What's new in Tejus.

Tejus no longer imports cookies from Chrome, Edge or Safari into its built-in browser. The component behind it read other browsers' saved cookies, which company antivirus flags as a password stealer, so it has been removed from the app.

Sign in to sites inside the built-in browser instead; saved logins in the browser vault work as before.

Long agent turns use less CPU and memory, parallel threads no longer flood the app with refreshes, and the old "server busy" popup is gone.

Running subagents show above the composer, and a subagent's thread shows the model it actually runs.

Switching a thread to another agent goes through Hand off and continues without replaying work; Codex threads resume with their full context.

Start a new thread and run your project's actions from the chat top bar, and close a chat tab or the docked terminal with Ctrl+W.

An empty chat shows a subtle animated background — search Settings for "ASCII background" to turn it off.

Steadier Stop, readable errors, project logos as project icons, Git diffs next to nested repositories, faster terminals on busy machines, OpenCode sign-in and Bun security updates.

Settings → About Tejus shows your version and whether an update is ready, with one button to check, download or restart.

New versions download in the background. When one is ready, a window in the middle of Tejus shows what's in it and restarts into it with one click — or choose Later and update from the rail, the notification bell or Settings.

Usage reads the same everywhere: every ring, bar and label shows how much of a limit you have used, the same way as Plan & Usage.

Settings → About Tejus introduces Prasanvi, the company behind Tejus, with support and legal links.

The full changelog in the app now lists every Tejus release, and Docs links open tejusai.com/docs.

Long conversations open, switch and scroll faster, and reloading a chat restores the whole conversation, including rollbacks made while you were disconnected.

Provider sessions are more dependable, reconnecting no longer repeats work, and Stop also pauses the follow-ups you queued.

When an agent starts subagents, open them from the Environment panel and follow their work without leaving the conversation.

Workspace search finds more, OpenCode v2 is supported, provider setup is more reliable, and many smaller fixes land across the app.

Tejus 1.0 is a new app: an agent workspace where several AI coding agents work side by side in threads and projects, with a full IDE built in — one install, one account.

Use the agents you already have — Codex, Claude, Cursor, OpenCode, Droid and more — next to Tejus Cloud models and your own API keys, all from one model picker.

A Background tasks panel shows what every agent and terminal is doing, in the workspace and in the IDE.

Security review for every agent: Tejus checks the changes from each turn and shows findings in the thread and the notification bell, with an optional fix-it-yourself pass for the agent.

Open any folder as a project: explorer, editor tabs, search, Source Control with Git history, terminals and web preview, with the chat on the right.

A familiar menu bar — File, Edit, Selection, View, Go, Terminal, Help — and the keyboard shortcuts you already know.

Live explorer: files and folders created by an agent, a terminal or another app appear right away, and Source Control updates on its own after a commit.

The chat knows what you are looking at: the open file and your selected lines go with your message. Dismiss the file pill to leave a file out.

Switch between the chat and the IDE without losing anything — running chats keep going, and new chats started in the IDE work in its folder.

The first launch after installing or updating no longer freezes for 10+ seconds; the window opens right away.

No more lag when you switch windows: Tejus stops re-checking every AI provider each time, and tools you have not installed are not probed again and again.

A dropped connection no longer shows raw WebSocket errors in the editor — your text is kept and saving resumes on its own.

Download Tejus 1.0 from this page today — the installer is now simply tejus-setup.exe.

From 1.0 on, new versions download in the background and install with one click when you restart, like Cursor.

Already on 0.9? The in-app update to 1.0 comes soon. Your saved API keys come along; you sign in once more with Google.

One slim title bar: the ≡ menu, your tabs, search, notifications and window controls now share a single row, so the workspace gets more room.

A calmer, flatter design language across every screen — frosted menus, soft cards, clearer active states — with real provider logos in the model picker.

Pick your text size and density in Settings → Themes; the whole interface scales with it.

Tejus now uses your system font and a new icon set, so it feels at home next to your other apps.

Frameless window on Windows with the native shadow, and no flash of the wrong colour on launch — including with custom themes.

