Skip to content

CONFIGURATION.md

Latest commit

 

History

History
67 lines (41 loc) · 3.02 KB

CONFIGURATION.md

File metadata and controls

67 lines (41 loc) · 3.02 KB
Raw

Configuration

Sift configuration comes from SQLite (settings and documents), environment variables (process-level overrides), and the in-app Settings screen.

Application data directory

By default the Node orchestrator writes:

  • SQLite database: data/sift.db (relative to the current working directory when the backend starts)

In development, that is usually the repo root’s data/ folder (created automatically).

SIFT_DATA_DIR

Override the directory that contains sift.db:

export SIFT_DATA_DIR=/path/to/writable/dir

The process must be able to create the directory and files there.

Backend HTTP server

Variable Default Description
PORT 4000 Listen port for Express
SIFT_LLM_BASE_URL (none; app defaults in DB/UI) Fallback base URL for OpenAI-compatible API
SIFT_LLM_MODEL (none) Fallback model id

The server binds to 127.0.0.1 only (localhost).

Settings stored in SQLite (settings table)

These are exposed via GET/POST /api/settings and the UI:

Key Meaning
watch_folder Directory monitored by the Tauri host (also persisted for UI)
llm_base_url Base URL for chat completions (often ends with /v1)
llm_model Model name passed to the LLM API

Defaults if unset (also used by the UI): http://127.0.0.1:11434/v1 and llama3.2 — aligned with Ollama’s OpenAI-compatible API on port 11434 (not 8080). See backend/src/llmDefaults.ts. Adjust the model to match ollama list on your machine (for example mistral, qwen2.5:14b).

LLM endpoint shape

The orchestrator calls OpenAI-compatible chat completions:

  • URL: {llm_base_url}/chat/completions (if llm_base_url already ends with /v1, paths are normalized in code).
  • Body: model, messages, temperature.

If the LLM is unreachable, ingest still completes using a heuristic summary and a lower confidence score.

Frontend

Variable Description
VITE_API_BASE Optional override for the API base (defaults to http://127.0.0.1:4000 in the client)

Tauri / Rust

Watcher and ingest calls are hardcoded to http://127.0.0.1:4000 in the host. If you change the backend port, update the Rust client (post_ingest in src-tauri/src/lib.rs) or add a shared config mechanism.

Related docs