Tongyi Lingma

Type: full-code · Vendor: Alibaba Cloud (通义灵码 / Lingma / Qoder CN) · Language: TypeScript (extension), Java (server) · License: proprietary (free individual tier + enterprise editions) · Status: active · Status in practice: mature

Links: homepage docs

Alibaba's IDE-native coding assistant on Qwen models with an Agent mode that autonomously breaks tasks into to-dos, edits across multiple files, and reflects on intermediate steps.

Description. Tongyi Lingma (通义灵码) is Alibaba Cloud's IDE coding assistant, powered by Qwen. It ships VS Code and JetBrains extensions and offers two operating modes: an Ask mode for instant Q&A and code completion, and an Agent mode that takes an objective, decomposes it into to-dos, performs multi-file editing, and dynamically adjusts the plan as it executes. The product also ships as an enterprise edition (in-VPC deployment) and as a separately-branded research agent called LingmaAgent (described in the FSE 2025 industry paper) for autonomous issue resolution against a code repository.

Agent loop shape. Two-mode IDE plugin. Ask mode runs a single Qwen call per user turn with the current file and cursor context. Agent mode runs an iterative planner-executor loop: the planner emits a to-do list from the user objective, the executor performs file edits and tool calls (read/write/grep/run-tests) per to-do, and the planner reflects after each tool batch to update the to-do list. The Qoder/Lingma editor is the host UI; tool calls go through the IDE's filesystem and process APIs.

Primary use cases

line- and function-level code completion in VS Code and JetBrains
agent-mode autonomous multi-file edits scoped to a workspace
natural-language code generation, refactoring, unit-test generation
issue triage and resolution via LingmaAgent (FSE 2025) for enterprise repos

flowchart TD USER[Developer in IDE] --> EXT[Lingma extension VSCode / JetBrains] EXT --> MODE{Mode} MODE -->|Ask| QWEN[Qwen Coder model] MODE -->|Agent| PLAN[Planner emit to-do list] PLAN --> EXEC[Executor file read / write / grep / run-tests] EXEC --> TOOLS[(IDE filesystem + process API)] EXEC -->|each batch| REFLECT[Reflect on results] REFLECT -->|update to-dos| PLAN EXEC -.confirm edits.-> HITL[Human-in-the-loop diff review] QWEN --> EXT EXEC --> EXT

Key concepts

Ask mode vs Agent mode (docs) — Single-shot Q&A vs. autonomous multi-step task execution.
LingmaAgent (docs) — Research agent for autonomous issue resolution via comprehensive repository exploration (FSE 2025 industry paper).
To-do plan refresh → plan-and-execute — Agent mode dynamically adjusts the active to-do list as each tool batch completes.
Repository exploration → agentic-rag — LingmaAgent's distinguishing behaviour: comprehensive multi-file exploration before proposing changes (FSE 2025).