Yapzek Local Agent

Yapzek Local Agent is a VS Code chat extension that lets you talk to local LLMs (Ollama or LM Studio) through a dedicated Yapzek chat surface. It also ships a control-panel sidebar, helper commands, and a light abstraction layer for adding more “agent tools” later on.

Highlights

• Yapzek Chat sidebar & window – open the custom Yapzek Chat view from View → Yapzek Chat or launch the detachable panel via Yapzek Agent: Open Chat Window whenever you prefer a standalone surface.
• Streaming replies – Ollama and LM Studio responses stream directly into the Chat panel for faster feedback, with buffered fallback automatically applied on failure.
• Local model bridge – single client interface that speaks both Ollama (/api/chat) and LM Studio (OpenAI compatible v1/chat/completions).
• Sidebar console – monitor active settings, jump into Chat, or open the configuration view from a dedicated activity-bar container.
• Command helpers – palette commands that capture the current selection, prime your clipboard, and focus the Yapzek Chat view so you can paste the prepared slash-command prompt.
• Toolbox commands – insert snippets, scan the workspace for text, or run shell commands in a dedicated Yapzek terminal without leaving the editor.
• Structured tooling – when the model emits a JSON "tool plan", Yapzek will create or edit files and run terminal commands automatically on your behalf.
• Mirrored command output – every automated action streams into both the chat transcript and a dedicated Yapzek Agent output channel, so you can review stdout in real time.
• Agent mode parity – pick "Agent" in the dropdown to mirror VS Code Chat’s agent role and aggressively prefer actionable tool plans.

Screenshots

Requirements

• VS Code 1.106.0 or newer (Chat Extensions API).
• Node.js 18+ (for building/running the extension).
• At least one local model server:
  • Ollama: ollama serve and ollama pull llama3 (or any supported model).
  • LM Studio: enable the REST server (Settings → Developer → REST API), note the port and optional API key.

Configuration

Open the command palette → Preferences: Open Settings (UI) and search for “Yapzek Local Agent”.

Usage

1. Start your preferred local model backend and confirm it answers REST calls.
2. Open the Yapzek Chat sidebar via View → Yapzek Chat or run Yapzek Agent: Open Chat Sidebar / Yapzek Agent: Open Chat Window from the Command Palette for a floating panel.
3. Use the helper commands from the palette to prefill prompts:
   • Yapzek Agent: Explain Selection
   • Yapzek Agent: Refactor Selection
   • Yapzek Agent: Generate Tests
   • Yapzek Agent: Insert Snippet (paste model output directly at the cursor)
   • Yapzek Agent: Workspace Search (jump to the next file containing a literal string)
   • Yapzek Agent: Run Terminal Command (send text to a dedicated terminal session)

Each helper copies the prepared text to your clipboard and focuses the Yapzek Chat view so you can paste it right away.

Custom Yapzek chat view

The dedicated sidebar/panel replicates the core VS Code Chat experience with a few extras:

• Dropdown mode selector (Ask, Explain, Refactor, Test) that maps to Yapzek’s slash commands.
• Toggle to include the current selection or file context with every turn.
• Streaming transcript with cancel/clear controls plus quick access to the Yapzek settings page.

Everything flows through this standalone surface—no @-mention or native Chat integration is required.

Tool automation

Yapzek now supports structured "tool plans" so the model can launch commands or edit files safely:

```json
{
	"explanation": "Add unit tests and verify formatting",
	"actions": [
		{ "tool": "writeFile", "id": "edit-tests", "path": "src/foo.test.ts", "contents": "// updated tests" },
		{ "tool": "runTerminal", "id": "run-tests", "command": "npm run test" }
	]
}
```

• Wrap the plan in a single ```json block.
• Paths are always workspace-relative.
• Use createFile for brand new files; use writeFile for edits (set append or overwrite flags as needed).
• runTerminal commands should be non-interactive (npm/yarn/pnpm scripts work best).
• Omit the plan entirely if you’re only providing an explanation and no automated steps are required.
• Yapzek injects these TOOL PLAN instructions into both the system and user prompts so compliant models know exactly how to respond. Non-compliant replies simply act as analysis with no actions executed.
• Automated actions now ship with live stdout/stderr mirroring in the Yapzek Agent output channel plus a compact summary message in the transcript so you always see what ran, what succeeded, and what needs attention.

Developing & Testing

npm install
npm run watch

• Press F5 (or run the “Launch Extension” configuration) to start a new Extension Development Host.
• Inside the test window, open the Yapzek Agent view from the activity bar and start chatting with the participant.

Run a one-off type-check/build (plus lint) before publishing:

npm run compile
npm run lint

Packaging & Distribution

The project bundles vsce as a dev dependency so you can produce VSIX artifacts locally:

npm run package

This command creates yapzek-local-agent-x.y.z.vsix in the workspace root. Share it directly or install it via the VS Code command palette (Extensions: Install from VSIX...).

Roadmap / Known Gaps

• Add markdown rendering (and copy buttons) to the transcript to improve long-form replies.
• Expand the toolbox to cover diff-aware edits, rename/copy operations, and smarter workspace search.
• Wire up packaged release automation (CI workflow plus signed VSIX artifacts).

Contributions and ideas are welcome—open an issue or start a discussion once the repo is published."# Visual-Studio-Code-Custom-Chat-Extension-LM-Studio-Ollama-"