CyberCode v1.1.22

CyberCode v1.1.22

Highlights

• Makes native desktop conversation switching and history anchors respond reliably while long transcripts are still settling.
• Adds a default model or published route choice for newly created conversations.
• Separates self-evolution memory into user understanding, current-project experience, and cross-project methods.
• Reduces repeated prompt and tool-schema overhead while preserving provider compatibility.
• Makes the expanded thinking and command activity panels more compact without losing live auto-follow.

Conversation Navigation

• Prepares local history and anchor metadata in the background so recent conversations can switch from cache.
• Keeps history anchors hidden until the local transcript and its rendered targets are ready.
• Re-centers an anchor through delayed WebKit layout changes and immediately hands scrolling back to the user on wheel, touch, pointer, or keyboard input.
• Keeps local history failures isolated from network state and preserves a retry path without discarding cached messages.

Chat Activity

• Keeps thinking and command activity aligned with the shared chat content column.
• Limits both expanded activity bodies to a compact 64 px viewport with internal scrolling.
• Continues following the latest streamed thinking and the newest running command unless the user deliberately scrolls upward.
• Refines spacing between user turns and all assistant-owned content, including thinking and tool activity.

Models And Routing

• Adds a localized setting for the model or published route used by new conversations.
• Preserves the existing behavior for users who leave the new default unset.
• Keeps provider and model choices synchronized without overwriting an explicit session selection.

Memory And Prompt Efficiency

• Stores reusable current-project experience separately from durable user preferences and global cross-project methods.
• Reviews ordinary conversations every six user messages and periodically distills bounded project evidence into global methods.
• Compiles communication, engineering, model, optimization, and user rules into one deduplicated prompt policy.
• Selects only the tool schemas relevant to the current request and falls back safely for providers without native tool references.
• Adds stable OpenAI and Codex prompt-cache affinity and preserves usage accounting across Chat Completions and Responses transports.

Verification

• Core suite: 164 files and 1,704 tests passed.
• Desktop suite: 130 files and 1,040 tests passed.
• Sidecar release suite: 20 tests passed.
• TypeScript checks and the production desktop build passed.
• Release packaging validates macOS ARM64, macOS x64, Windows x64, and Linux x64 assets, including a ZIP download for every platform and a portable Windows ZIP.

CyberCode v1.1.21

CyberCode v1.1.21

Highlights

• Adds a small blue completion indicator to conversations that finish in the background.
• Keeps the currently selected conversation free of redundant completion indicators.
• Moves the cached follow-up bar closer to the message composer for a tighter, more natural workflow.

Background Completion Indicators

• Marks a conversation only after its full assistant turn, tool activity, and background Agent work have genuinely settled.
• Uses the same stable completion boundary as the completion sound, preventing premature indicators during streaming or reconnection.
• Shows the blue indicator only when another, unselected conversation finishes.
• Clears the indicator immediately when the conversation is selected or when the user sends or queues a new message in it.
• Includes localized accessibility labels in Chinese, English, Japanese, and Korean.

Cached Follow-ups

• Reduces the visual gap between the cached follow-up bar and the message composer.
• Preserves the existing composer position and spacing when no cached follow-up is present.

Verification

• Desktop test suite: 130 files and 1,027 tests passed.
• TypeScript checks and the production desktop build passed.
• Release packaging produces macOS ARM64, macOS x64, Windows x64, and Linux x64 assets, including ZIP downloads for every platform and a portable Windows ZIP.

CyberCode v1.1.20

CyberCode v1.1.20

Highlights

• Publishes a cached follow-up into the conversation immediately when the user chooses Join task, while removing it from the cache bar in the same interaction.
• Cleans invisible and pathological whitespace from copied thinking content without flattening normal paragraphs or code indentation.
• Stops completed Agent launch receipts from continuing to show a running shimmer after the assistant turn has ended.

