😼 ManulAI Local Agent

ManulAI is a local AI coding assistant for Visual Studio Code powered entirely by your own Ollama runtime.

It is built for developers who want workspace-aware chat and local tool execution inside the editor without cloud APIs, remote inference, or account-based workflow. ManulAI keeps the chat in the right-side Secondary Sidebar, forwards local file context to Ollama, and runs file and terminal actions through Ollama native tool-calling flow.

The Manul goes hunting and never returns without its prey.

Status: Alpha. Developed by a single person.

ManulAI is already useful for real work, but it is still being battle-tested on real-world projects. Bugs, rough edges, and behavioral changes are expected while the product matures. The priority is transparent local behavior, predictable tool execution, and strong Ollama-first integration rather than polished marketing promises.

Why Use ManulAI

ManulAI is designed for developers who already want Ollama as the model runtime and need a practical assistant inside VS Code:

• local-first by default
• no cloud AI dependency or remote model API
• chat and tools stay close to the code you are editing
• works across any programming language opened in VS Code
• keeps attached file context and conversation history available during the session
• supports both plain chat and agent-driven tool execution

What It Does

Right-Side Chat UI

• dedicated ManulAI chat view in the Secondary Sidebar
• launcher view in the Activity Bar
• chat stays beside the editor instead of replacing the main work area
• conversation history is kept in memory for follow-up requests

Local Ollama Integration

• uses your local Ollama server through /api/chat
• supports native Ollama tool-calling flow with tool role responses
• no hard request timeout inside the extension; long-running local responses can finish naturally
• model selection is exposed inside the extension UI and settings

Agent Mode And Chat Mode

ManulAI has two working modes:

• Agent Mode enables local tools and lets Ollama continue the tool loop automatically
• Chat Mode disables tools and responds as plain chat only
• in chat-only mode, ManulAI should not claim that files were changed
• Auto-Approve can be turned on to execute tools immediately or off to require confirmation for every tool call

File Context Flow

• attach the active editor file to the chat
• attach files from the Explorer context menu
• browse and attach files from disk
• keep attached file context visible in the UI
• forward attached content into the model context for the next requests

Built-In Workspace Tools

Agent Mode currently exposes these tools to Ollama:

• read_active_file
• read_specific_file
• create_or_edit_file
• write_to_file
• replace_in_file
• execute_terminal_command
• delete_file
• list_workspace_files

These cover the main local coding tasks: reading files, targeted edits, full rewrites when necessary, file creation, file deletion, listing workspace directories, and running local shell commands.

Safer Editing Behavior

The extension now pushes stricter file-editing rules into the agent prompt:

• make surgical edits for small requests instead of rewriting entire files
• never remove content that the user did not explicitly ask to remove
• prefer targeted replacement over full-file overwrite when possible
• read the file before editing so the model has the full structure
• do not claim a file changed unless a real tool changed it

This exists specifically to reduce destructive edits like replacing an entire README when the request was only to remove one line or one image block.

Direct Command Handling

For common requests, ManulAI also has direct handlers before the full agent loop:

• title rename in Markdown files
• LICENSE author rename

This helps simple edits complete faster and more predictably.

Commands

The extension contributes these VS Code commands:

• ManulAI: Open Chat
• ManulAI: Open Secondary Sidebar
• ManulAI: Select Ollama Model
• Attach to ManulAI Chat
• Attach Active File to ManulAI Chat
• Attach Explorer Selection to ManulAI Chat

Configuration

The extension exposes these settings:

• manulai.ollamaModel — local Ollama model used for chat and tool calling
• manulai.ollamaBaseUrl — base URL for the local Ollama server
• manulai.agentMode — turns tool-enabled agent behavior on or off
• manulai.autoApprove — skips approval prompts for tool execution when enabled
• manulai.systemPrompt — extra system prompt text prepended to each Ollama request

Default values:

• model: llama3.2
• Ollama base URL: http://localhost:11434
• agent mode: true
• auto-approve: false

Notes

• icon.png is used as the extension icon in the VS Code manifest
• media/manulai-icon.svg is used for the contributed sidebar container and view icon
• the project is intentionally Ollama-only and local-first
• the README describes current behavior and avoids cloud-oriented setup or product marketing fluff

What's New

• 0.0.1 (Alpha Release): Initial public alpha with right-side chat UI, local Ollama integration, workspace file attachments, native tool-calling support, agent/chat mode separation, approval controls, directory listing and file deletion tools, and stricter prompt rules for safer file edits.

License

This project is licensed under the Apache License 2.0. See the LICENSE file included in the extension package for details.