# Canopy 0.3.4: test tab recovery, OpenCode usage, and Remote

> A task-based reading of the v0.3.4 release notes with a short installed-build trial for closed tabs, OpenCode Statistics, and Remote continuity.

Canonical HTML: https://canopyide.dev/blog/canopy-034-agent-workflow-release-guide
Article date: 2026-09-28

Canopy's latest listed downloadable release, v0.3.4, was published on 5 August 2026. Its notes list many fixes and additions. Three are especially useful to test in an actual coding-agent workflow: a restore window for a closed agent tab, OpenCode token usage and cost in Statistics, and a Remote portal parity update. The release list establishes that those changes shipped in the tagged build; the steps below help you decide whether they solve your own problem. They are not a hands-on benchmark of the installer.

## Start with the tagged build

Open the v0.3.4 GitHub release and choose the asset for your operating system and architecture. The page lists the release commit, platform downloads, and SHA-256 values. After installation, record the app version and platform shown by the build you are actually using. Canopy's main-branch README explicitly says it may contain features newer than the latest installer, so use the release notes as the shipping boundary and treat current-main screenshots as leads for a separate trial. If the version differs, repeat this exercise against its own release notes instead of assuming identical controls.

- Keep the release tag and installation platform beside each observation.
- For a trust check, compare the downloaded file's digest with the value for that exact asset.
- Do not assign the release's August date to this article; it was written later.

## Try recovery after closing one agent tab

The v0.3.4 notes include a restore window when an agent tab is closed. Use a disposable repository and a short, harmless CLI session. Record its project, checkout, branch, last prompt, and whether the CLI process is idle or running. Close the tab using the normal control, then try the release's restore affordance promptly. Confirm the restored view refers to the same project and checkout, and inspect whether the process is still live. A visible tab and a live agent process are separate states; this trial should not be used to infer that the agent survives an app exit or a sleeping computer. The longer session-lifecycle guide covers those boundaries.

- Test one tab first; do not use an unsaved production change for a recovery trial.
- Note what restored: tab, terminal output, CLI conversation, or process state.
- Save a branch and next-action checkpoint even if the tab comes back.

## Read OpenCode usage as an estimate with a source

The release notes say OpenCode token usage and cost were added to Statistics and also mention an OpenCode workspace edit-attribution fix. Run one small OpenCode task in the tagged build, then inspect its session, CLI, model, token, and cost rows. Record the time shown for the observation and the task outcome. Compare it with the OpenCode or provider account view when deciding what was actually charged. A cost estimate derived from reported tokens is not a provider invoice, and a visible edit list should still be checked against Git status and the final diff. This trial is especially useful if you switch CLIs to manage usage; it does not prove every CLI reports the same categories.

*Use one short task to separate visibility from proof of billing or accepted work.*

| Question | Where to check | What it does not establish |
| --- | --- | --- |
| Which session used OpenCode? | Canopy session and Statistics rows | That the feature works |
| What did the row estimate? | Tokens, model, time, cost basis | The actual provider charge |
| What changed in the repo? | Git status, commit, and diff | That every edit was attributed perfectly |

## Test Remote continuity on the same host

The v0.3.4 release notes say the Remote portal was brought closer to desktop parity. For a meaningful trial, enable Remote deliberately on a trusted network, open the phone view, and compare one live agent session and project with the desktop host. Ask one harmless question from the phone and verify that the same CLI session received it once. After a disconnect, reconnect and inspect the host, session, branch, and latest result before resending. Remote controls a running desktop host; it is not a cloud agent that continues after the host stops. Close or rotate access after the trial, especially if a public tunnel was enabled. The release-note title alone does not enumerate every phone control, so write down which controls you actually saw in your installed build.

- Use a disposable task and avoid including a real PIN or token in shared notes.
- Record whether the portal showed the relevant diff, services, and PR, instead of assuming complete parity.
- If one surface differs, report the version, platform, route, and reproduction in a GitHub issue.

## Judge the release by one completed task

A useful v0.3.4 trial ends with a branch, a running result, and a reviewed diff, not just a tour of controls. Start or resume one supported CLI, make a bounded change, check the app locally, close and restore a tab only when safe, inspect usage if that CLI exposes it, and review the final Git state. The same release also lists work on agent status, attention routing, PR quick actions, and stacked PR merging; test those separately if they matter to your workflow. Keep feature claims tied to the release notes and any installed observation. If the trial helps, inspect the MIT-licensed repository and consider starring it; if it fails, report the exact reproduction so the next build can improve.

- Record Canopy version, platform, CLI version, account, repository, and task.
- Give each observation a result: passed, failed, or untested.
- Do not substitute a release-note line or current-main README for a tested user journey.

## Copyable resources

### v0.3.4 workflow trial card

Keep PINs, provider credentials, private prompts, and repository secrets out of a public report.

````text
Canopy release and platform: [ ]
Downloaded asset and digest checked: [ ]
CLI/version/account type and disposable repository: [ ]
Task, starting branch/commit, and acceptance check: [ ]
Closed agent tab: [time]; restore result: [tab/output/process/session]
OpenCode Statistics, if used: [session/model/tokens/estimate/as-of]; provider comparison: [ ]
Remote route and host: [LAN/tunnel, version]; same session after reconnect: [ ]
Agent status/attention or PR quick action trial, if relevant: [ ]
Running app URL and observed behavior: [ ]
Final diff, checks, and accepted result: [ ]
Missing control or bug reproduction: [ ]
````

## Frequently asked questions

### Is v0.3.4 still the latest Canopy download?

It was marked latest on GitHub's release list when this article was drafted on 28 September 2026. Check the live Releases page before downloading because a newer build can replace it.

### Does tab restore mean an agent keeps running after Canopy closes?

No. The release notes describe a window for restoring a closed agent tab. Test the process state separately; app exit, host sleep, and CLI resume are different conditions.

### Does OpenCode's Statistics cost equal my bill?

No. Treat the displayed cost as a usage estimate and compare it with the account or provider record for actual charges.

### Does Remote work when the desktop host is off?

No. Canopy Remote controls a running host. Reconnect to that host and verify the same session before sending another instruction.

## Sources and further reading

- [Canopy v0.3.4 release notes and platform assets](https://github.com/FluidWorksApp/canopy-ide/releases/tag/v0.3.4)
- [Canopy current-main README and release caveat](https://github.com/FluidWorksApp/canopy-ide/blob/main/README.md)
- [Canopy v0.3.4 closed-tab restore change](https://github.com/FluidWorksApp/canopy-ide/pull/438)
- [Canopy v0.3.4 OpenCode Statistics change](https://github.com/FluidWorksApp/canopy-ide/pull/439)
- [Canopy v0.3.4 Remote portal change](https://github.com/FluidWorksApp/canopy-ide/pull/434)

## Related Canopy pages

- [What happens to a coding agent when you close Canopy?](https://canopyide.dev/use-cases/what-happens-to-agents-when-canopy-closes.md)
- [Understanding AI coding agent usage and estimated cost](https://canopyide.dev/blog/understanding-ai-coding-agent-costs.md)
- [Resume a local coding agent from your phone](https://canopyide.dev/use-cases/resume-coding-agent-from-phone.md)
- [Which coding-agent CLIs work in Canopy, and what differs?](https://canopyide.dev/guides/coding-agent-cli-support-in-canopy.md)
- [How to inspect an open-source AI IDE before installing it](https://canopyide.dev/guides/inspect-open-source-ai-ide-before-installing.md)

Canopy runs installed coding CLIs; CLI accounts, model selection, and provider billing remain separate. Check the installed release before relying on version-specific behavior.