Eigent Release Notes v0.0.89: Server Refactor v1, Haiku 4.5, and Security Auditor Skill

Eigent v0.0.89 is a foundational release. Under the hood, the entire server has been restructured around a domain-driven architecture with versioned API routing — a change that makes Eigent's backend cleaner, more scalable, and far easier to contribute to. On the surface, this release adds Claude Haiku 4.5 to the model picker, ships a built-in security auditor skill, and brings meaningful quality-of-life improvements to the chat UI.

For anyone running Eigent as their open source cowork platform or evaluating it as a cowork alternative, v0.0.89 represents a more mature, production-ready foundation. Let's dive in.

---

⚙️ Server Refactor v1: Domain-Driven Architecture

The biggest change in this release — and one that sets the foundation for everything ahead.

Huge thanks to @4pmtong, reviewed by @Wendong-Fan, @bytecii, and @fengju0213, for completing the v1 server refactor. This was a significant architectural undertaking that transforms how Eigent's backend is organized.

Previously, the server used a flat controller/component/service layout. That worked when Eigent was smaller, but as the platform has grown — adding more models, more triggers, more OAuth providers, more MCP integrations — the flat structure became harder to navigate and extend. The new domain-driven layout solves this.

What's changed:

• Domain modules — the server is now organized by domain: chat, config, mcp, model_provider, oauth, trigger, user. Each domain owns its own api/, schema/, and service/ layers
• Core infrastructure — shared utilities previously scattered across component/ are now consolidated in core/
• Shared package — cross-cutting concerns (auth, middleware, exception handling, logging, HTTP utilities, types) live in a dedicated shared/ package
• Trace ID support — request tracing via contextvars and middleware, making debugging distributed agent tasks easier
• Versioned API routing — all frontend API calls have migrated from /api/... to /api/v1/..., covering chat, config, MCP, triggers, OAuth, user, and model provider endpoints
• Legacy cleanup — removed component/auth.py, component/permission.py, component/stack_auth.py, and component/time_friendly.py

This is the kind of foundational work that makes Eigent a more serious open source cowork platform — one that contributors can navigate confidently and teams can build on sustainably.

🔗 PR: https://github.com/eigent-ai/eigent/pull/1509

---

🔐 Security Auditor: Built-in Skill for Code Security

Every eigent cowork agent now ships with security awareness out of the box.

Thanks to @statxc, reviewed by @a7m-1st and @Pakchoioioi, for contributing skill-security-auditor as a default example skill. This skill gives all Eigent agents a structured, professional-grade approach to code auditing, secrets detection, and vulnerability scanning.

The security auditor skill is a great example of what makes Eigent a compelling cowork alternative: instead of running a separate security tool, you can delegate security reviews directly to an Eigent agent as part of your normal workflow.

What's included:

• 6-step audit workflow — reconnaissance, secrets detection, OWASP Top 10 scanning, dependency audit, configuration review, and auth/authorization review
• Vulnerability pattern library — detailed detection patterns for injection, authentication flaws, cryptography issues, deserialization, SSRF, and file operation vulnerabilities across Python, JavaScript, TypeScript, Go, and Java. Framework-specific checks for Django, Flask, Express, and Spring Boot
• Secrets pattern library — regex patterns for 20+ secret types (AWS, GCP, Azure, GitHub, Slack, Stripe, SendGrid, and more), database connection strings, and private keys — with false-positive reduction rules
• scan_project.py — a full project scanner that checks for hardcoded secrets, dangerous function calls, and insecure configuration patterns. Supports text and JSON output
• scan_secrets.py — a focused secrets scanner with false-positive filtering (skips placeholders, env var references, and test files by default), with --include-tests flag available
• Zero dependencies — both scripts run with stdlib only. No pip installs required

No changes to Electron, frontend, or backend were needed — the skill is auto-discovered from resources/example-skills/ via the existing seedDefaultSkillsIfEmpty() mechanism.

