This release adds 3 notable features for engineering teams evaluating rollout.
✓ No known CVEs patched in this version
ReleasePort's take
Light signalRelease v0.5.7 refines UI wording, routing logic, and performance of multi‑file writes while fixing several edge‑case bugs.
Why it matters: The update clarifies subscription setup across all touchpoints, optimizes model selection routing, accelerates multi‑file text writes by up to ~30% (preserving order), and resolves UI glitches that could mislead users about project scope or context budgeting.
Summary
AI summaryUpdates What Changed, Release Status, and Release Notes For Testers across a mixed release.
Changes in this release
| Type | Severity | Summary | CVE |
|---|---|---|---|
| Feature | Medium |
Clarified subscription setup wording across first-run dialog, Settings, install/start/sign‑in/model selection, usage/quota, reconnect, uninstall, and fallback states. Clarified subscription setup wording across first-run dialog, Settings, install/start/sign‑in/model selection, usage/quota, reconnect, uninstall, and fallback states. Source: llm_adapter@2026-05-23 Confidence: high |
— |
| Feature | Medium |
Added route‑source labels for model options to distinguish Codex, GitHub Copilot, Claude/Gemini, OpenRouter, direct provider APIs, and local model servers. Added route‑source labels for model options to distinguish Codex, GitHub Copilot, Claude/Gemini, OpenRouter, direct provider APIs, and local model servers. Source: llm_adapter@2026-05-23 Confidence: high |
— |
| Feature | Medium |
Added guarded local file/folder copy support for workspace tools and registered the required desktop command/capability. Added guarded local file/folder copy support for workspace tools and registered the required desktop command/capability. Source: llm_adapter@2026-05-23 Confidence: high |
— |
| Feature | Medium |
Added tool activity and coverage summaries to the coding review panel showing reads, searches, edits, terminal runs, browser checks, web use, and missing capability reasons. Added tool activity and coverage summaries to the coding review panel showing reads, searches, edits, terminal runs, browser checks, web use, and missing capability reasons. Source: llm_adapter@2026-05-23 Confidence: high |
— |
| Feature | Medium |
Updated support/funding defaults and documentation to make Cash App the public funding path while keeping Stripe/PayPal as hosted‑link‑only and excluding secrets from builds. Updated support/funding defaults and documentation to make Cash App the public funding path while keeping Stripe/PayPal as hosted‑link‑only and excluding secrets from builds. Source: llm_adapter@2026-05-23 Confidence: low |
— |
| Performance | Medium |
Improved 9Router subscription routing to prefer matching models when accounts are connected and show Codex context‑window controls only for active Codex routes. Improved 9Router subscription routing to prefer matching models when accounts are connected and show Codex context‑window controls only for active Codex routes. Source: llm_adapter@2026-05-23 Confidence: high |
— |
| Performance | Medium |
Made multi‑file text writes faster while preserving ordered per‑file results and added stale‑edit recovery guidance to retry failed hash‑based edits instead of stopping early. Made multi‑file text writes faster while preserving ordered per‑file results and added stale‑edit recovery guidance to retry failed hash‑based edits instead of stopping early. Source: llm_adapter@2026-05-23 Confidence: high |
— |
| Performance | Medium |
Improved full‑computer workspace mode by lazily resolving host roots for precise absolute paths while keeping selected‑folder chats project‑scoped. Improved full‑computer workspace mode by lazily resolving host roots for precise absolute paths while keeping selected‑folder chats project‑scoped. Source: llm_adapter@2026-05-23 Confidence: high |
— |
| Performance | Medium |
Improved terminal startup resilience with longer frontend timeout and native interactive‑shell timeout to reduce false failures on slower Windows shells. Improved terminal startup resilience with longer frontend timeout and native interactive‑shell timeout to reduce false failures on slower Windows shells. Source: llm_adapter@2026-05-23 Confidence: high |
— |
| Bugfix | Medium |
Fixed no‑project UI that could show project Git controls even when the chat was not tied to a project. Fixed no‑project UI that could show project Git controls even when the chat was not tied to a project. Source: llm_adapter@2026-05-23 Confidence: high |
