminal.
maxAgents Optional parallel CLI limit. Default 0 means no fixed count limit. Any positive value queues additional chats automatically; idle processes yield space when needed.
turnTimeoutSeconds Active-work deadline; pauses while awaiting permissions or answers. Default 3600.

After upgrading from an earlier runtime, stop the old runtime with Stop Runtime and All Agents when its work is finished, then reconnect to enable shared project conversations. Reloading the extension alone keeps the previous runtime alive. Restart the runtime to apply executable, process-limit or timeout changes. Restart the extension window after changing home. If a CLI cannot start, check its executable with --version and select a working absolute path. Chepio does not repair global installations or copy credentials. Remote SSH, WSL, browser VS Code and native Windows runtime execution remain unsupported.

Privacy and limits

History and previously created worktrees live under <home>/agent-runtime. New Git conversations use read-only proposals followed by an explicitly selected worktree; existing chats retain their workspace. A shared conversation excludes other Chepio writers in that workspace while a process owns it; unrelated external programs remain outside this guard. History is plaintext private local data; CLI-owned transcripts can also exist in the executor's own storage. Messages and project content can reach the configured CLI's provider under its own authentication and policies. A worktree isolates concurrent edits, not arbitrary programs. Only use trusted projects and trusted CLI configuration, including plugins, hooks and permission rules.

Chepio preserves normal CLI permission checks and forwards supported requests. Codex uses untrusted approval policy, workspace-write sandbox and user review; Claude uses default permission mode. No bypass flags are used. Actions the CLI already permits need not produce a UI approval. Unsupported interactions and approval details too large to display are refused. Disconnecting the UI leaves approvals pending; waiting for your decision pauses the turn deadline. Always allow resolves pending and future native action requests only for the selected running process. Questions and host terminal actions still require personal decisions. The runtime flag resets when the process ends; the saved chat choice is reapplied to its next writable process and does not change other chats. The shield beside New Chat stays visible: green means approval for each action, a red highlighted checkmark means Always Allow. The shield is always interactive, including before startup, during changes and while disconnected. Its color shows your selection; the tooltip distinguishes applying it from saving it for the next connected writable process. Rapid clicks apply the latest choice in order. A rejected change restores the actual mode. The webview retains the choice across reloads; a new chat starts with manual approvals.

For work that needs a host terminal command, the agent can present an exact list with its reason, folder, shell, environment and time limit. Review it, then choose Allow and run or Deny. Approval runs that list once in dedicated terminals and returns the output/exit codes to the same agent so it can continue. Each command starts in the displayed initial folder and environment; a failed or interrupted command prevents later commands. Changing the list requires a new request. Stop, terminal close or a lost runtime connection stops the owned execution; reconnection never replays it. Shell integration must work before any command is sent.

These commands run as your OS account on the workspace host. They can change host files and invoke privileged tools, so inspect the complete list before approval. If a command asks for an OS password, enter it directly in the terminal; never paste it into Chat. The agent receives command output, which can itself contain sensitive data. Terminal closure cannot undo completed changes or guarantee killing detached/root processes. No provider sandbox setting is changed. Synthetic reset/replay scenarios are in TERMINAL_ACTIONS.md.

No fixed retained-chat count limit; session lists are paged. Per-chat/protocol limits: 512 KiB / 4,096 cached events per session with the complete journal paged from disk, 1 MiB local socket frames, 16 MiB provider frames for bounded image payloads, sixteen combined pending interactions per process, 32 KiB of full approval details and 64 KiB of total pending data per process. Structured questions have their own typed 32 KiB limit; they do not use the permission-display limit. Pending-request overflow declines/cancels only the new request and explains how to continue; model handoffs use bounded excerpts with a full-journal reference; history growth no longer blocks a request at 256 KiB or 8 MiB. There is no automated retention deletion; Delete on the canvas removes one stopped chat at a time. A truncated/corrupt history file stops runtime startup for manual recovery; it is not silently rewritten. SIGTERM and explicit shutdown terminate owned process groups; an uncatchable daemon kill can leave orphaned tools and requires inspection before resuming.

