Eigent Release Notes v1.0.7: Split Sessions, Clearer Task Logs & Reusable Work
Keep Sessions side by side, review what agents are doing, and turn completed Tasks into Skills and Automations
Douglas Lai
Eigent v1.0.7 gives you more ways to arrange, understand, and reuse your work. Open up to three Sessions side by side in a Space, move a Session into its own window, and read a clearer account of the actions behind each Task. When a Task produces a useful result, turn it into a Skill request or an Automation draft from the Session menu.
This release also makes approvals more specific, keeps model choices scoped to their Space, and improves Automation scheduling and run delivery. A new Space overview brings files, Sessions, and recurring work together, while fixes across recovery, sign-in, and quitting help you return to work with its context intact.
🪟 Keep Sessions Side by Side or Give One Its Own Window
Following two conversations should not require switching between them after every reply.
From a Session's sidebar menu, choose Open in → Split view to place it beside the Session already open. Up to three Sessions can share the view. Drag the dividers to resize them, and click or move keyboard focus into a pane to make it the active Session. The layout and pane sizes are remembered per Space.
What's new:
- Up to three Session panes with independent conversations, composers, and previews
- Separate Session windows on desktop, or a new browser tab on the web
- Show window and Bring back here controls for a Session opened elsewhere
- Restored desktop Session windows after restarting the app
- A toggleable Summary column beside the conversation, with a popover when the window is too narrow to dock it
Each detached Session is shown in one place at a time. Cmd/Ctrl+W closes its window, and Open in main window returns you to the main app. Supporting changes coordinate background work across windows and tabs to reduce duplicate Automation, queue, and Remote Control starts.
🔗 PRs: #2067, #2068, #2069, #2072, #2131, #2132
📋 Read the Work Behind a Reply
A useful work log tells you what the agent did in language you can scan.
The default chat view now uses readable action titles such as “Running npm test,” “Searched for OpenAI updates,” and “Updated the plan,” with matching action icons. The work-log header opens while a Task runs and folds when it finishes, while respecting your own choice to open or close it.
Builds with the experimental Narrative timeline enabled also gain Information density controls: Minimum, Moderate, and Full. Repeated tool actions can share one row, or you can show every call. Dev mode switches to the detailed trajectory view. Narrative also distinguishes failed, stopped, and interrupted calls; the default chat's tool rows continue to distinguish running from done.
What's improved:
- Message actions below your request: Copy, Edit and resend, and Expand or Collapse
- A Session title menu for Pin, Rename, End, and Delete
- A wider conversation column with aligned messages, work logs, files, and feedback controls
- Steadier scrolling while follow-up replies stream, plus a Scroll to bottom control
Edit and resend puts a copy of an earlier request into the composer. Your original message remains in history, and an existing unsent draft is preserved. If the earlier request included files, attach them again before sending.
🔗 PRs: #2096, #2097, #2100, #2102, #2123, #2124, #2129
✅ See Exactly What Needs Your Approval
Two file-write requests should tell you which two files they concern.
Approvals and agent questions now sit in a dedicated card above the message box. Approval cards name the file, URL, or command involved, with a Review details section for more context. Command approvals display the command line rather than an internal permission identifier, and the event timeline retains the target after you decide.
When another request follows immediately, Review next request gives you a chance to read it before its decision buttons appear.
What's improved:
- Specific approval targets in live cards, restored cards, and timeline records
- Answerable pending questions after a window reload, when the Run is still waiting
- Questions and answers kept together in the Task's work log after completion and reopening
- Memory change reviews with before-and-after details and Confirm or Reject controls
Question cards also have clearer focus styling, while the disabled message box keeps its saved draft without appearing active.
🔗 PRs: #2095, #2106, #2110, #2119, #2120, #2136, #2141
🧩 Turn a Useful Task into Reusable Work
Once a Task works well, you can use its request and result as the starting point for the next one.
The Session menu now includes Turn into skill and Turn into automation, using the latest completed Task with a result.
- Turn into skill prepares an editable request for the bundled skill-creator Skill. Review it in the composer before sending. If you already have a draft or attachments, Eigent keeps them and asks you to resolve the draft first.
- Turn into automation opens the Automation editor with the Task's request filled in. When the result includes structured Automation details, those help fill the draft too. Review the schedule and required inputs before creating it.
The actions prepare work for review. A Skill request is sent only when you send it, and an Automation is created only when you save it.
🔗 PR: #2099
⏰ Manage Automations Across the Whole Space
An Automation belongs to the Space where you use it, even when you switch conversations.
The Automations list now includes the whole active Space. New Automations receive their own Session, so their work can queue separately from your conversation. Runs waiting behind another Task are labeled as waiting.
What's improved:
- Space-wide discovery that keeps Automations visible when you switch Sessions
- Local-time schedules for new Automations, when supported by the server, so daily and weekly work can retain its local hour across daylight saving changes
- Direct Run now delivery that queues the requested execution without relying solely on a WebSocket notification
- Recovery of recent pending runs after reconnecting and during periodic checks
- Duplicate-run protection when a Session opens or the same execution arrives through more than one delivery path
Run history refreshes while an execution is pending or running, and Not started explains when Eigent did not receive the run in time. Keep Eigent open and signed in on the computer that runs the Automation.
Existing Automations retain their Sessions and UTC schedules. Without server support for time-zone scheduling, new schedules use the previous UTC behavior as well.
