Eigent Release Notes v1.0.6: Models, Automations & More Reliable Sessions

Eigent v1.0.6 gives Models and Automations dedicated views for configuring and inspecting your work. Saved model configurations are easier to manage and select. Automations puts instructions, schedules, and actual run history together, with starter examples to help you set up recurring work.

This release also adds a Profile step to onboarding and addresses the interruptions that get in the way of longer Tasks. Approvals remain actionable after recovery, completed results appear more consistently, and switching Sessions preserves your unsent draft and terminal history.

---

🧠 Manage Your Models in One Place

Using the same model through two different API keys should not make those configurations interchangeable.

Models is now a top-level collection in Settings. You can add, edit, remove, and validate saved configurations, inspect their availability, and control which ones appear in the model picker. Each configuration keeps its own credentials and identity, so a Session pinned to one setup does not silently switch to another setup's API key.

What's new:

• A Models collection with saved configurations, search, availability, and visibility controls
• Provider-grouped selection in the input box, with thinking effort kept beside your message
• Distinct saved setups for defaults, Sessions, and Workers, even when the provider and model names match
• Retryable loading when the provider list cannot be retrieved, while existing choices remain visible during refresh

The release also fixes thinking-effort errors that prevented bring-your-own-key models from starting, including inside Spaces with a Bundle installed. For models whose supported efforts are unknown to Eigent, the provider's default is used. Recognized reasoning models use a supported level when the requested one is outside their known range. Registered models retain their capability checks.

Thanks to @Douglasymlai for the Models redesign, with contributions from @4pmtong, who also fixed bring-your-own-key thinking-effort handling.

🔗 PRs: #1954, #2050

---

⏰ See What Your Automations Actually Ran

A schedule tells you when work should happen. Run history tells you what happened.

The redesigned Automations page starts with six shared examples and keeps a compact automation list beside the selected item's details. Each detail view brings together its instructions, schedule, next run, and execution history. Expand a run to inspect stored output or the reason it failed or was missed.

What's new:

• Starter examples to help you create routine Tasks
• A detail view with instructions, schedule, next run, and actual run history
• Direct controls to run now, pause or resume the schedule, edit, and delete
• History that stays readable while background updates refresh existing runs

Run now is available even when an Automation is paused. It queues a manual execution without turning the schedule back on.

Scheduling fixes also improve date handling across month and year boundaries, reject invalid hours and minutes, and preserve stored schedules containing ranges or lists when the editor cannot represent them.

Thanks to @hy5kate for the Automations redesign, with contributions from @Douglasymlai and @4pmtong. Thanks to @4pmtong for the additional schedule fixes.

🔗 PRs: #2016, #2033, #2051

---

👋 Choose Your Work Profile

First-launch onboarding now includes a Profile step. Choose from 18 work roles, including Engineering, Finance, Design, Student, and Founder, or use Others. You can skip the step and change your choice later in Settings → General → Profile.

The role is a local preference in this release. It does not change agent behavior or select different Automation examples.

Thanks to @hy5kate for the onboarding and Settings update, with contributions from @Douglasymlai.

🔗 PR: #2009

---

🔄 Recover a Session and Answer Its Next Request

Returning to a Session should leave you with a clear next action.

v1.0.6 fixes cases where an approval arrived after a window reload but stayed hidden, or a new approval after Resume was treated as read-only history. Pending checks retry after a timeout, and finished approvals remain historical records.

If the app restarts while an agent is waiting for your answer to a question, the Task is now marked interrupted. Resume lets the agent ask again instead of leaving the Session waiting for a process that no longer exists.

What's improved:

• Actionable approvals after Resume and window reloads
• Clearer stop reasons that distinguish time limits, cancellation, expired requests, and other interruptions
• Accurate recovery results so a successful Resume no longer leaves the turn marked failed because of its earlier recoverable error
• Model outage messages that explain when the model is unavailable

When another Task is holding a Space after an interrupted file operation, Eigent now explains the wait and can link to the blocking Session. If a command's outcome is unknown, Resume stays unavailable and the app explains how to review the files and cancel the interrupted Task.

Thanks to @4pmtong for the recovery, approval, and Task-status improvements.

🔗 PRs: #2012, #2020, #2021, #2022, #2026, #2042

---

📂 Keep Drafts, Terminal History, and Finished Files Within Reach

Switching Sessions now preserves each Session's unsent message and any review comments added to it while the app is running. Return to the Session and your draft is there. Terminal tabs also retain their older output and scroll position when the preview panel opens and closes.

Completed work is easier to inspect, too. Recovered final answers and Task review pick up the recorded output files, and local files open correctly from Summary → Files. PDFs render in the Files panel, with an external-open option if an embedded preview fails.

Version history now includes retained checkpoints from Tasks that worked directly in the Space's folder or a selected worktree. Those changes can appear in Session and Task versions instead of only in the technical timeline.

Thanks to @4pmtong for draft preservation, terminal history, file previews, and checkpoint visibility.

🔗 PRs: #1969, #1991, #1992, #2017, #2043, #2044

---

📦 Check Bundles Before Sharing and Recover Failed Imports

Saving a Space as a Bundle now checks its Skill, MCP, model, and file references on your device before publishing. The dialog lists problems and lets you check again. Unresolved references can stay in a draft, while publishing waits for blocking issues to be resolved. These checks cover the author's device; recipients may still need to configure their own access.

A failed import also has a way forward. Where cancellation is available, you can cancel it and review another Bundle while retaining the Space draft, saved local values, and files already written. An installation actively writing files cannot be cancelled through this action.

Thanks to @4pmtong for Bundle reference checks and import recovery.

🔗 PRs: #1995, #1998

---

🔁 Upgrade Notes

Session drafts are kept in memory while the app runs. They survive Session switches, but are cleared when the app restarts.

Recurring Automations continue to use fixed UTC schedules. Their local display time can shift when daylight saving time changes. Stored ranges or lists that the schedule editor cannot display are preserved when you save other edits.

Interrupted commands with unknown outcomes are not automatically rerun. Review the Space's files and follow the recovery message before continuing.

---

❤️ More Control Over Everyday Work

Eigent v1.0.6 delivers:

• A Models collection for managing saved configurations and selecting providers
• An Automations dashboard with starter examples, schedules, and inspectable run history
• A Profile choice in onboarding and Settings
• More dependable approvals and clearer Session recovery
• Drafts and terminal history that survive Session switches
• More consistent file previews and visible Task checkpoints
• Bundle reference checks and recovery from failed imports

Download Eigent to try v1.0.6, and tell us how it works with your models and recurring Tasks.

🔗 Release: Eigent v1.0.6

🔗 Full changelog: v1.0.5 → v1.0.6

Let's keep building.