Build and verify

From the repository root:

cargo build -p chepio-orchestrator --locked
npm --prefix vscode-extension ci --ignore-scripts
npm --prefix vscode-extension test
python3 tests/test_agent_runtime.py
python3 tests/test_shared_conversation.py
python3 tests/test_chat_recovery.py
python3 tests/test_model_selection.py
npm --prefix vscode-extension run test:host
npm --prefix vscode-extension run test:icons
# Focused Current Processes UI check with the same isolated fixture
npm --prefix vscode-extension run test:processes
npm --prefix vscode-extension run test:canvas-chat
python3 tests/test_vscode_build.py
python3 vscode-extension/scripts/prepare-runtime.py
npm --prefix vscode-extension run package -- --target darwin-arm64
# Output: releases/chepio-agents-darwin-arm64-<version>.vsix (ignored by Git)
python3 tests/test_vscode_package.py

test:icons checks local manifest SVG loading for both providers in light/dark themes, transparent artwork and missing/failed image fallback. Its provider extensions and marks are synthetic; no official extension or account is needed. All host scenarios use isolated profiles and temporary projects. test:host and test:processes also use a temporary data home and fake CLIs. Set VSCODE_EXECUTABLE_PATH to an existing VS Code executable to avoid downloading the test editor. On macOS its path is /Applications/Visual Studio Code.app/Contents/MacOS/Code. On headless Linux run the host tests under xvfb-run -a. prepare-runtime.py --target <Rust triple> builds that target and verifies its native executable header before packaging. It remaps the builder's home directory to ~ in embedded source paths. --binary <path> accepts an already-built binary with the same checks; build it with the same --remap-path-prefix, because package tests reject absolute home paths. Static musl builds can serve both Linux and Alpine; the script rejects an ELF interpreter for those packages. The sidecar records the target and SHA-256, checked against the VSIX target during package tests. Never label an ARM binary as x64.

This version is approved for personal local use. Commercial distribution needs a separate authentication decision; see ADR-0022.

Packaging requires an explicit request; source-only testing and merges do not create a VSIX. The manually dispatched Package VS Code extension workflow builds the six platform variants and uploads artifacts for review. It does not publish. After owner approval, publish each reviewed VSIX through vsce publish --packagePath <artifact.vsix> using the chepiotech publisher; publisher access must be established separately. Do not put Marketplace credentials in the repository.

To inspect the same fixture interactively, set CHEPIO_UI_REVIEW=1 when running host tests. The test log prints the preview's temporary path; create its continue file to close it and clean up, or close the test window. This mode pauses for up to 30 minutes after assertions pass and never calls real models.

Validation on 2026-10-28: deterministic runtime/adapter tests and VS Code 1.143.1 extension-host tests with fake CLIs on macOS arm64, including first-run welcome gating, adding without worktrees, first-message send (superseded by the ADR-0038 proposal workflow in 0.9.0), account/model/usage display, real webview layout, 5,000-message virtualization, lazy tool expansion, historical reply identity, autofocus, draft isolation, settings/update states, theme persistence and bundled font loading. Native OAuth exchanges are not exercised by these tests. A separate metadata-only manual check on Claude Code 2.1.170 returned both real subscription windows on 2026-10-28 without a model prompt. A disposable Alpine ARM64 Dev Container (Dev Containers CLI 0.89.0) passed runtime/client conversation, model, denial and shutdown checks using fake CLIs. This does not establish remote extension-host UI acceptance or Marketplace auto-installation before first publication. Initialization-only handshakes passed against Codex 0.163.0-alpha.10.2 and Claude Code 2.1.170 in temporary unauthenticated homes. Protocol definitions were also checked against the installed clients. Authenticated model turns remain a separate, owner-approved acceptance check; fake tests do not prove account access or compatibility with every CLI release.

The input starts as one editable line with its action buttons visible. A newline, paste or visual wrap expands it automatically; removing text or sending returns it to one line. Its height animates for 200 ms, with no manual collapse/expand buttons. Reduced motion skips the transition. Drafts, selection and attachments are retained. Only the compact working-status strip sits above the input.

Chat details opens a bounded curtain down from the top of the chat. It starts closed and remembers your choice. Inside are separate disclosures for Prompts, Technical reasoning, action totals, Auto approvals, Plans and changes, workflow and context. Each history scrolls within a bounded area; the curtain occupies at most 45% of the chat height. Recent activity appears first; Show earlier activity exposes older groups.

Prompts duplicates every user message in chronological order, starting with the first prompt, fifty per page. The original messages remain in the conversation. The prompt history has a fixed 200 px viewport (reduced for short windows), so long prompts cannot cover the chat. Errors retain their readable explanation and diagnostic source.

Pending questions, permissions and completion controls are under Needs your attention. A new decision opens that section and the curtain without moving keyboard focus. Closing it keeps the request pending; an unchanged update does not force it open again. Auto approvals does not grant terminal consent: host commands still require personal review of the exact list. Model selection stays available while the input is compact.