Conversation Experience

• Uses an optimistic, atomic update for immediate steering: the current assistant text is finalized first, then the new user message appears in the correct timeline position with its images and files intact.
• Keeps delivery tracking internal so queued work still blocks premature completion sounds without leaving a visible cache row behind.
• Prevents queued and processing acknowledgements from restoring the cache row or creating duplicate user bubbles.
• Restores a rejected message to the cache as a retryable failure and removes the optimistic bubble when the server explicitly refuses delivery.
• Preserves compatibility with older sessions that only receive the final steering acknowledgement.

Thinking And Tool Activity

• Normalizes isolated carriage returns, Unicode spaces, zero-width characters, repeated blank lines, and abnormally long space runs before thinking text is displayed or copied.
• Preserves meaningful paragraph breaks and ordinary code indentation in thinking details.
• Treats an Agent launch receipt as active only while its assistant turn is genuinely active, removing stale running highlights once the turn finishes.

Verification

• Desktop test suite: 130 files and 1,023 tests passed.
• TypeScript checks and the production desktop build passed.
• Release packaging produces macOS ARM64, macOS x64, Windows x64, and Linux x64 assets, including ZIP downloads for every platform and a portable Windows ZIP.

CyberCode v1.1.19

CyberCode v1.1.19

Highlights

• Fixes premature task completion so the send button, completion sound, and completed state appear only after the full AI turn, including follow-up reasoning and tool work, has actually finished.
• Prevents old streamed events from being replayed as new chat bubbles after an SDK reconnect, while keeping automatic recovery for real network interruptions.
• Automatically compacts and retries once when switching models causes an existing conversation to exceed the new model's context window.
• Refines the conversation timeline with stable thinking and tool summaries, clickable message anchors, and clearer live execution feedback.

Chat Reliability

• Adds one authoritative turn-completion gate across the server and desktop state so temporary idle or result events cannot finish an active task early.
• Keeps the stop control active while queued messages, tools, background work, or follow-up output are still pending.
• Plays the selected completion sound only after the final turn settles, with built-in bell, ding, and knock choices.
• Extends idle WebSocket handling to match the SDK keepalive interval and deduplicates UUID-bearing events after reconnects.
• Preserves a replacement SDK socket when an older connection closes and suppresses late output from stopped turns.
• Monitors bundled child processes and closes orphaned sidecars cleanly when their desktop owner exits.

Conversation Experience

• Keeps the current thinking and tool activity directly below the latest assistant bubble throughout a turn, then folds both only when the turn is complete.
• Consolidates repeated command, read, edit, and tool activity into one summary and omits zero-count categories.
• Gives expanded thinking and tool details bounded, scrollable areas with consistent headers, sizing, and animation.
• Adds a compact message anchor rail with hover previews, reliable click navigation, active-anchor highlighting, and a smooth two-pulse highlight on the destination user bubble.
• Aligns message actions near the lower edge of long bubbles and keeps conversation controls centered with the composer width.

Context And Sessions

• Detects provider prompt-too-long responses, performs automatic context compaction, and retries without trapping the conversation in an error state.
• Persists server-side message anchors for long sessions and restores scrolling without losing the active conversation position.
• Improves project/session list reuse and desktop runtime request handling to reduce repeated loading during navigation.
• Adds clearer localized status and completion-sound settings in Chinese, English, Japanese, and Korean.

Verification

• Passed 1,648 root service and integration tests, including real WebSocket, reconnect, image fallback, prompt compaction, and turn-completion coverage.
• Passed 1,017 desktop tests and 20 sidecar packaging tests.
• Passed TypeScript checks, the desktop production build, Rust checks, and release-diff validation.

Downloads

• Publishes macOS ARM64, macOS Intel, Windows x64, and Linux x64 builds.
• Includes a standard ZIP download for every platform alongside native installers and updater assets; Windows also includes a portable ZIP.

CyberCode v1.1.18

CyberCode v1.1.18

Highlights

• Adds practical vision-aware routing so a text-first route can keep using its default model while image requests temporarily switch to a configured vision model.
• Replaces error-prone free-form task and modality values with localized choices in both the route node and its inspector.
• Keeps route definitions stable after image requests, so temporary capability routing never changes the user's default text model.