🧠 Keep Model Choices with Their Space
Choosing a different model for one Space should leave your other Spaces' choices intact.
A model selected in a Session is now saved as that Space's model for later new Sessions. Only Settings → Models changes the global default, which supplies the starting model for new Spaces. Existing Sessions keep their own selections.
The conversation also shows Model changed to… between Tasks whose recorded models differ. It uses the model that actually ran the Task and leaves unknown identities unlabeled.
What's new:
- Space-scoped model selections with clear fallback messages if a saved model becomes unavailable
- Visible model changes between Tasks in the conversation
- BytePlus ModelArk Managed Agents as a remote sub-agent provider, alongside Gemini Agents API
- Clearer paths when credits run out: Upgrade plan, Use your own API key, and Refresh
Set up ModelArk in Settings → Sub Agents with an API key and existing agent and environment IDs from BytePlus. Follow-ups in the same Eigent Session reuse the remote Session's context and sandbox files. The redesigned Sub Agents page shows configuration and active-provider status, with one provider enabled at a time.
ModelArk tool approvals cannot be answered inside Eigent. Its agent's tool permission policy must be configured in ModelArk for the delegated work.
🔗 PRs: #2061, #2084, #2098, #2134, #2139
🗂️ See Your Space Before Starting the Next Task
On wide windows, a new overview panel sits beside the Workspace composer. See the Space's location, recent files, Session and Automation counts, and Remote Control status. Show version history for a folder Space, or reorder and hide the overview's sections to suit your work. The arrangement is saved per Space on this device.
The Dispatch page also has a clearer Remote Control layout: active links with expiry, Copy link and Revoke link actions, one shared Activity log, and a Channels rail. Channels marked as coming soon remain unavailable.
Starting a new Session feels more immediate, too. Pressing Send opens the local Session while server setup continues. If setup fails, Eigent preserves the request for retry and explains the error.
🔄 Return to Work with a Clearer Record
Several fixes address what happens around interruptions, updates, and long-running Tasks:
- Interrupted work stays inspectable: completed actions, uncertain outcomes, and written files remain available in the Task receipt and Session summary
- Browser-aware Resume: Tasks with recorded browser use can reconnect their browser tools when an authorized Eigent browser is available
- More orderly follow-ups: queued messages wait for the previous Task's status to settle, with fewer repeated busy errors
- Clearer shutdown: quitting checks Tasks that are starting, running, or waiting across Spaces; a restart during approval is recorded as an interruption rather than your rejection
- Terminal environment cleanup: completed Tasks release their environment copies, while interrupted Tasks retain theirs for Resume
- Better return after sign-in: fixes prevent individual Remote Control failures from signing out the whole app and remember the account's last Space
- Offline history deletion: remove Session history when the local backend is unavailable, leaving its files on disk
- Preview and interface polish: PDFs show their filenames and fit narrow panes; dates follow the app language; icons, scrollbars, and composer controls are more consistent
New Tasks and resumed attempts also receive a fresh date, time, and time zone, so an app left open overnight does not keep giving agents its startup date. Permanent model-capability refusals retain their failure reason after a reload.
🔗 PRs: #2088, #2092, #2093, #2094, #2103, #2104, #2107, #2108, #2109, #2111, #2125, #2126, #2127, #2137, #2138
🔁 Upgrade Notes
Split view arranges separate Sessions for viewing. Tasks within each Session continue to follow that Session's queue.
Existing Automations keep their UTC schedules. New local-time schedules require a supporting server; check the schedule preview and next-run time when creating recurring work.
The global model default now applies to new Spaces. Existing Sessions keep their model, and ordinary new Sessions use the Space's saved selection before its Bundle configuration or the global default. The opt-in managed parallel-Session path continues to use the global default.
Information density and Dev mode are available in builds with the experimental Narrative timeline enabled. Readable work-log titles, message actions, and the other default-chat improvements do not require that view.
❤️ More Ways to Work with Your Sessions
Eigent v1.0.7 delivers:
- Split view and separate windows for keeping Sessions within reach
- Clearer task logs, message actions, approvals, and question history
- Skill requests and Automation drafts built from completed work
- Space-wide Automations with stronger scheduling and run delivery
- Space-scoped model choices and ModelArk remote sub-agents
- A Space overview, clearer Remote Control, and more dependable recovery
Thanks to @Douglasymlai for the features and fixes across this release.
Download Eigent to try v1.0.7 and tell us how it fits the way you work across Sessions.
🔗 Release: Eigent v1.0.7
🔗 Full changelog: v1.0.6 → v1.0.7
Let's keep building.
Recent Posts
What Is the Grok API guidelines Parameter? Custom Content Moderation, Explained
Learn what the Grok API guidelines parameter does, which content categories it supports, how to set it, and the request limits and errors to expect.
Eigent
What Is GPT-6.1 Terra? OpenAI's Cheaper Computer-Use Model
GPT-6.1 Terra is OpenAI's fast, cheaper model for computer use, visual reasoning, and coding. Here's what's new, the pricing, the benchmarks, and how to access it.
Eigent
Kimi K3.1 Pro: What's New, Benchmarks, and How It Compares
Kimi K3.1 Pro is Moonshot AI's new open-weight MoE model with a 1M-token context. See its benchmarks, coding and finance gains, open-weight date, and how it compares.
Eigent
