Ashibalt AI
Beta — actively developed. Feedback is welcome!
What is it
Ashibalt AI is a full-featured AI coding agent for Visual Studio Code. It doesn't just answer questions — it can autonomously edit files, run terminal commands, search your project, diagnose errors, and iteratively solve complex tasks.
Features
- Agent Mode — autonomous code editing, file creation, terminal commands, iterative problem solving via agent loop
- Chat Mode — read-only AI assistant with your codebase context
- Multi-Provider — Ollama (local, free), OpenRouter, Mistral, DeepSeek.
- Model Browser — search and add models directly from the UI
- Snapshot System — every file edit creates a recoverable snapshot with inline Accept / Reject buttons
- 13 Tools —
read_file, edit_file, create_file, delete_file, list_files, search, terminal, xray_codebase, tasks, diagnose, lsp, fetch_url, web_search
- Autonomous Terminal — the agent runs commands in a dedicated terminal with automatic output capture, interactive prompt detection (y/n, password, selection), and user confirmation UI
- Semantic Project Analysis[LSP] —
xray_codebase tool provides a structural overview of any codebase: file tree with function/class signatures, constants, variables, and line numbers — supporting Python, TypeScript, JavaScript, Go and more
- Task Tracking —
tasks tool lets the agent create and manage a structured task list displayed in the chat UI with auto-clear on new requests
- Syntax Checking — tree-sitter based analysis for 14+ languages (TypeScript, Python, Rust, Go, C/C++, Java, Ruby, etc.)
- Context Management — automatic context compression near limits, context window management (up to 256K)
- Metrics — real token usage, prompt cache display, context window utilization
- Sessions — persistent chat history with switching and search
- Automatic Provider Selection - More details in Changelog.
Quick Start
- Install the extension from VS Code Marketplace
- Open the Ashibalt sidebar (icon in the Activity Bar)
- Open ⚙️ Settings, choose a provider and enter your API key:
- Ollama — install Ollama, run a model locally (free)
- OpenRouter — get an API key at openrouter.ai
- Mistral / DeepSeek — enter API key
- Pick a model and start coding!
Privacy
- API keys are stored locally in VS Code's secure secret storage
- Data is only transmitted between your machine and the chosen AI provider
License
MIT — see LICENSE.
Links
| |