Routing

• Adds task choices for vision, coding, reasoning, audio, and general requests.
• Adds modality choices for image, text, and audio inputs.
• Resets incompatible condition values when switching condition types instead of silently preserving an invalid value.
• Preserves free-form values for numeric and operational conditions such as context, cost, health, and quota.
• Provides Chinese, English, Japanese, and Korean labels for every new routing choice.

Verification

• Passed 1,624 root service tests, 951 desktop tests, and 19 sidecar packaging tests.
• Passed TypeScript checks, the desktop production build, Rust checks, multilingual documentation build, route sidecar smoke tests, and GitHub Actions workflow linting.
• Added a live proxy regression proving text requests use the text-only provider, image requests use the vision provider, and the published route remains unchanged.

Downloads

• Publishes macOS ARM64, macOS Intel, Windows x64, and Linux x64 builds.
• Includes a standard ZIP download for every platform alongside the native installers and updater assets.

CyberCode v1.1.17

CyberCode v1.1.17

Highlights

• Adds a complete visual routing workspace with editable Agent, condition, distribution, parallel, result, relay, model, and output nodes.
• Makes provider/model capabilities adaptive instead of relying on brittle hard-coded model lists, while keeping image uploads usable for tool-assisted processing on non-vision models.
• Expands Agent Node into a practical multi-key gateway with human-readable provider aliases, separate model and route pickers, compact connection examples, and localized documentation.
• Adds localized Skill and Plugin marketplaces with search, categories, popularity and recency sorting, official icons, installation flows, and clearer evolution visibility.
• Introduces built-in CyberCode-to-CyberCode P2P model sharing backed by a configurable public WebRTC signaling endpoint.

Routing And Providers

• Adds draft autosave, preview, publish, rollback, undo/redo, drag-and-drop node creation, reconnectable edges, context menus, validation, and responsive graph editing.
• Lets distribution nodes add and remove output branches explicitly from both the node card and inspector; removed branches disconnect safely without deleting downstream model nodes.
• Prevents duplicate semantic connections and keeps manually configured distribution branch counts stable across saves while preserving legacy one-branch routes.
• Separates Kimi Code membership access from the Moonshot/Kimi API catalog and applies provider-specific thinking settings without leaking one provider's model into another.
• Probes and caches image capability dynamically, normalizes provider-specific model metadata, and keeps unsupported-image requests recoverable so MCP or local vision tools can still be invoked.
• Improves local-model warmup, timeout recovery, context-window handling, retry behavior, and model-switch isolation.

Agent Node And P2P

• Supports multiple named access keys with independent permissions, default targets, quotas, rotation, draft protection, and copy actions.
• Uses stable, editable public aliases so third-party agents no longer need long internal provider identifiers for ordinary setup.
• Separates route and model selection in the connection generator and defaults client-facing model configuration to auto.
• Adds compact in-app Anthropic/OpenAI-compatible examples plus detailed Chinese, English, Japanese, and Korean guides.
• Adds encrypted P2P signaling and relay-channel foundations so two CyberCode installations can share approved models without exposing upstream credentials.

Conversations, Projects, And Evolution

• Improves queued-message behavior during active generation with independent editing and automatic ordered delivery after the current response.
• Reworks project-grouped conversations, project filtering, temporary sessions, folder selection, creation flows, and sidebar performance.
• Adds animated per-turn and context token usage with clearer units and a more readable detail panel.
• Makes skill evolution visible through automatic learning, preference and working-style extraction, proposal review controls, and a default automatic mode.
• Improves image attachment previews, restored history images, long-session loading, Markdown rendering, diff rendering, and large transcript/index processing.

Marketplaces, Migration, And Computer Use

• Adds server-backed Skill and Plugin marketplace catalogs, localized metadata, official plugin icons, filtering, sorting, packaging checks, and resilient loading states.
• Extends USB and Agent migration with clearer progress, portable manifests, safer path handling, download fallback, and compatibility fixes.
• Improves Computer Use setup, bundled dependency validation, permissions guidance, runtime restoration, and cross-platform helper packaging.
• Fixes stale-cache crashes in Agent Node, USB migration API compatibility, routing chart consistency, and several compact-window interaction regressions.