— |
| Bugfix | Medium |
Replaced embedded data URLs with compact metadata placeholders in provider‑visible context (message, research, token estimates). Replaced embedded data URLs with compact metadata placeholders in provider‑visible context (message, research, token estimates). Source: llm_adapter@2026-05-23 Confidence: high |
— |
| Bugfix | Medium |
Tightened automation prompts and inbox summaries to avoid empty sections, protocol narration, duplicate source sections, and unwanted source cards when no web‑search tool ran. Tightened automation prompts and inbox summaries to avoid empty sections, protocol narration, duplicate source sections, and unwanted source cards when no web‑search tool ran. Source: llm_adapter@2026-05-23 Confidence: high |
— |
| Bugfix | Medium |
Fixed model changes racing against stale active‑chat references by applying updates through the current active chat id. Fixed model changes racing against stale active‑chat references by applying updates through the current active chat id. Source: llm_adapter@2026-05-23 Confidence: high |
— |
| Bugfix | Medium |
Prevented source‑card leakage from non‑web tool results by extracting bridge sources only from `web_search` calls. Prevented source‑card leakage from non‑web tool results by extracting bridge sources only from `web_search` calls. Source: llm_adapter@2026-05-23 Confidence: high |
— |
| Bugfix | Medium |
Corrected automation outputs that could echo hidden run instructions, include empty blocked‑approval sections, or append raw source lists in final inbox results. Corrected automation outputs that could echo hidden run instructions, include empty blocked‑approval sections, or append raw source lists in final inbox results. Source: llm_adapter@2026-05-23 Confidence: high |
— |
| Bugfix | Medium |
Resolved context‑budget overestimates and prompt bloat caused by large embedded media/data URLs being serialized into text‑oriented context. Resolved context‑budget overestimates and prompt bloat caused by large embedded media/data URLs being serialized into text‑oriented context. Source: llm_adapter@2026-05-23 Confidence: high |
— |
| Bugfix | Medium |
Scoped default Vitest command to the public `src` tree so private overlays under `.gilbert` are not collected as app tests during local release validation. Scoped default Vitest command to the public `src` tree so private overlays under `.gilbert` are not collected as app tests during local release validation. Source: llm_adapter@2026-05-23 Confidence: high |
— |
Full changelog
Gilbert Codex v0.5.7
Gilbert Codex v0.5.7 is a desktop reliability update for the Windows public alpha. It keeps the release focused on the real app experience: subscription setup should be understandable, model routes should be clear, workspace tools should be safer and faster, automation output should read like a normal result, and the release docs should match the build a user can actually test.
Release Status
- Version:
0.5.7 - Tag:
v0.5.7 - Previous public release:
v0.5.6 - Primary packaged target: Windows x64
- Release vehicle: GitHub Actions
Releaseworkflow plus Tauri NSIS installer - Installer family:
Gilbert-Codex-0.5.7-x64-setup.exe - Update feed asset:
latest.json - Updater signature asset:
Gilbert-Codex-0.5.7-x64-setup.exe.sig - Checksum asset:
Gilbert-Codex-0.5.7-x64-setup.exe.sha256 - Published SHA-256: pending final release build
- Local validation status: local Windows installer build passed on May 22, 2026
Do not fill the published checksum until the final GitHub Release installer artifact has been built and uploaded. Keep the GitHub Release body aligned with this file.
What Changed
- Clarified subscription setup across the first-run provider dialog and Settings. Install, start, sign-in, model selection, usage/quota, reconnect, uninstall, and fallback states now use subscription-focused wording instead of confusing sandbox labels.
- Added route-source labels for model options so the composer and model picker can distinguish Codex subscription routes, GitHub Copilot routes, Claude/Gemini subscription routes, OpenRouter routes, direct provider APIs, and local model servers.
- Improved 9Router subscription routing selection so connected accounts prefer matching models and Codex context-window controls only appear when a Codex subscription route is active.
- Added guarded local file/folder copy support for workspace tools and registered the desktop command/capability needed to copy binary assets or folders without treating them as text rewrites.
