Build and operate
in a single window

All your servers on one screen. Sessions auto-recover when dropped, and Korean output never breaks.

• ✓ Zero remote install — run Claude Code and Codex locally, only sshd on the server (works on air-gapped or no-install-permission servers too)
• ✓ Split multiple SSH sessions + auto-reconnect on drops
• ✓ No garbled Korean output (-EncodedCommand handling)
• ✓ Saved connection profiles, SFTP file CRUD

Monaco engine with LSP for 14 languages. Ctrl+I inline AI fixes code with RAG context and lets you review via diff.

• ✓ Monaco + LSP for 14 languages
• ✓ Inline AI (Ctrl+I) + RAG context injection
• ✓ diff Apply/Reject, image and PDF viewer, SFTP remote editing

Control remote desktops over VNC and run commands on Windows servers via WinRM. Pull other apps screens into a panel with desktop mirror and diagnose them with AI Vision.

• ✓ VNC viewer — screen control + file upload/download + capture
• ✓ WinRM — remote PowerShell execution on Windows servers
• ✓ Desktop mirror — bring another app screen into a panel, AI Vision diagnostics

Claude and Codex multi-turn streaming. An agent that operates panels, DB, SSH, and files directly through the MCP bridge.

• ✓ Claude + Codex, multi-turn streaming
• ✓ MCP bridge (40~60 tools for panel/db/ssh/file/preview and more)
• ✓ Agent mode to directly manipulate the workspace
• ✓ Background Tasks — up to 5 concurrent AI jobs independent of chat

When one AI isn't enough. Spin up multiple agents on a board, split the work as a team, and stay in control with a run queue and unified approvals.

• ✓ Multi-session board — see multiple agents (Claude, Codex) as a node/edge graph with real-time flow visualization
• ✓ Team mode — auto-relay and split work via agent_send, create role-based teammates via agent_create, peer-aware prompts
• ✓ Run queue — own-turn, same-folder serialization, provider-budget gates + a file-conflict guard keep parallel work safe
• ✓ Unified approval inbox — approve every agent's requests in one place, with per-agent SSH/DB environment isolation
• ✓ Board serialization — board, conversations, and progress are restored across restarts

Sync designs from claude.ai/design and Figma and render them live with no external server. Inject tokens and components into AI code generation and edit them on the spot by just asking.

• ✓ claude.ai/design & Codex Figma integration — pull components/screens and render live in an iframe (zero external server)
• ✓ Multi-platform preview for Web, iOS, Android and Flutter (Dart) in device frames
• ✓ AI design grounding — inject tokens/components into code generation for consistent UI code
• ✓ Token export (Flutter ThemeData, CSS variables) + visual regression checks

5 engines in one place. Turn natural language into SQL and roll back risky queries using backup tables.

• ✓ Unified connections across 5 engines
• ✓ Natural language → SQL, large-scale virtual scroll grid
• ✓ Transaction control + backup-table-based rollback

Version control even on PCs without git, thanks to a bundled MinGit. Track and revert changes and push straight to GitHub.

• ✓ MinGit bundle — works even on PCs without git installed
• ✓ ChangeSet change tracking and revert
• ✓ Work history logging + git checkpoint restore
• ✓ GitHub OAuth publish / clone

Search 6 sources — code, DB, Git, chat, history, docs — by meaning. Embeddings run locally and never leave your machine.

• ✓ sqlite-vec-based hybrid search across 6 sources
• ✓ Semantic + keyword combined + reranking
• ✓ Local embeddings — your code never leaves the machine

Trace runtime errors back to editor markers. Embed your dev server as a preview and verify it with AI vision.

• ✓ Trace run-result errors back to editor markers
• ✓ Embedded dev server Live Preview
• ✓ AI vision verification — screenshots, console, visual regression
• ✓ Per-chat Python virtual environment — run code, auto-install packages

Drop a single probe line (@BLYCK@) and wherever it fires — terminal, browser, or mobile — it collects into one buffer. The AI reads runtime values directly with debug_read, so it fixes without guessing and without anyone pasting the console.

• ✓ Standard debug line @BLYCK@ {label,vals,level,loc} — one format for every platform
• ✓ AI observes runtime values directly via debug_read (zero need for humans to copy-paste the console)
• ✓ 11-language helpers (debug_shim) + format-guaranteed probes (debug_probe) — auto no-op in release builds
• ✓ AI auto-investigates on errors and asserts (Mode B, OFF by default, 4-layer safety)
• ✓ Ctrl+Shift+D live debug panel — humans see the same stream the AI sees, in real time

The AI stores the capabilities, rules, and facts it verified in this workbench into exact-match memory and pulls them back automatically next session. Working alongside RAG fuzzy search, it stops false "I can't" calls.

• ✓ Store verified capabilities, rules, and facts in exact-match memory (a counterpart to RAG fuzzy search)
• ✓ Successful actions (deploy, build, DB connect) auto-promoted to capability candidates (approval gate)
• ✓ Relevant memories auto-injected into the next session — prevents false "I can't do that" calls
• ✓ Stale memories auto-pruned (decay); capabilities and rules are preserved

Export your chat history, settings, and connections in one click; unpack on another PC and the paths convert automatically. Encrypted bundles carry even DB/SSH passwords safely.

• ✓ Export/import chat history, settings, and connections as a .blyckbundle
• ✓ Automatic path remapping — Windows↔macOS, 4 directions, chat history carries over
• ✓ AES-256-GCM encrypted bundle — safely includes DB/SSH passwords and GitHub token
• ✓ Merge mode — add to an existing install without replacing, auto-backup before import

Auto-updates, license management, and localization in 7 languages. A native desktop app for Windows and macOS.

• ✓ Auto-update (update.blyck.dev)
• ✓ License management (7-day grace)
• ✓ i18n in 7 languages, Windows and macOS support