Distribution

• Publishes signed and notarized macOS ARM64 and Intel builds, Windows x64, and Linux x64.
• Provides a verified standard ZIP download for every platform alongside DMG, DEB, AppImage, NSIS, updater manifests, portable assets, checksums, and signatures.
• Validates embedded sidecars, Code Graph grammars, Agent Browser, RTK, Computer Use resources, portable archives, and P2P signaling before a Release can be published.

Verification

• Passed the complete root service and desktop Vitest suites, routing compatibility regressions, TypeScript and production builds, Rust checks, sidecar validation, and multilingual documentation builds.
• Verified new distribution branch editing, legacy graph compatibility, Agent Node multi-key flows, provider capability routing, marketplaces, migration, conversation queues, and packaged-resource contracts.

CyberCode v1.1.16

CyberCode v1.1.16

Highlights

• Makes repeated session switching substantially faster and prevents the chat view from remaining stuck on history loading after several switches.
• Loads recent conversations through bounded, reusable requests and a short-lived background cache while keeping the active conversation responsive.
• Hardens Computer Use so desktop approval requests authenticate correctly and every release platform must prove the bundled runtime works without network access.
• Improves cross-platform session and project handling, including Windows path display and project-scoped tabs when session identifiers overlap.

Session Reliability And Performance

• Starts history reads before opening another long-lived chat socket, suspends idle connections, and ignores stale socket callbacks from replaced connections.
• Deduplicates concurrent foreground history reads, bounds background prefetch waits, extends the useful cache lifetime, and applies request timeouts instead of leaving the interface loading indefinitely.
• Keeps the current and previous chat panels warm for quick switching while restoring the correct bottom position only for the visible panel.
• Prevents stale history responses from one project from overwriting a same-ID session opened from another project.
• Builds recent-project choices immediately from locally available sessions, refreshes Git metadata with bounded concurrency, and caches results without hiding newly created or removed projects.

Computer Use

• Passes the desktop server authentication token into agent sessions and attaches it to Computer Use approval requests.
• Adds regression coverage for the authenticated approval bridge and desktop child-process environment.
• Extends the sidecar release smoke test to reject unauthenticated runtime requests, restore the embedded runtime in a fresh temporary profile, and verify Python and dependencies while all download URLs are deliberately unreachable.
• Runs that offline execution test independently on macOS ARM64, macOS Intel, Windows x64, and Linux x64 before packages can be published.

Desktop Polish

• Makes token context usage rely on the latest real context snapshot instead of cumulative tool-turn usage, avoiding misleading 100% readings during long agentic turns.
• Improves cached-chat scrolling, compact message spacing, Markdown rendering stability, tab targeting, and session/project navigation.
• Handles both slash and backslash path separators in project names and compact path labels.

Distribution

• Updates GitHub-hosted release, runtime, installer, and Playwright workflows to Node 24-compatible official Actions.
• Continues to publish signed and notarized macOS ARM64 and Intel builds, Windows x64, and Linux x64, with a standard ZIP download for every platform.

Verification

• Passed 1,431 root service and adapter tests, 745 desktop Vitest tests, and 6 Playwright desktop/compact end-to-end tests.
• Passed desktop TypeScript and production builds, Rust checks, documentation builds, Action workflow linting, and whitespace validation.
• Passed two complete four-platform cloud build rehearsals, including packaged-resource extraction, ZIP integrity checks, macOS signing/notarization, and real offline Computer Use runtime startup on every platform.

