VERSE
078e5e817c
- macOS, Linux badge and AI suggestions showcase in README - Update user guide and docs for development
30 lines
1.4 KiB
Markdown
30 lines
1.4 KiB
Markdown
# AI engine architecture (`internal/ai/`)
|
|
|
|
The AI subsystem provides real-time, context-aware command completions powered by cloud or local LLM providers (Groq, Ollama).
|
|
|
|
## Core components
|
|
|
|
- `internal/ai/client.go`: Defines the `Client` interface and handles HTTP requests to OpenAI-compatible chat completion endpoints.
|
|
- `internal/ai/env.go`: Captures runtime context snapshots (`EnvSnapshot`) including CWD, previous command, exit code, and recent history.
|
|
- `internal/ai/prompts.go`: Constructs system and user prompts optimized for shell command completion.
|
|
- `root/wrapper.go`: Manages async debounce timers, request cancellation (`context.WithCancel`), and ghost text injection.
|
|
|
|
## Provider interface
|
|
|
|
All providers use a unified HTTP pattern matching OpenAI's `/v1/chat/completions` format:
|
|
|
|
```go
|
|
type Client interface {
|
|
Suggest(ctx context.Context, prompt string, env *EnvSnapshot, currentCmd string) (*AISuggestion, error)
|
|
}
|
|
```
|
|
|
|
## Request lifecycle
|
|
|
|
1. User types in the prompt buffer.
|
|
2. `root/wrapper.go` triggers a debounce timer (`debounce_ms`, default 500ms).
|
|
3. If typing continues, previous in-flight contexts are cancelled (`aiCancel()`).
|
|
4. An `EnvSnapshot` is created capturing CWD, last executed command, and up to 3 recent history entries.
|
|
5. Provider sends an HTTP POST request with structured JSON payload.
|
|
6. Response is parsed and rendered as inline ghost text via `overlay.InjectAISuggestion()`.
|