One consistent motion language everywhere, and animations turn off when your OS asks for reduced motion.

Faster to start and switch: editor themes and file icons load only when needed, and hidden panes no longer cost anything.

Untitled and extension-less files now get syntax highlighting, hints and completions — Tejus detects the language from the content, and a new language picker in the status bar lets you set it yourself.

Minimap on the right edge of the editor (Settings → General to turn it off), plus a visible scrollbar.

Images open as a real preview — size, dimensions, checkerboard for transparency — instead of a “binary file” notice.

Edit → Cut, Copy, Paste and Selection → Select All are in the menus.

Search inside files: flip the toggle next to the search box from NAMES to IN FILES and jump straight to the matching line.

Right-click gained Rename (F2), Duplicate, Cut, Copy and Paste; Ctrl+X/C/V work on the selected file or folder.

Pick your shell: PowerShell 7, Windows PowerShell, Command Prompt or Git Bash from the + menu, with a default in Settings.

If a shell can’t start, the terminal tells you why instead of staying blank.

The + button now stays put when you have many tabs open, and every toolbar button explains itself on hover.

Settings is in the File menu, and the Go, View, Terminal and Help menus have real entries.

Clone a repository straight from the Welcome screen.

Switch or create branches, and add or change remotes (GitHub, GitLab, Bitbucket or any Git host) from the new ⋯ menu.

Recent commits are listed in the panel; error messages are shown in full — and if Git doesn’t know who you are yet, one click sets your name and e-mail.

Generate commit message now keeps what you already typed as the intent.

Fewer false “Something went wrong” notices, and a Send report button when one is real.

Account e-mails carry the Tejus logo.

DeepSeek models could reject a chat outright once the conversation contained a reasoning-only reply. Those empty turns are now filtered out — in the app and on our servers, so the fix reached everyone immediately.

The Retry button on a failed chat had an invisible label — it worked, you just couldn't read it.

When Tejus Cloud declines a request for a reason that won't change (daily limit reached, capacity, provider account), the app now reports it straight away instead of retrying three times first.

Tejus Cloud now has a fair-use ceiling across all users: if the free tier is exhausted for the day, you get a clear message and a pointer to your own API key or a local model — never a raw provider error.

Update checks and optional diagnostics are far less chatty, so the app is a better citizen on metered connections and our backend stays fast as the beta grows.

Kimi is temporarily off the cloud model list while its provider account is restored; DeepSeek V4 Flash and V4 Pro are unaffected, and your own API keys always work.

Google sign-in hardened end to end: every callback must echo a one-time cryptographic check, and the token-injection path that browsers could poke at is gone entirely.

Workspace trust: the first time you open a folder, Tejus asks whether you trust it. Untrusted folders never auto-run project scripts (typecheck/lint/test after agent edits) and disable auto-approve — so opening someone else's repository is safe by default.

MCP tools now always show an approval card before running, like every other mutating tool.

The agent treats file, terminal, web and tool content strictly as data — instructions hidden inside a repository or web page are ignored and flagged.

The web-fetch safety filter now also blocks IPv4-mapped IPv6 and NAT64 addresses that could have reached local services.

No more white screens: if a panel crashes, Tejus shows a recovery card with Reload — and you can send us a crash report with one click (only when you click; nothing is sent automatically).

New in the Help menu: Send feedback… and Email support… — tell us what broke without leaving the app.

The daily free limit is now enforced accurately even when many requests run in parallel, with clearer messages and a per-user rate limit that keeps the free tier fair for everyone.

SHA-256 checksums and a signing notice on the download page, so you can verify what you install.

Screen readers now announce streaming AI output and status changes; animations respect your reduced-motion preference.

Google sign-in no longer fails when the account picker or 2FA takes a while — the app now waits for you.

New Settings → Account page: sign in and out, see your plan and today's usage at a glance.

New accounts receive a branded activation e-mail; Tejus Cloud unlocks the moment you click the link.

In-app notifications when anything changes on your account: quota granted, plan changed, a new cloud model available.

Switching projects switches everything: chats are per project, and security findings reset and rescan for the new folder.