CyberCode v1.1.15

CyberCode v1.1.15

Highlights

• Reworks the model and routing workspace for compact desktop windows, with stable navigation, responsive provider grids, and dialogs that remain usable at low resolutions.
• Separates custom providers from local runtimes and places custom connections first, while preserving existing provider and model settings.
• Adds a unified Data Migration workspace for Agent imports and USB-based CyberCode portability across macOS ARM64, macOS Intel, Windows x64, and Linux x64.
• Makes USB migration substantially faster by excluding rebuildable caches and generated dependencies, copying files concurrently, and selecting only the current platform app by default.
• Ships the verified private Computer Use runtime inside every desktop package without requiring users to install Python or prepare dependencies.

Providers And Routing

• Fixes model-level branding so DeepSeek and other child models use their own logo instead of inheriting an unrelated provider logo.
• Standardizes model, API format, Web Cookie, and media-provider selectors with the same keyboard-accessible dropdown component.
• Adds localized, provider-specific Web Cookie credential instructions, exact field names, safe clipboard import, and clearer security guidance.
• Keeps provider filtering useful with an empty search query and improves compact layouts for intelligent routing, Agent Node, status metrics, and route actions.
• Keeps custom, official API, aggregator, OAuth, Web Cookie, media, and local catalogs distinct and consistently ordered.

Data Migration And Portability

• Combines Agent migration and USB portability in one keyboard-accessible page while retaining compatibility with the previous migration route.
• Copies settings, sessions, skills, plugins, explicitly selected projects, portable path metadata, launchers, and optional application packages into an atomic portable bundle.
• No longer selects every discovered project or all four application packages automatically; projects are opt-in and the current platform is selected by default.
• Excludes rebuildable CyberCode runtime, index, cache, log, and temporary data, plus generated project directories such as node_modules, target, dist, virtual environments, and Code Graph indexes.
• Uses bounded concurrent file copying, live current-item updates, an explicit old-bundle cleanup stage, and a 30-second no-data download timeout with automatic mirror failover.
• Maps project references across drive letters and mount points so the same portable data can reopen projects on Windows, macOS, and Linux.
• Adds data-only fallback when portable application assets are temporarily unavailable and prevents filesystem roots, user profiles, and configuration-owning directories from being mistaken for projects.
• Preserves checksums, first-run profiles, updates to existing portable directories, Windows argument forwarding, atomic replacement, and rollback behavior.

Computer Use And Distribution

• Packages the Computer Use runtime as a verified opaque payload so macOS notarization and Linux AppImage dependency scanners do not recursively inspect its private Python libraries.
• Packages the Linux Bun sidecar behind a checksum-verified launcher, preventing AppImage tooling from rewriting or rejecting the embedded executable while preserving transparent first-run startup.
• Restores the original runtime archive only on first use, then validates payload and archive checksums, extraction paths, Python, and required imports before atomic activation.
• Keeps managed network download and mirror fallback available when a bundled payload is unavailable or invalid.
• Fixes Windows runtime archive creation under Git Bash by converting runner paths before invoking tar.
• Adds Linux AppImage and Windows portable ZIP assets, plus a checksummed portable.json manifest used by USB migration.
• Publishes a standard ZIP download for every supported desktop platform alongside installers and updater manifests.

Reliability

• Prevents stale generated Computer Use resources from leaking into local or release builds when embedding is disabled.
• Preserves portable Code Graph, scheduled-task, and session project paths without changing unrelated local paths.
• Improves migration cancellation, progress accuracy, mirror fallback, compact-window overflow handling, startup resource validation, and token-usage control styling.
• Adds multilingual Data Migration documentation and updates provider-routing guides to match the final catalog structure.

Verification

• Passed the complete scoped root service, adapter, routing, migration, conversation, and utility test suites.
• Passed the complete desktop Vitest suite, TypeScript checks, production builds, sidecar preparation, Rust checks, multilingual documentation builds, and GitHub Actions validation.
• Verified real USB migration data locally: filtered configuration scanning completed in about 0.16 seconds and copied about 83 MiB in about 0.31 seconds, excluding runtime and index caches.
• Verified a real bundled runtime payload through offline restoration and Python dependency validation with zero network requests.
• Verified Linux DEB and AppImage extraction, checksum restoration, and real sidecar startup in a clean GitHub-hosted Linux build.
• Validated checksums, release asset rules, portable archive generation, whitespace, and packaged resource checks.

