Explicit AI — Agile Development Assistant

A VS Code extension for AI-driven development with explicit context control. No auto-indexing, no hidden magic — you control exactly what the AI sees. Works with any OpenAI-compatible API: LM Studio, Ollama, OpenAI, Groq, Together, and more.

No telemetry. No data collection. Your code stays local.

Features

Three Workflow Modes

Switch between modes with Tab (or click the badge in the textarea):

Inline Completions

Ghost text autocomplete as you type — powered by your local model. Tab to accept. Configurable via explicitAI.inlineCompletions.

Multi-Modal (Image Input)

Paste screenshots or images into chat to ask about them. Supported by vision-capable models. Configurable via explicitAI.allowImages.

Drag & Drop File Mentions

Drag files from the VS Code explorer into the chat input — automatically inserts #relative/path at cursor and attaches the file for context.

Skills System

Load best-practice skill instructions into the AI's context with @skill <name>:

@skill react        # Load React best practices
@skill testing      # Load testing best practices
@skills             # List all available skills

Skills are stored as markdown in .explicitai/skills/. Bundled skills included: react, testing. Create your own — just add .md files.

@ and # Syntax

• @agentName message — invoke a specific agent (autocomplete on @)
• #relative/path — attach a file (autocomplete on #)
• @skill name — load a skill
• @skills — list available skills
• @codebase — semantic search across codebase
• @terminal — include recent terminal errors
• @workspace — include project file structure
• @errors — include diagnostics
• @git — include uncommitted changes
• @selection — include current editor selection

Built-in Agents

Auto-Compaction at 60%

The conversation automatically compacts when token utilization reaches 60% — older messages are summarized to free context space and prevent hallucination. No manual intervention needed. A visual [auto] indicator appears in the chat when it triggers.

Self-Healing

After every write/edit, the extension auto-fixes lint and compile errors using VS Code code actions. Configurable via explicitAI.selfHealing.

Agile Planning

Full sprint-based agile workflow built into the chat:

• Plans & Tasks — Create plans with goals, break into tasks with priorities, story points, acceptance criteria, and dependencies
• Sprints — Time-boxed sprints, task assignment, velocity tracking
• Retrospectives — What went well, what to improve, action items
• Auto-Plan — Describe a requirement and the AI suggests a task breakdown

Project Memory

Persistent memory across sessions:

• Decisions, preferences, patterns, warnings, knowledge
• Keyword-based retrieval into AI prompts
• Manual remember/forget from the chat UI

Built-in Developer Tools

Read tools execute immediately. Write/shell tools require approval (except in Agent mode with undo support).

MCP Integration

Model Context Protocol support with manual approval for every action:

• Configure servers in .explicitai/mcp.json
• Toggle individual tools (filesystem, terminal, HTTP)
• Enable/disable servers from the UI

LSP Tools

The AI can query language intelligence directly:

• getDiagnostics — errors and warnings for a file
• getDefinition — go to definition
• getReferences — find all references
• getHoverInfo — type info at position
• getDocumentSymbols — outline of a file
• getCodeActions — available quick-fixes
• getImplementations — find implementations
• getTypeDefinition — go to type definition

Thread History

• Conversations auto-save as threads
• Switch between threads or revert to a previous state
• Threads track mode (chat/plan/agent) and active agent

Commands

Context Management

• Token budget tracking with visual utilization bar
• Automatic compaction at 60% utilization
• Manual compact button
• Stack detection auto-injects project info

Setup

1. Install LM Studio (or any OpenAI-compatible provider) and load a model
2. Start the local server (default: http://localhost:1234)
3. Install the extension from VS Code marketplace
4. Open the Explicit AI sidebar from the activity bar

Configuration

All settings under explicitAI.* in VS Code settings:

Architecture

src/
├── agents/          # Custom AI agent registry
├── chat/            # Chat session management
├── commands/        # VS Code command implementations
├── completions/     # Inline completion provider
├── core/            # LLM client, config, memory, planner, skills, etc.
├── mcp/             # Model Context Protocol integration
├── threads/         # Conversation thread persistence
├── tools/           # Built-in workspace tools + LSP tools
├── ui/              # Webview host, panel, provider
└── extension.ts     # Entry point

Data stored in workspace:

.explicitai/
├── agents/          # Agent JSON configs
├── threads/         # Conversation snapshots
├── skills/          # Markdown skill files
├── plans.json       # Plans, tasks, sprints, retros
└── memory.json      # Persistent project memory

Development

npm install
npm run build        # Compile TypeScript + validate HTML
npm run watch        # Watch mode for development

Press F5 in VS Code to launch the extension in a development host.

License

Local use only.