A Leader plans, a DAG orders the work, expert agents execute, and evidence closes the loop. LingXiao is not a model wrapped in a text box; it is a local engineering runtime for models, tools, permissions, tasks, state, and verification.
LingXiao is built for long-running, complex, high-stakes AI engineering work. From one goal, it can form a plan, create dependency-aware tasks, dispatch role-based worker experts, call real tools, and preserve build output, test results, browser screenshots, logs, Git changes, and session state as auditable evidence.
Expert Team Runtime
The Leader plans, dispatches, and verifies; workers operate as research, frontend, backend, QA, review, design, or operations specialists.
Task DAG Orchestration
Complex goals become dependency-aware task graphs, making parallel work, blockers, failures, and repairs visible.
Real Tool Kernel
Files, Shell, Git, browser automation, HTTP, Office, MCP, skills, and workflows are unified under one tool registry.
Full State Sync
WebUI / TUI / CLI share the same runtime state: task board, agents, permission requests, terminal output, and artifacts.
Evidence Loop
Delivery returns to builds, tests, screenshots, logs, contracts, and diffs. Failed gates become explicit repair work.
Recoverable Sessions
SQLite, Git, memory, work notes, and archived artifacts preserve engineering context across interruptions.
Entry
WebUI / TUI / CLI / Desktop
Unified goal entry with live visibility into tasks, agents, tools, permissions, and evidence.
Session Kernel
SessionManager / SessionRuntime
Creates, restores, pauses, and synchronizes sessions while maintaining modes and event streams.
Command
LeaderAgent
Understands goals, creates plans, builds the DAG, selects roles, and verifies evidence.
Orchestration
TaskBoard DAG
Tracks dependencies, blockers, dispatch, completion, failure, repair, and evaluation state.
Execution
AgentPool / Worker Experts
Specialists for research, coding, testing, review, design, documentation, and operations.
Evidence
SQLite / Git / Memory / Artifacts
Stores session state, diffs, work notes, artifacts, verification evidence, and recovery trails.
| Scenario | How LingXiao Handles It |
|---|
| Product Delivery | Coordinates requirements, contracts, frontend/backend work, browser verification, and docs. |
| Code Governance | Tracks technical debt, migrations, test gaps, refactors, and proof-backed repair plans. |
| Internal Automation | Turns local scripts, Office/PDF tasks, MCP, Git, workflows, and models into reusable processes. |
| Long-Running Goals | Eternal Goal keeps pursuing a target until complete, blocked, or ready for a human decision. |
| Layer | Technology |
|---|
| Runtime | Node.js 24, TypeScript (ESM) |
| Web Server | Fastify, @fastify/websocket |
| Web UI | React, Zustand, Vite |
| TUI | Ink |
| Desktop | Electron, electron-builder |
| Database | SQLite (WAL + FTS5) |
| CLI | Commander.js, Inquirer, Chalk |
| LLM | OpenAI SDK, Anthropic SDK, Vercel AI SDK |
| Dependency | Requirement |
|---|
| Node.js | >=24.0.0 |
| npm | Bundled with Node 24 |
| Git | Recommended for diffs, commits, and recoverable trails |
| OS | Linux / macOS / Windows / WSL |