Unavailable models and missing CLI executables can continue with another connected, eligible model. A preferred model remains saved for future turns while an unavailable one is temporarily replaced. Quota exhaustion preserves completed work and passes its saved progress to the replacement after the old process stops. Diagnostics and retry logs remain available inside collapsed Technical reasoning blocks; no available model produces one short notice. Old chats keep their IDs and working folders. The full private journal is retained on disk, while each model request gets an explicit bounded excerpt and archive reference. See history and recovery.

Branch-name questions show buttons containing only the suggested names. Send another name through the regular chat composer. Selecting a reply sends that name to the agent; Git actions still follow normal permissions. Ordinary errors retain their raw details in a disclosure.

Review a packaged release with mock scenarios

After preparing and packaging the runtime, run on macOS Apple Silicon:

python3 vscode-extension/scripts/manual-review.py --code '/path/to/Visual Studio Code.app/Contents/MacOS/Code'

The launcher installs the current darwin-arm64 VSIX into a fresh temporary profile, creates isolated sample projects and fake providers, and opens a visible review window. It does not install into the normal profile. The temporary folder is printed and retained for feedback. Use Mock scenarios in the status bar to open chat, agent and process scenarios; the included REVIEW.md lists interactions and replay/reset instructions. The helper is not shipped in the VSIX. No new provider requests, accounts or credentials are needed.

The chat now supports an explicit follow-up choice while a turn runs: queue one task for the same chat or confirm a parallel worktree. Accepted tasks remain with their original chat when switching views; failures can be retried or returned to the draft. Queues live only while the extension window stays open. The scroll-to-latest button, JSON highlighting and grouped permission-policy history are available in the same chat.

Run npm --prefix vscode-extension run test:release for the combined native chat regressions.

Run npm --prefix vscode-extension run test:send for chat selection/send recovery, cancellable preparation and native image preview. A removed selected chat keeps its draft identity; sending starts a new conversation instead of silently discarding the request. During account/model refresh, Preparing message… is visible and Stop cancels that wait without sending later. Capability refusals remain visible in the conversation. Full module check assignments and their limits are documented in Regression checks.

Rust fenced code blocks preserve the original source and use lexical syntax colors, editor font settings and four-space tab display. Copy returns the exact source; unknown languages remain plain text.

Questions and development flow

Audit & finish in Current Processes first asks where to collect the project's work. Choose from two commit-derived branch suggestions, the default branch, or a typed name before work starts. All local and remote branch/worktree changes are collected into one result, existing tasks are finished, and the combined result runs every required check. Selecting the default branch authorizes local integration and an origin push; a feature destination creates one open GitHub PR and keeps its branch remotely. Successful cleanup leaves only the default branch in Current processes. Chats and source folders remain available.

The project status shows preparation, combining, implementation, checks, delivery and cleanup. Open audit chat, Stop audit and Resume audit remain available. Failures keep the cards and recovery evidence visible. If source branches changed, Refresh branches and restart includes the newer work and retains the previous candidate. GitHub CLI access is required for PR creation; remote/protection failures cannot be reported as completion. This operation does not package or deploy the application.

Retained chats, worktrees and historical recovery refs are preserved. Hover keeps process cards at their current size. Click a card title or branch to open its chat. Keyboard focus can enlarge a card for reading without moving the canvas; manual canvas zoom remains available.

Pending questions stay in Chat details → Needs your attention until you send every answer. Open the panel to answer; Answer later, close and Escape keep the process paused. Switching chats retains your in-memory selections. An orange question icon on the process card opens the waiting chat. Stop explicitly ends the process.

Model refresh reads the installed provider CLI catalog. If it lists older model versions, the installed CLI may need updating; Add Model ID accepts an exact ID without claiming that your account can use it.

New Git chats answer ordinary messages without a branch prompt. Before code changes, the compact picker shows two suggested English names and the main branch name, without extra labels or fields. Send a feature/fix name (up to five words and 60 characters), or an available existing task branch name, through the regular chat composer. Invalid names stay editable. The main branch button returns to read-only discussion; project changes still require a task branch. Start work… reopens a deferred choice without a model call, including after clarification or reopening the chat. You can keep chatting while this choice is open. Only invalid branch-proposal JSON gets a validation retry. History and attachments remain in the same chat; existing chats retain their chosen workspace. The prompt-history header stays visible as you scroll its contents. UI-generated continuation instructions are kept out of your prompt list.

