Eigent Open Source Cowork Release Notes: Privacy Refactor, GPT-5.4 & Multi-Modal Improvements 🚀

This release focuses on cleaning up the user experience and keeping Eigent's model ecosystem current. From a streamlined privacy flow to a new model option and important toolkit upgrades, v0.0.87 is all about polish and reliability.

As always, these improvements are community-driven. Let's dive in.

---

🔐 Simplified Privacy Consent Flow

This is one of the most impactful UX changes in this cycle.

Huge thanks to @eureka928, with reviews from @a7m-1st and @Wendong-Fan, for completely refactoring the privacy consent experience.

Previously, users were prompted with a four-toggle privacy settings page inside the chat interface — an interruption that added friction to the core workflow. In v0.0.87, that entire flow has been replaced by a single, optional "Help Improve Eigent" telemetry toggle.

What's changed:

• Removed runtime consent gating from ChatBox and InputBox components
• Replaced the multi-toggle privacy page with a single optional toggle
• Added help_improve field to the backend UserPrivacySettings model
• Created a reusable Radix UI Checkbox component for consistent form controls
• Updated internationalization across all 11 supported locales
• Cleaned up unused components — the old Dialog/Privacy component is gone

The result? Smoother onboarding, and a chat interface that stays focused on what matters — getting work done with your agents.

🔗 PR: https://github.com/eigent-ai/eigent/pull/1239

---

🤖 GPT-5.4 Model Support

Model flexibility keeps expanding.

Thanks to @fengju0213, Eigent now supports GPT-5.4 as an available model option, keeping the platform up to date with the latest from OpenAI.

Because of this PR, you now get:

• GPT-5.4 in the model selector
• Access to improved reasoning and instruction-following capabilities
• Full compatibility across all agent types

Whether you're building complex multi-step workflows or running simple Q&A agents, having access to the newest models means you can always choose the best tool for the job.

🔗 PR: https://github.com/eigent-ai/eigent/pull/1465

---

🧠 Upgraded Multi-Modal Agent Toolkit

An important architectural fix for anyone relying on vision capabilities.

Thanks to @Wendong-Fan for replacing the deprecated ImageAnalysisToolkit with the newer ScreenshotToolkit.

What's improved:

• Image analysis tasks now properly use the user's configured model
• No more silently falling back to a hardcoded default
• Your model selection is respected across all task types, including screenshot capture and image analysis

This is a meaningful change — it ensures that when you pick a model, that model is actually the one doing the work across every workflow, including multi-modal ones.

🔗 PR: https://github.com/eigent-ai/eigent/pull/1464

---

📊 Better Success Rate Display for Triggers

Small detail. Big clarity win.

Shoutout to @Desel72, with reviews from @Zephyroam and @bytecii, for fixing how trigger success rates are displayed.

When a trigger is waiting for its first execution, the success rate previously showed a misleading "0%." Now it shows "–" until actual execution data is available.

What's changed:

• Success rate displays "–" instead of "0%" for unexecuted triggers
• Threshold constants extracted into shared location for cleaner code
• More honest and accurate status reporting

No more second-guessing whether your trigger has actually failed or just hasn't run yet.

🔗 PR: https://github.com/eigent-ai/eigent/pull/1424

---

⚡ Faster Dev Mode Startup

Another win for developer experience.

Thanks to @4pmtong for optimizing the development startup flow, merged by @Wendong-Fan.

Previously, the app downloaded and installed uv on every startup when versions changed or the .eigent/bin directory wasn't initialized. In dev mode, Eigent now searches for a system-installed uv on your PATH first, avoiding redundant downloads and leveraging existing caches.

What's improved:

• Searches system PATH for uv before falling back to .eigent/bin
• Uses platform-appropriate commands: where.exe (Windows) and which (macOS/Linux)
• Replaces execSync with execFileSync to prevent shell injection vulnerabilities
• Development-mode only — production builds remain unaffected

Faster iteration cycles for everyone contributing to Eigent.

🔗 PR: https://github.com/eigent-ai/eigent/pull/1448

---

📦 NPM Wrapper Version Tracking

A frustrating upgrade bug, squashed.

Thanks to @4pmtong, approved by @bytecii, for fixing a version management issue with npm wrappers.

When users upgraded Eigent (e.g., from 0.0.85 to 0.0.86), wrapper scripts continued pointing to the old Python virtual environment. This caused wrappers to reference obsolete dependencies after an upgrade.

What's fixed:

• Wrappers now store the actual pythonPath instead of a static marker
• Automatic detection when the path changes during upgrades
• Wrapper scripts are recreated automatically — no manual cleanup needed

Upgrades just work now.

🔗 PR: https://github.com/eigent-ai/eigent/pull/1458

---

🐛 Bug Fixes

This release also includes several targeted bug fixes to improve stability:

• Fixed help_improve privacy bug — Resolved a UI validation issue caused by an unexpected backend field, with a follow-up patch on the server side. (#1455 — @a7m-1st)

• Fixed trigger-induced UI issues — Corrected header spacing on Windows, restored file viewer scrolling, and hid unnecessary execution settings from schedule trigger configurations. (#1453 — @a7m-1st)

• Fixed scroll for HTML & PDF — Resolved scrolling issues when viewing HTML and PDF content in the file viewer. (#1470 — @4pmtong)

• Added missing react-checkbox dependency — Fixed a missing package required by the updated privacy settings UI. (#1467 — @4pmtong)

• Fixed README typos — Minor documentation cleanup. (#1461 — @layla)

---

❤️ Community Keeps Raising the Bar

This sprint delivered:

• A streamlined privacy consent flow
• GPT-5.4 model support
• An upgraded multi-modal agent toolkit
• Better trigger status reporting
• Faster dev mode startup
• Reliable npm wrapper upgrades

Every single one of these improvements came from contributors who cared enough to improve what they use.

That's the power of open source — and the power of open source cowork.

If you've been thinking about contributing models, UX, agents, tooling, or cowork workflows, this is your moment.

Let's keep building.