Restore previous session — the last project and its open files reopen on launch.

A Retry button on chat errors, friendlier error messages, and automatic retry on flaky connections.

Attach images to chat (button, drag & drop, or paste) for vision-capable models.

File → New Window opens a fresh, empty window.

Adding your own API key automatically selects that provider's recommended model; unavailable models are clearly marked.

Security panel: download the full report, fix all findings with one click, and see a summary when a scan completes.

Scans run in the background and only on the open project — no more freezes.

The app process is now Tejus.exe, and every remaining internal name has been rebranded.

Installers are now digitally signed (Authenticode) in addition to the updater's cryptographic signature — SmartScreen and antivirus warnings are gone.

In-app updates verify both signatures before installing.

Plan mode can use its own model (Settings → Models) with a safe fallback chain — an unavailable choice can never break a run.

Anthropic (Claude) with your own key fixed: requests now route natively, sidestepping a browser CORS limitation.

Cloud models added in our catalog appear in the app within a second — no update needed — and new ones carry a NEW badge.

Fresh visual identity: the new Tejus mark across the app, installer and website.

Update notifications appear once per release as a toast, then live quietly in the notification bell with full changelogs.

Moonshot AI (Kimi) is now a built-in provider — add your Moonshot API key in Settings → Models and use Kimi models alongside the existing BYOK providers.

Auto mode: a toggle beside Agent/Plan in the AI panel lets agents run mutating tools — file edits and shell commands — without per-action approval cards.

Amber pulsing indicator while Auto is on; the pill stays in sync with Settings → Agents → Bypass permissions (flip either, the other follows).

Safety rails stay on in Auto mode: the sensitive-file deny-list (.env, keys, credentials) still blocks reads and writes.

Windows installer refreshed (~66 MB NSIS). macOS and Linux builds still to follow.

Initial public beta of Tejus — a Tauri 2, terminal-first AI-native workspace.

Windows x64 NSIS installer (~66 MB) available now. macOS and Linux builds ship next.

No telemetry. API keys stored via the OS keychain on macOS and Windows.

Built on Tauri 2 + Rust — native PTY, no Electron runtime.

Multi-tab terminal with a WebGL-rendered xterm.js surface.

Native PTY backend with support for zsh, bash, pwsh, fish, and cmd.

Split panes, inline search, link detection, and true-color output.

Shell integration for cwd and prompt markers; WSL supported as a workspace environment.

Private terminal tabs with AI-context redaction.

CodeMirror 6 editor with multi-language support (TypeScript, JavaScript, Rust, Python, Go, C/C++/Java/C#, PHP, HTML/CSS, JSON, Markdown, and more).

Inline AI autocomplete and reviewable AI edit diffs.

First-class Vim mode with motions, registers, and marks.

Prebuilt editor themes (Tokyo Night, Nord, GitHub, Gruvbox, and more).

Multi-agent and sub-agent workflows with tools for file read/write/edit, bash, search, and plan.

BYOK for OpenAI, Anthropic, Google, Groq, xAI, Cerebras, OpenRouter, DeepSeek, Mistral, and OpenAI-compatible endpoints — or local via LM Studio, Ollama, and MLX.

Voice input via OpenAI Whisper (requires an OpenAI API key), slash commands, skills, MCP servers, and per-project memory via TEJUS.md.

Semantic codebase indexing, @codebase attachments, and auto-compact for long conversations.

Source control panel with stage, unstage, commit, and branch actions.

Commit graph history and per-file diffs without leaving the workspace.

Auto-detected local dev server preview in a sandboxed iframe.

Image and PDF viewers alongside the editor.

Opt-in Security review: live local pattern scanning as you type, plus AI deep-review on agent-written files.

Advisory or Agent self-fix — critical issues can be fed back to the agent for correction before you approve (never hard-blocks).

Optional AI code review of git-changed files for logic and correctness — always asks before spending tokens.

Hardened AI tool surface for filesystem, network, and IPC — mutating tools require approval.

SSRF and DNS rebinding defenses on outbound HTTP; sandboxed preview and path guards for sensitive files.