CyberCode Computer Use Runtime v1

Platform-specific private runtimes downloaded automatically by CyberCode. Users do not need to install Python.

CyberCode v1.1.10

CyberCode v1.1.10

Highlights

• Rebuilds the model-provider experience around clear first-level catalogs for official API keys, model aggregators, OAuth accounts, Web Cookie sessions, local runtimes, and image, audio, and video services.
• Adds provider model discovery, manual import, and supported-provider background synchronization without overwriting existing user configurations.
• Adds health-aware intelligent routing with ordered fallbacks, availability checks, and route-aware model selection across desktop and TUI workflows.
• Adds Agent Node endpoints compatible with OpenAI and Anthropic clients, including multiple access keys, model scopes, monthly quotas, rotation, revocation, and isolated usage accounting.

Agent Node

• Migrates or discards legacy single-key and hot-reload cache entries safely, preventing the Node page from crashing or asking users to restart CyberCode.
• Adds stable, readable, editable provider aliases, migrates opaque legacy aliases, and invalidates stale caches that could expose internal provider UUIDs.
• Protects unsaved key policies during key switching, renaming, and rotation, and adds confirmation before rotating a key.
• Simplifies the in-app connection guide to Protocol, Base URL, API Key, and Model=auto; advanced fixed targets and test requests remain collapsed until needed.
• Separates routes and direct models in the connection generator, keeps auto as the simple default, and removes internal provider-ID terminology from the primary workflow.

Desktop Experience

• Speeds up provider and Agent Node pages with read-through caching, progressive loading, and locally available catalogs instead of blocking the whole drawer on remote requests.
• Adds consistent catalog search and access, cost, and modality filters, including correct filtering when the search box is empty.
• Keeps Kimi Code and Kimi as separate API-key entries while excluding OAuth runtime records from API-key and local-provider catalogs.
• Restores the right-side workspace tools, adds a native Git workspace, and improves compact-window layout, modal behavior, dropdowns, external links, and startup readiness.

Intelligence And Portability

• Adds first-class Web Cookie and media-provider catalogs with localized provider names, visible risk guidance, and provider-specific credential flows.
• Adds TUI commands for /provider, /routing, and /node, using the same built-in provider bridge and routing runtime as the desktop app.
• Extends agent migration to WorkBuddy, Kimi Code, and Pi while preserving native skill, memory, instruction, and project formats.
• Improves Code Graph lifecycle handling, RTK integration, attachment fallback, temporary-session memory, and long-running task recovery.

Reliability

• Stops active agent-browser work when generation is cancelled and prevents late tool output from reviving a completed turn or leaving context usage stuck at full.
• Sends desktop-server authentication to IM adapter project, session, Git, and task-list requests so commands such as /projects and /new work after connecting.
• Prevents invalidated or manually refreshed read-through requests from overwriting newer cached data.
• Tests configured Web Cookie providers concurrently with bounded parallelism, reducing long waits without overwhelming providers.
• Retries transient Apple timestamp-service failures during signed macOS builds while still failing immediately for non-transient signing errors.
• Uses a Windows baseline Bun sidecar target to avoid unsupported CPU instruction crashes and validates every embedded runtime before publishing.

Documentation And Distribution

• Adds complete Chinese, English, Japanese, and Korean guides for providers, model synchronization, intelligent routing, Agent Node access, and matching TUI commands.
• Opens the Agent Node guide in the current interface language and keeps the primary examples free of internal IDs and real credentials.
• Cloud builds cover macOS Apple Silicon, macOS Intel, Linux x64, and Windows x64, with a ZIP download for every platform alongside the normal installer and updater assets.

Verification

• Passed Agent Node desktop UI, cache normalization, gateway, TUI routing, provider-service, IM adapter, conversation cancellation, and read-through-cache regression suites.
• Passed desktop TypeScript checks, VitePress multilingual documentation build, real-page Agent Node verification, packaged-resource validation rules, and Git whitespace checks.