- Made multi-file text writes faster while preserving ordered per-file results, and added stale-edit recovery guidance so failed hash-based edits can re-read and retry instead of stopping early.
- Improved full-computer workspace mode by resolving host roots lazily for precise absolute paths while keeping selected-folder chats project-scoped.
- Hid project-only workspace/Git controls in no-project chats unless a real root or full-computer scope is available, reducing misleading controls in general chats.
- Cleaned provider-visible context by replacing embedded data URLs with compact metadata placeholders in message context, research context, and token estimates.
- Tightened automation prompts and inbox summaries so scheduled task results avoid empty sections, protocol narration, duplicate source sections, and source cards when no web-search tool ran.
- Added tool activity and tool coverage summaries to the coding review panel so users can see reads, searches, edits, terminal runs, browser checks, web use, and missing capability reasons.
- Improved terminal startup resilience with a longer frontend timeout and a native interactive-shell startup timeout, which should reduce false failures on slower Windows shells.
- Updated support/funding defaults and docs so Cash App is the public funding path while Stripe/PayPal remain hosted-link-only and secrets stay out of builds.
Fixed
- Fixed model changes racing against stale active-chat references by applying model/provider updates through the current active chat id.
- Fixed source-card leakage from non-web tool results by only extracting bridge sources from
web_searchcalls. - Fixed automation outputs that could echo hidden run instructions, include empty blocked-approval sections, or append raw source lists in the final inbox result.
- Fixed context-budget overestimates and prompt bloat caused by large embedded media/data URLs being serialized into text-oriented context.
- Fixed no-project workspace UI that could show project Git controls even when the chat was not tied to a project.
- Fixed local release validation by scoping the default Vitest command to the public
srctree so ignored private overlays under.gilbertare not collected as app tests.
Release Notes For Testers
- Build and install the Windows x64 setup executable locally before tagging or publishing.
- Create or sign into a local Gilbert user after install.
- Check Settings > Providers and Settings > Subscriptions for subscription install/sign-in/fallback flow.
- Test a subscription model route and a direct/OpenRouter fallback route so the composer label can be compared.
- Test a no-project chat, a selected-folder project chat, and full-computer mode before release.
- Test a local file copy or binary asset copy only inside approved workspace roots.
- Confirm automation inbox results and coding review activity read cleanly after a tool-using run.
Validation Commands
npm.cmd run build
npm.cmd run rust:fmt:check
npm.cmd run rust:check
npm.cmd run app:installer
npm.cmd test
git diff --check
The GitHub release workflow performs the signed Windows updater build when v0.5.7 is pushed or the workflow is dispatched with this version input. This local prep pass intentionally does not push the update.
Local Prep Validation
Passed on May 22, 2026:
npm.cmd run app:installer
npm.cmd test
npm.cmd run build
npm.cmd run rust:fmt:check
npm.cmd run rust:check
git diff --check
The local installer build generated the Windows NSIS setup executable successfully. rust:check completed with the existing dead-code warning for DictationStopRequest.dictionary. Published release checksums and updater signatures still need to come from the final release artifact.
Known Limits
Windows x64 is the packaged target for this release. macOS support is in development and should be out soon, but it still needs native release validation before it should be presented as official. Linux remains source-first until native packaging and QA are complete.
ChatGPT GPT-5.3 Spark is a known read-only route in this build. It can inspect workspace files, but write/edit tool calls are not completing reliably yet. Use the other available Codex/ChatGPT, OpenRouter, or local routes for file-changing work.
The installer is not Authenticode code-signed unless a release build is produced with a valid code-signing configuration. SmartScreen warnings are expected on some machines.
Private runtime overlays, provider keys, local databases, logs, OAuth secrets, downloaded SDKs, and signing credentials stay out of public Git and should not be bundled except through the intended release workflow or local machine setup.
Weekly OSS security release digest.
The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.
No spam, unsubscribe anytime.
Share this release
About UrbanWafflezz/GilbertCodex
All releases →Related context
Related tools
Beta — feedback welcome: [email protected]