Automatic permission details are grouped by turn. The latest-message control uses animated dots while working and a static arrow when idle. It is visible while the latest message is offscreen. Sending resumes following. Scrolling away pauses following; pending messages resume after four seconds without interaction. Text selection holds the reading position. Reduced-motion and high-contrast settings are respected.

If you set a positive Max Agents limit and all parallel slots are occupied, a sent message shows Queued and starts automatically when a slot is free. Its text and attachments are already saved. Stop cancels that waiting request; it does not stop another chat. Closing the editor leaves the background queue running. Stopping the runtime cancels queued execution while retaining history for explicit continuation. Existing custom Max Agents values are preserved; restart the runtime after changing that setting. The default is 0 (no fixed parallel count limit) for profiles without an explicit override. Set an existing custom value to 0 to remove its limit. OS resources and provider quotas still apply. Model access/quota fallback remains independent of this local process limit.

Every send refreshes connected CLI availability, so a recovered model can be retried immediately. Refusals are remembered only within that turn. GUI launches include known CLI/interpreter locations without changing the editor environment. Run npm run test:chat-update and npm run test:recovery for isolated native tests of these flows.

Task and release workflow

Choose a task branch by button or type its name in the chat. The original request starts automatically with the project’s existing verification commands. A preliminary answer does not finish the task: the runtime continues until implementation and required checks finish, or input/permission/a visible error requires attention. Failed checks return to the agent for repair. If no check command is discoverable, provide it when asked.

Workflow shows progress, results and manual recovery controls. Start task recovers an already selected branch that has no task; Prepare task manually defines custom scope/checks. Stop preserves the conversation and work. A new message resumes an uncommitted automatic task. Completion returns to verified local review. Commit, push/PR and merge each have their own explicit controls and evidence; model replies never mark a task verified or merged. Project tasks link back to the same chat. Use Prepare release for a frozen candidate, provider access, three independent review rounds, corrections and exact-build personal testing. Publication remains a separate approved operation.

The detailed contract and current limits are in chat-to-release. npm run test:lifecycle runs the manual delivery scenario; npm run test:progress covers automatic startup, continuation and verification. manual-review.py --development --scenario progress --code <VS Code executable> opens persistent interactive mocks without packaging. After compiling and building the debug runtime, use python3 vscode-extension/scripts/manual-review.py --development --scenario audit --code <GUI-executable> from the repository root for persistent interactive mocks without packaging. --scenario lifecycle exercises delivery controls; an explicitly requested --vsix uses the separate package review flow. All profiles are isolated and use fake providers. Native tests verify transport/UI behavior, not live model execution quality.

Add requirements while a task runs

Type in the composer and press Enter to update the same task. Codex accepts the update in its active turn; Claude saves it and visibly confirms delivery after the current response. A rejected update keeps its draft and attachments. The single action button shows Stop throughout preparation and active work, even with text in the composer; stopping retains that draft. When idle it returns to Send. Pending questions and permissions still require their own answers. Idle conversations start an ordinary next turn. The compact settings popover uses an accessible close icon.

Debug existing chats

The runtime built from this source supports chepio-agentd debug list, debug show CHAT_ID and debug events CHAT_ID. They read saved chats across projects in the selected Chepio home, including when the runtime is stopped. A tool-enabled agent receives the exact command and home on its next turn; ask it to inspect a chat ID and cite event sequences. Native permissions still apply. Existing tool-free branch discussions/reviews do not gain tools.

See saved-chat debugging for filters, pagination, source-build commands and isolated mocks. Journals are local private evidence; the reader never changes chat state or uploads logs. Previously discarded stderr and deleted histories cannot be restored.

Chat permission control checks

When Close card encounters failed checks, it asks the original task chat's agent to repair its source workspace, then reruns the same checks on a new combined candidate. Repair has at most three attempts and can be stopped. Missing tools or personal decisions remain visible; no other chat/model is chosen silently. A successful repair still requires your Merge and close confirmation and retains the branch, working folder and chat history. Run npm run test:card-repair to replay this flow with a fake agent and synthetic failed check.

Run npm run test:permissions for the isolated native header-attachment and Always Allow scenarios. The mock guide is CHAT_PERMISSIONS.md; it includes reset/replay and interactive permission, question, attachment and long-history checks.

  • Contact us
  • Jobs
  • Privacy
  • Manage cookies
  • Terms of use
  • Trademarks
  • Your Privacy Choices
  • Consumer Health Privacy
© 2026 Microsoft