xMind is a native agent platform and coding assistant for local projects. The CLI, webpage and VS Code extension share one C++ backend for agents, tools, models, sessions, approvals and execution history. Embedded xlang3 runs compatible Python scripts and pure-Python libraries and performs SQLite I/O.
A coding agent uses the same runtime as a general agent. Start with the normal agent/tool loop; add skills, named agents or graphs when the task needs them. Agents and graphs can be authored with YAML/Markdown and trusted custom xlang3 programs through the native authoring API. See the architecture, diagram and documentation index.
The local development installation is 0.1.43, including the coding-diagnostics
fix. This is a development preview, without a
Marketplace or general-release claim. The xlang3 SDK is pinned at
58a8aab9b40a464e38a741129b4804ac36e27a53.
| Check | Verified scope |
|---|---|
| Installed native runtime | All 154 local contracts pass in 994.99 seconds with checked inputs and binaries unchanged. A separate real TypeScript check returns 0/0/1/0 diagnostics for clean/changed-clean/invalid/repaired buffers. |
| Installed clients | 0.1.43 passes 429 extension and 147 browser fixtures locally and on exact-source hosted CI; package and installed payloads verify. |
| Installed coding | Actual configured-Claude CLI and browser checks pass: one reviewed edit, fresh clean diagnostics and compiler exit 0. The preview refreshes automatically; reload retains 17 messages/eight usage records. |
| Earlier application lifecycle | Separate installed 0.1.37/0.1.38 tasks verify a Vue edit, 32 xlang3 tests, production build, FastAPI/Vite startup, and explicit service shutdown. |
| Preserved work | After installation, 27 checked older owners, 11 runtime inventories, 10 webpages, settings and prior profile directories remain intact. |
| Still open | Actual VS Code pane activation, Rust semantic startup, hosted native acceptance, broader protocol recovery and process-bound agent execution. |
The installed runtime refreshes invalidated push-only language-server analysis and returns an unavailable tool result when its transport fails, allowing other coding checks to continue. Missing diagnostics are never reported as clean. Workspace and provider metadata remain responsive while an editor query waits.
Current validation status, complete native result, installed 0.1.43 coding scope, earlier application lifecycle.
- Install the supplied verified Windows x64 VSIX with Extensions: Install from VSIX. VS Code 1.140 or newer is required.
- Open and trust the local project folder, then run xMind: Open Workspace.
xMind uses the opened folder; no
.code-workspacefile or separate server launch is required. For several opened roots, select the active root explicitly. - The agent sidebar is on the right. Choose the saved provider and a discovered model beside the composer. Settings can enroll a provider if none is saved. Existing configured keys remain on the backend and do not need re-entry.
- Send a coding task. Inspect file-change proposals and approve the exact edits. Configured trusted command profiles can run without a per-command prompt.
Use xMind: Open Browser View for the webpage attached to the same workspace and profile. The local adapter enrolls the browser privately; no server-token copy is needed. Closing a view leaves backend execution alive.
Machine-level runtime, provider/process YAML and profile-root settings are documented in the extension guide. Paths are not imported from the opened project. After changing trusted runtime/process configuration, use xMind: Apply Trusted Runtime Configuration for a profile-preserving handoff.
A verified native distribution contains xmind.exe and its paired runtime,
modules and pure standard-library source. Keep that directory intact. With
xmind.exe on PATH, start the managed console for a project:
xmind --workspace D:\Projects\Example chatThe default profile root is %USERPROFILE%\.xMind\p. If the editor uses a
custom agentflow.profileRoot, supply that same path with --profile-root to
share its backend and sessions. Reconnecting reuses the profile's saved provider
settings; omit first-launch configuration options unless you intend to change
the trusted launch configuration. No key belongs on the command line.
xmind --workspace D:\Projects\Example sessions
xmind --workspace D:\Projects\Example profile-info
xmind --workspace D:\Projects\Example workspace init
xmind --workspace D:\Projects\Example workspace add-folder D:\Projects\Shared
xmind --workspace D:\Projects\Example workspace infoworkspace init registers the first folder before group inspection or folder
attachment. Adding a folder attaches it to that workspace; it does not copy or
delete files. Existing runs retain their admitted folder snapshot.
Explicit --port is the separate operator-selected backend mode and requires
its authentication configuration. Console commands.
Use --mcp-config FILE to import trusted MCP server configuration into a
managed profile. Applying a changed document requires --upgrade-backend;
the backend validates saved revisions and retired server IDs before handoff.
VS Code exposes the machine setting agentflow.mcpConfigPath and
xMind: Apply Trusted Runtime Configuration. MCP setup.
The webpage uses the same renderer and native command/event contracts. Its top bar opens Workspace Settings for switching workspaces, creating one from an existing folder and attaching additional folders. The left pane browses files; the right sidebar contains the current conversation, activity, approvals and model selection. Browser launch and behavior.
For source builds use Windows x64, a C++20 Visual Studio toolchain, CMake, Node.js and the supported xlang3 SDK. The guarded build entry point is:
.\Tools\agentflow.ps1 -Action BuildDevelopment covers explicit SDK/library paths, isolated builds and native runtime packaging. Python commands, tests and pip run with xlang3; CPython executables and CPython native extensions are not used. Pure Python libraries from PyPI are allowed. xMind-required native gaps are fixed in an isolated xlang3 branch.
xMind OSS runs locally with SQLite, CLI, webpage and VS Code. Generic Local and Nexus connection profiles belong to the public architecture; authenticated Nexus enrollment and distributed worker execution remain unfinished. Private Nexus owns the team server, PostgreSQL, WebRTC and standalone Electron IDE.
OpenCode and LiteLLM are references for workflow and provider research. xMind implements its own C++ runtime and protocols; compatibility and feature parity with those projects are not acceptance criteria.
This repository preserves upstream xMind history and its Apache-2.0 license. The old xlang implementation and Python prototype are recoverable through Git history and are not part of the current product.