rumus rumus
← Back to Community
R

📝 Changelog

Rumus · February 5, 2026 at 12:32 PM
Announcement

Rumus v0.1.9

Bug Fixes

  • Fixed an issue where the window did not follow the mouse while dragging

  • Fixed the Esc key not responding after the terminal entered interactive mode

Rumus v0.1.8

Features

UI & Experience

  • Both left and right sidebars are now resizable — drag the edge to adjust width

  • Double-clicking the title bar now correctly toggles maximize/restore

  • Simplified the onboarding flow by removing the secret key setup step

  • Updated application icon

AI / Agent

  • Removed the interactive command dispatch to terminal — the agent no longer sends sudo/ssh commands to the user's terminal for manual execution

Linux

  • Automatic terminal renderer fallback to Canvas on X11

  • Acrylic theme is now disabled on X11 with automatic fallback to Dark theme

  • Added missing desktop icon sizes

Bug Fixes

  • Host management dialog now correctly shows the secret key recovery flow instead of PIN unlock when the vault is not initialized

  • Fixed an issue where secret key files could sometimes fail to load during key recovery

  • Fixed title bar double-click not working

Rumus v0.1.7

Features

AI / Agent

  • Interactive commands dispatched to the terminal now end the agent's tool loop, allowing users to complete the interaction before the agent continues. The agent prompts users to notify it once they're done.

UI & Experience

  • Animated skeleton bubbles (bouncing dots) now appear while waiting for the AI's first token, replacing the previous "Generating..." text in the header

  • Unified all dialog scrollbars for consistent styling across Keychain, Snippets, Accounts, Known Hosts, and Snippet List dialogs

  • Optimized the host edit dialog layout — icon color picker merged inline, Shell Type moved from advanced settings to the basic info tab

  • Improved the onboarding flow logic

Bug Fixes

  • Fixed keyboard shortcuts not working in Vim mode

  • Fixed an error when dragging non-terminal tabs to a new window

  • Fixed an issue where backslashes were stripped from Windows paths, preventing terminals from launching

  • Fixed host management reading the key file path instead of the file contents after selecting a private key file

  • Fixed SSH sessions freezing during data‑intensive transfers.

Rumus v0.1.6

Bug Fixes

  • Fixed an issue where the PATH environment variable failed to load correctly on macOS. (#18; thanks Robot)

  • Optimized the display of the autocomplete menu.

  • Fixed an issue where pressing Tab failed to trigger completion when the input exactly matched a suggestion.

Rumus v0.1.5

Features

AI / Agent

Plan Mode

Plan Mode enables the AI to break down complex tasks into structured, multi-step execution plans before taking action. Instead of immediately running commands, the agent first creates a plan outlining its approach — complete with ordered steps, reasoning, and a clear goal. The plan is presented to you for review and approval before any execution begins. As the agent works through each step, progress is tracked in real time with visual indicators (pending, in-progress, completed, failed).

Plan Mode is always available — the AI automatically decides when a task is complex enough to benefit from a plan. No manual toggle is needed.

Sub Agent

Sub Agent (also called Sub Task) allows the AI to delegate complex subtasks to independent child agents. When the primary agent encounters a problem that benefits from focused research, risk assessment, or parallel investigation, it spawns a sub agent that operates autonomously with its own context — while forwarding all tool calls and outputs back to the main conversation for full transparency.

Sub Agent is always available — the AI decides when task delegation is appropriate. No manual toggle is needed.

MCP (Model Context Protocol)

MCP is an open standard (originally introduced by Anthropic) that lets AI agents connect to external tools and data sources through a unified protocol. In Rumus, you can add MCP-compatible servers (e.g., filesystem tools, database connectors, API wrappers) and the agent will automatically discover and use their tools during conversations.

To configure MCP: go to Settings → AI → MCP, click Add, and paste a server configuration in JSON format. Each server has an enable/disable toggle. When enabled, the server connects automatically when the AI sidebar opens.

Skills

Skills are reusable instruction packs that extend the AI agent's knowledge and behavior. Each skill consists of a name, description, and detailed instructions. The agent sees all enabled skill names and descriptions, but only loads the full instruction content on demand — keeping token usage efficient while giving the agent access to specialized knowledge when needed.

To manage Skills: go to Settings → AI → Rules and Skills. Click Add to create a new skill, or upload a skill file. Each skill has an enable/disable toggle.

Cloud Web Search

Cloud Web Search uses the Rumus Search API to provide the agent with faster, more reliable web search results compared to the default local DuckDuckGo scraping. Search results include richer metadata such as extra snippets, content age, and site information.

To configure: go to Settings → AI → General and toggle Cloud Web Search. Enabled by default. When disabled, the agent falls back to local browser-based DuckDuckGo Lite search.

