ASIAT is a terminal-native, Ink-powered TUI coding assistant that orchestrates code understanding, editing, and project automation for local repositories.

• Interactive TUI chat with an input box and scrollable history.
• Natural language intent parsing for explain, refactor, run, git, and file creation workflows.
• TypeScript-aware code indexing via ts-morph, persisted in a local SQLite database with lightweight embeddings for vector similarity.
• LLM abstraction layer supporting OpenAI or self-hosted LLaMA-compatible endpoints.
• Patch engine and file utilities for applying unified diffs or generating new files.
• Git integration for branch, status, diff, commit, and push workflows.
• Command executor that runs build, test, lint, or migration commands inside the target project.
• Session memory retaining the latest conversation turns for better follow-up interactions.

npm install
npm run build
npx asiat --project /path/to/your/project

During development you can run the CLI directly with TypeScript support:

npm run dev -- --project /path/to/your/project

Set the following environment variables as needed:

• OPENAI_API_KEY – required for OpenAI provider.
• ASIAT_LLM_PROVIDER – openai (default) or llama.
• ASIAT_LLM_MODEL – override default model name.
• ASIAT_LLM_BASE_URL – custom endpoint for non-OpenAI providers.

1. CLI bootstrap loads configuration and initializes shared services.
2. Ink TUI renders chat history, status bar, and command input.
3. Intent parser classifies user requests to drive downstream actions.
4. Code indexer walks the repository, extracts symbols, and stores embeddings.
5. Git engine exposes porcelain status, branching, commits, and push helpers.
6. LLM gateway normalizes access to remote or local language models.
7. Patch system applies unified diffs and manages file reads/writes.

• The SQLite database is stored in data/index.db by default.
• Embeddings are deterministic and CPU friendly so ASIAT works offline.
• Extend intent handling in src/tui/app.tsx to add new behaviors.
• The patch engine expects standard unified diff format when applying patches.

MIT