The official desktop application and central distribution gateway for the BackEnd software ecosystem.
A local-first developer command center that unifies coding agents (Claude Code, Antigravity CLI), IDEs (Antigravity, Cursor, VS Code), test suites, Git repositories, and verified ecosystem application management under strict human-in-the-loop cryptographic approval.
Every view below is an authentic high-resolution capture of the running Hub desktop client, operating against authentic SQLite database records and local repository state.
The operational central view answering "What is running?", "What needs approval?", and "What failed?". Highlights project health, pending approvals, active workspace locks, and discovered coding tools.
Target individual repositories on disk. Hub tracks branch state, commit health, and assigns exclusive filesystem locks to prevent concurrent agents from colliding in the same folder.
Honest reporting across 15 distinct capability flags. Real executable version discovery for Claude Code, Antigravity, Cursor, VS Code, and Ollama, separating headless execution from interactive-only tools.
Inspect the exact command, working directory, and prompt before execution. Cryptographic SHA-256 payload snapshot guarantees that unapproved changes cannot slip in between review and execution.
Chain complex developer workflows across all 7 primitives: run missions, trigger test suites, launch IDEs, pause for human code review, execute verification checks, and package context handoffs.
Non-destructive repository observer. Tracks modified files, untracked artifacts, current commit hashes, and packages structured diff context to hand off between consecutive agent runs.
An immutable, timestamped record of every action taken in your environment. Records execution duration, process exit codes, operator approvals, and error details stored permanently in local SQLite.
Manage local updates, check releases against sinvaux.com, view database schema health, inspect paired BackEnd tools (Scaffold and Aftermath), and configure process timeouts.
Today's developer tools force a false choice: either let autonomous agents execute arbitrary shell commands blind, or manually copy-paste terminal outputs across five disconnected windows. Hub provides a third way: automated orchestration bounded by cryptographic human-in-the-loop gates.
Before any consequential mission or test runs, Hub captures a SHA-256 snapshot of the exact executable path, CLI arguments, and working directory. If anything is modified between review and execution, the step aborts instantly.
No two workflows or background missions can execute simultaneously on the same repository directory. Hub halts overlapping commands with structured WORKSPACE_BUSY rejections.
All binaries are invoked directly through OS-level Process spawning with validated arguments and explicit PATH lookup. Hub never passes raw strings to cmd.exe or powershell.exe.
Workflows in Hub are not opaque black boxes. Every multi-agent sequence is composed of seven transparent, auditable step primitives.
| Primitive | Target Category | Approval Policy | Execution Behavior |
|---|---|---|---|
RUN_MISSION |
Agents & CLI Tools | Requires Approval | Direct process execution (shell: false), capturing live circular stdout/stderr buffers with cancellation hooks. |
RUN_TESTS |
Test Runners | Requires Approval | Auto-detects Vitest, Jest, PyTest, or Cargo suites; enforces timeout limits and captures exit codes. |
OPEN_WORKSPACE |
IDEs (Antigravity, Cursor, VS Code) | Requires Approval | Launches native desktop IDE at the project root directory via detached, non-blocking platform commands. |
OPEN_FILE |
IDEs & Editors | Optional | Opens a specific target file directly in the user's preferred IDE at the precise line of interest. |
REVIEW_PAUSE |
Human Operator | Mandatory Checkpoint | Halts execution until the operator inspects changes, checks diffs, and explicitly signals continuation. |
VERIFY |
Verification Harnesses | Automated | Runs non-destructive verification scripts, measuring duration and recording pass/fail scorecards in SQLite. |
HANDOFF |
Multi-Agent Context | Automated | Packages git branch status, dirty file lists, and previous step outputs into structured context for the next agent. |
Hub discovers installed development tools on your Windows system dynamically, querying exact binary versions and tracking fifteen distinct capability flags.
Discovers claude.exe via user path. Executes headless non-interactive prompts with -p, streaming output buffers, structured exit codes, and instant cancellation.
Discovers Google Antigravity installations. Safely opens workspaces and individual files. Truthfully reports interactive-only status rather than fabricating prompt execution.
Discovers cursor.cmd and code.cmd on Windows PATH. Seamlessly coordinates editor window opening during handoff and review stages of workflows.
Discovers ollama.exe on the local machine. Coordinates offline language model prompts without cloud dependencies or external API usage.
Standalone desktop application for Windows 10/11 x64. Includes complete SQLite database engine, local process manager, Git observer, multi-agent workflow runner, and ecosystem distribution engine.