Other AI Improvements

  • Parallel tool execution — multiple independent tool calls now run concurrently

  • Automatic retry on HTTP 524 timeout errors

  • Full tool_calls and tool_results context now persisted and restored when reopening sessions

  • Conversation rename request delayed 500ms to avoid racing with the main generation request

  • Tool targetName persisted to database for display continuity

  • tool_call / tool_result message pairing guaranteed in streaming

Workspace

  • Bidirectional workspace sync — remote changes now automatically apply to local state

  • The monitoring card supports monitoring of the currently running computer

UI & Experience

  • We have re-optimized the display effects of acrylic themes across various systems

  • Now, under the acrylic theme, you can manually control the transparency of the background color (#17; thanks donny.rui)

  • Acrylic theme color refinements and removal of redundant background overlays

  • Sidebar now includes Files / Scripts tab switcher with a script list and file upload button

Linux

  • Added manual download update prompt for deb/rpm installation users

  • Optimized startup speed

  • SwiftShader software renderer detection with automatic Canvas rendering fallback

  • Removed Nvidia X11 WebKitGTK DMA-BUF environment variable workaround

  • Removed Linux acrylic noise texture effect

  • Clipboard fix for Linux

Bug Fixes

  • Fixed tool execution status being lost across sessions

  • Fixed Zod v4 tool parameter serialization issue

  • Fixed user_info missing causing session restore failure

  • Fixed deleted rules still being injected into system prompt

  • Fixed Progress card status icon stuck on loading

  • Fixed autocomplete menu not closing on Esc

  • Fixed free-tier users hitting "Database not initialized" and "BadFileKey" errors on first PIN setup (#14; thanks WEP)

  • Fixed bash terminal color rendering, web_search parameter compatibility, and log optimization

  • Fixed Nvidia GPU auto-detection and WebKitGTK DMA-BUF workaround on Linux

Rumus v0.1.4

Features

AI / Agent

  • Rebuilt the Agent core architecture from the ground up

  • Added image support for Agent conversations, including paste-to-upload and zoom controls

  • Introduced token-based context management using js-tiktoken, replacing the fixed message count approach

  • Added Max Mode setting — when disabled, context is limited to 20k tokens; when enabled, uses the model's full context length

  • Automatic LLM summarization when context usage reaches 80% threshold

  • Added auto-naming toggle for conversations — choose between AI-generated titles or using the first message

  • Hide inline completions when the terminal enters interactive mode

  • Introduced multi-terminal support with per-tab isolated chat sessions

  • Each tab's agent now retrieves its own terminal context instead of sharing global state

  • Agent tools support a target parameter to execute commands in a specific terminal

  • Sidebar history automatically filters by the current tab's profile, host, or workspace

  • Detached mode: viewing history from a different context disables command execution

Workspace

  • Added executeInTerminals / executeInAllTerminals for batch execution across multiple terminals

  • Workspace-aware AI agent context retrieval and command execution

  • Added onboarding prompt when creating the second card in a workspace

Auth & Account

  • Added a note on the login onboarding step explaining that sign-in is required during beta

UI & Experience

  • Refined the chat input: removed Markdown preview, moved zoom buttons to the top-right corner, added slide transition animation, replaced the paper-plane send icon with an arrow

  • Added a ligature hint in terminal font settings to inform users about automatic ligature rendering

  • Added a first-run onboarding flow

  • Improved community link entry points

Linux

  • Improved overall display and styling on Linux

Bug Fixes

  • Fixed frequent auth session expiration

  • Fixed access token not being refreshed on a timer

  • Fixed errors during auto-naming and summary generation with built-in models

  • Fixed cross-device sync resetting Max Mode and auto-approve tool settings

  • Fixed CWD tracking using a stale static value (#11; thanks Wang zixu)

  • Fixed default model ID format not matching built-in model ID format

  • Fixed selected model being reset after refreshing the model list

  • Fixed terminal tab session never being created, preventing message sending

  • Fixed newly created sessions missing terminal context and session info callbacks

  • Fixed error when opening chat history with no session created

  • Fixed browser engine download dialog appearing when there is no active session

  • Enabled opening historical conversations directly from the Chat History page without switching to a terminal tab

  • Fixed Agent thinking process not being displayed

  • Fixed garbled Chinese input on macOS in production (#6; thanks Wang zixu)

  • Fixed missing syntax highlighting for command output on Linux

  • Changed all path comparisons to case-insensitive to fix replacement issues on case-mismatched paths (#8; thanks Wang zixu)

  • Fixed detected bash shell being discarded during initialization

0 replies 258 views

0 Replies

Sign in to reply