🔗 PR: https://github.com/eigent-ai/eigent/pull/1479

---

📊 Token Usage Clarity in the Chat Header

Small UI change. Big difference in understanding what your agents are consuming.

Thanks to @dataCenter430, reviewed by @a7m-1st, for reworking the token display in the chat header so it's actually informative rather than just a number you have to mentally decode.

Previously, the token count shown next to "Chat" in the header was cryptic — a raw number like 19994406 with no label and no context. Now it's a proper informational element.

What's new:

• Hover tooltip — hovering the token count reveals a "Token usage" label with a short explanation
• Task vs. project breakdown — the tooltip shows both the current task token count and the project total (e.g., "This task: 1,234 · Project total: 5,678")
• cursor-help styling — visual affordance to signal the value is interactive
• i18n support — new locale keys (chat.token-usage, chat.token-usage-this-task, chat.token-usage-project-total) added across all supported languages

When you're running multi-step agent workflows inside your eigent cowork workspace, understanding token consumption at both the task and project level is essential for managing costs and planning capacity.

🔗 PR: https://github.com/eigent-ai/eigent/pull/1501

---

🤖 Claude Haiku 4.5 Model Support

Faster, lighter agents when you need them.

Thanks to @Abishek-Newar for adding Claude Haiku 4.5 to Eigent's model roster. Haiku 4.5 is Anthropic's fastest and most cost-efficient model in the Claude 4 family — ideal for high-volume tasks, quick classification steps, and lightweight agents that don't need the full reasoning power of Sonnet or Opus.

What's added:

• Claude Haiku 4.5 available in the model selector across all agent types
• Full compatibility with existing Eigent cowork agent workflows
• Cost-efficient option for workloads that benefit from speed over depth

Having Haiku 4.5 in the lineup gives you the full Claude 4 spectrum inside your eigent cowork workspace — pick the right model for each job without leaving the platform.

🔗 PR: https://github.com/eigent-ai/eigent/pull/854

---

🔢 Locale-Aware Token Count Formatting

Numbers should be readable. Now they are.

Thanks to @dev-miro26 for a focused but impactful improvement: token counts throughout the app now use Number.toLocaleString() to render with locale-aware thousands separators.

A count like 19994406 becomes 19,994,406 for en-US users, 19.994.406 for de-DE users, and so on — automatically, based on the user's regional settings.

Where it applies:

• Chat header token count in HeaderBox
• History sidebar active and historical project token counts

A small quality-of-life win that makes usage tracking much easier to glance at — especially in longer agent sessions.

🔗 PR: https://github.com/eigent-ai/eigent/pull/1475

---

🐛 Bug Fixes

This release also brings several targeted fixes:

• Fixed trigger action card appearing mid-stream — The "Add trigger action" card now only appears after the assistant's final END message has fully finished typing, preventing premature UI elements while the markdown is still streaming. (#1474 — @kiannidev)

• Fixed long task text overflowing the ChatBox task card — Added min-w-0 to allow the task text container to shrink correctly inside the flex row, and replaced the non-effective overflow-wrap-anywhere class with [overflow-wrap:anywhere] so long task content wraps properly. (#1505 — @fengju0213)

• Reverted native macOS Electron styling — The macOS vibrancy/native window styling introduced in v0.0.88 has been reverted while stability improvements are worked on. (#1512 — @4pmtong)

---

❤️ Community Keeps Raising the Bar

This sprint delivered:

• A full domain-driven server refactor with versioned API routing
• Claude Haiku 4.5 added to the model lineup
• A production-grade security auditor skill, built and contributed by the community
• Smarter token usage display in the chat header
• Locale-aware token count formatting throughout the UI
• Three important bug fixes

v0.0.89 is a release that makes Eigent more capable on the surface and more solid underneath. Whether you're using it as your primary open source cowork tool, evaluating it as a cowork alternative to closed platforms, or contributing to the eigent cowork ecosystem — this release moves the needle in meaningful ways.

Let's keep building.